[Senate Hearing 117-723]
[From the U.S. Government Publishing Office]


                                                        S. Hrg. 117-723

                    DISRUPTING DANGEROUS ALGORITHMS:
                        ADDRESSING THE HARMS OF
                         PERSUASIVE TECHNOLOGY

=======================================================================

                                HEARING

                               BEFORE THE

                 SUBCOMMITTEE ON COMMUNICATIONS, MEDIA,
                             AND BROADBAND

                                 OF THE

                         COMMITTEE ON COMMERCE,
                      SCIENCE, AND TRANSPORTATION
                          UNITED STATES SENATE

                    ONE HUNDRED SEVENTEENTH CONGRESS

                             FIRST SESSION
                               __________

                            DECEMBER 9, 2021
                               __________

    Printed for the use of the Committee on Commerce, Science, and 
                             Transportation

                  [GRAPHIC NOT AVAILABLE IN TIFF FORMAT]

                Available online: http://www.govinfo.gov

                               __________

                    U.S. GOVERNMENT PUBLISHING OFFICE
                    
53-136  PDF               WASHINGTON : 2023   


       SENATE COMMITTEE ON COMMERCE, SCIENCE, AND TRANSPORTATION

                    ONE HUNDRED SEVENTEENTH CONGRESS

                             FIRST SESSION

                   MARIA CANTWELL, Washington, Chair
AMY KLOBUCHAR, Minnesota             ROGER WICKER, Mississippi, Ranking
RICHARD BLUMENTHAL, Connecticut      JOHN THUNE, South Dakota
BRIAN SCHATZ, Hawaii                 ROY BLUNT, Missouri
EDWARD MARKEY, Massachusetts         TED CRUZ, Texas
GARY PETERS, Michigan                DEB FISCHER, Nebraska
TAMMY BALDWIN, Wisconsin             JERRY MORAN, Kansas
TAMMY DUCKWORTH, Illinois            DAN SULLIVAN, Alaska
JON TESTER, Montana                  MARSHA BLACKBURN, Tennessee
KYRSTEN SINEMA, Arizona              TODD YOUNG, Indiana
JACKY ROSEN, Nevada                  MIKE LEE, Utah
BEN RAY LUJAN, New Mexico            RON JOHNSON, Wisconsin
JOHN HICKENLOOPER, Colorado          SHELLEY MOORE CAPITO, West 
RAPHAEL WARNOCK, Georgia                 Virginia
                                     RICK SCOTT, Florida
                                     CYNTHIA LUMMIS, Wyoming
                 Melissa Porter, Deputy Staff Director
       George Greenwell, Policy Coordinator and Security Manager
                 John Keast, Republican Staff Director
            Crystal Tully, Republican Deputy Staff Director
                      Steven Wall, General Counsel
                                 ------                                

          SUBCOMMITTEE ON COMMUNICATIONS, MEDIA, AND BROADBAND

BEN RAY LUJAN, New Mexico, Chair     JOHN THUNE, South Dakota, Ranking
AMY KLOBUCHAR, Minnesota             ROY BLUNT, Missouri
RICHARD BLUMENTHAL, Connecticut      TED CRUZ, Texas
BRIAN SCHATZ, Hawaii                 DEB FISCHER, Nebraska
EDWARD MARKEY, Massachusetts         JERRY MORAN, Kansas
GARY PETERS, Michigan                DAN SULLIVAN, Alaska
TAMMY BALDWIN, Wisconsin             MARSHA BLACKBURN, Tennessee
TAMMY DUCKWORTH, Illinois            TODD YOUNG, Indiana
JON TESTER, Montana                  MIKE LEE, Utah
KYRSTEN SINEMA, Arizona              RON JOHNSON, Wisconsin
JACKY ROSEN, Nevada                  SHELLEY MOORE CAPITO, West 
JOHN HICKENLOOPER, Colorado              Virginia
RAPHAEL WARNOCK, Georgia             RICK SCOTT, Florida
                                     CYNTHIA LUMMIS, Wyoming

                            C O N T E N T S

                              ----------                              
                                                                   Page
Hearing held on December 9, 2021.................................     1
Statement of Senator Lujan.......................................     1
Statement of Senator Thune.......................................    44
    Prepared statement...........................................    44
Statement of Senator Klobuchar...................................    49
Statement of Senator Fischer.....................................    51
Statement of Senator Schatz......................................    53
Statement of Senator Blackburn...................................    55
Statement of Senator Markey......................................    57
Statement of Senator Lee.........................................    59
Statement of Senator Baldwin.....................................    60
Statement of Senator Scott.......................................    62
Statement of Senator Cantwell....................................    64
Statement of Senator Peters......................................    67
Statement of Senator Rosen.......................................    69
Statement of Senator Blumenthal..................................    70

                               Witnesses

Rose Jackson, Director, Democracy + Tech Initiative, Digital 
  Forensic Research Lab, Atlantic Council........................     3
    Prepared statement...........................................     5
Jessica J. Gonzalez, Co-Chief Executive Officer, Free Press 
  Action.........................................................    17
    Prepared statement...........................................    18
James Poulos, Executive Editor of the American Mind, Claremont 
  Institute......................................................    26
    Prepared statement...........................................    28
Dr. Dean Eckles, Associate Professor of Marketing, MIT Sloan 
  School of Management...........................................    30
    Prepared statement...........................................    32

                                Appendix

Imran Ahmed, Founder and CEO, Center for Countering Digital Hate, 
  prepared statement.............................................    79
Letter dated December 9, 2021 to Senator Ben Ray Lujan and 
  Senator John Thune from the Disinfo Defense League.............    81
Response to written questions submitted to Rose Jackson by:
    Hon. Kyrsten Sinema..........................................    96
    Hon. Raphael Warnock.........................................    99
Response to written questions submitted to Jessica J. Gonzalez 
  by:
    Hon. Kyrsten Sinema..........................................   100
    Hon. Raphael Warnock.........................................   103
Response to written questions submitted to James Poulos by:
    Hon. John Thune..............................................   104
Response to written questions submitted to Dr. Dean Eckles by:
    Hon. Kyrsten Sinema..........................................   110
    Hon. John Thune..............................................   112

 
                    DISRUPTING DANGEROUS ALGORITHMS:
                        ADDRESSING THE HARMS OF
                         PERSUASIVE TECHNOLOGY

                              ----------                              


                       THURSDAY, DECEMBER 9, 2021

                               U.S. Senate,
        Subcommittee on Communications, Media, and 
                                         Broadband,
        Committee on Commerce, Science, and Transportation,
                                                    Washington, DC.
    The Subcommittee met, pursuant to notice, at 10:54 a.m., in 
room SR-253, Russell Senate Office Building, Hon. Ben Ray 
Lujan, Chairman of the Subcommittee, presiding.
    Present: Senators Lujan [presiding], Cantwell, Klobuchar, 
Blumenthal, Schatz, Markey, Baldwin, Rosen, Thune, Fischer, 
Blackburn, Lee, and Scott.

           OPENING STATEMENT OF HON. BEN RAY LUJAN, 
                  U.S. SENATOR FROM NEW MEXICO

    [Technical problems.]
    Senator Lujan.--today is holding a hearing on disrupting 
dangerous algorithms, addressing the harms of persuasive 
technology, to examine the risks posed by algorithmic 
amplification. As we have heard from testimony in previous 
hearings this year, algorithms, automation, and artificial 
intelligence have deep consequences for the health of our 
democracy and civil discourse, equal access to critical 
information, and the mental and physical health of our 
children.
    This hearing will focus on solutions that Congress and 
regulators can take to ensure these important innovations 
benefit all Americans and safeguard against potential harms. 
Today, more of our lives are lived online than ever before. 
Over the pandemic, the number of Americans using telehealth 
services more than tripled. More of us rely on the Internet for 
work, for school.
    Online platforms like Facebook, Twitter, YouTube, TikTok, 
Snapchat, and Amazon connect us to news, economic opportunity, 
and our friends and family. Online platforms helped Americans 
stay connected and access basic services during the pandemic. 
However, this exchange has come with a cost. Founded on 
innovative and disruptive business models, these companies have 
come to rely on algorithms. Data scientists and engineers at 
these companies design these algorithms to predict the content 
that is most likely to provoke a response and in turn amplify 
that content to users. Rather than users choosing what they see 
online, these algorithms make the decision for them to maximize 
growth and revenue.
    Civic health and well-being of consumers come second. If 
there was ever any doubt that business models and consumer 
interests were in conflict, Frances Haugen's testimony put 
those notions to rest. Her work and the documents she uncovered 
showed repeated underinvestment from Facebook in civic 
integrity, and in the health of minority communities at home 
and non-English speaking populations globally.
    She showed that Facebook had internal warnings that the 
platform was changing the global political conversation to be 
more extreme because of the content that it amplified. She 
showed that the algorithms favored emotions like anger and 
outrage due to deliberate decisions made at the highest levels 
of the leadership of these companies. We are here to chart a 
course forward.
    The Internet was created in the United States. ARPANET, the 
First Amendment, Section 230, each of these contributed to U.S. 
leadership in tech and innovation. It is time for the United 
States to step up and again to protect these principles in the 
future while also accounting for the harms. We know there is a 
lot to do, but there is much that our constituents across the 
country agree on here. First, constituents want to have control 
over their personal information. Second, they want transparency 
for what tech platforms do with that information.
    Third, they want oversight that can ensure these 
protections. And finally, they want accountability when 
platforms are negligent or actually harm Americans. It's all 
common sense. I don't know how we can't find common ground in 
each and every one of these areas, and it is clear based on the 
hearings that have come out of the Commerce leadership, this 
action is needed.
    Now we know Americans want more control over their data. 
These automated algorithms rely on user data to target 
consumers at the exact moment they are most likely to cause a 
change in behavior. Laxed oversight of data brokers, opaque and 
complicated collection practices, and insufficient disclosure 
all contribute to the harms these persuasive algorithms have to 
our health and well-being. Second, we need to know what is 
going on within these companies.
    Ms. Haugen described the internal sentiment of Facebook 
that, ``if information is shared with the public, it will just 
be misunderstood.'' This is frankly insulting to the American 
people. We need more public transparency that allows for 
meaningful public oversight and discourse over the true impact 
these platforms have on our daily lives. Third, we need to 
strengthen our existing institutions.
    The Federal Trade Commission, the States Attorneys General 
seek to protect consumers against harmful practices. With 
adequate resources, they can step up to this role. That is why 
I introduced the FTC Technologies Act to improve the 
Commission's ability to understand the dangers and protect 
consumers against harmful technologies. We must also strengthen 
local journalists and broadcasters to continue to act as checks 
to power and to live up to their roles as trusted voices in the 
communities that they serve.
    Finally, we need accountability when tech platforms are 
negligent and actively cause harm. That is why I also 
introduced the Protecting Americans from Dangerous Algorithms 
Act with the support of Senator Whitehouse and Representatives 
Malinowski and Eshoo. Platforms that have consequences for 
actively amplifying content that deprives Americans of their 
civil rights or contributes to terrorist activity. We may have 
a lot to do, but we are also making progress. The 
reconciliation package before the Senate makes historic 
investment in privacy enforcement at the Federal Trade 
Commission.
    The bill greatly improves the FTC's ability to go after bad 
actors by expanding its fining authority, and with a number of 
bipartisan bills to consider, Congress has the opportunity to 
make meaningful reforms. So I thank our witnesses for 
testifying today. Who is joining us remotely, Ms. Jessica 
Gonzalez is the Co-Chief Executive Officer at Free Press 
Action. Ms. Gonzalez has long been an advocate--advocating for 
the protection of minority communities and civil rights online, 
and I look forward to her perspective on that topic.
    Ms. Rose Jackson is the Director of Democracy in Tech 
Initiative at the Digital Forensic Research Lab with the 
Atlantic Council. She will provide important context for the 
business models that incentivize these algorithms and the 
global context in which they operate. Dr. James Poulos is 
Executive Director of the American Mind at the Claremont 
Institute, and he will provide a broader context for algorithm 
and digital regulation.
    Professor Dean Eckles is an Associate Professor of 
Marketing at the MIT Sloan School of Management and will 
provide testimony on the technical details on how these 
algorithms work and what measures, effective regulation should 
we consider. I want to thank Ranking Member Thune and his staff 
for their work to hold this hearing. And with that, I 
recognize--I will recognize the Ranking Member for remarks when 
he joins us. And I think we are going to go to our witnesses. 
And Ms. Jackson, we will hear from you first.

     STATEMENT OF ROSE JACKSON, DIRECTOR, DEMOCRACY + TECH 
  INITIATIVE, DIGITAL FORENSIC RESEARCH LAB, ATLANTIC COUNCIL

    Ms. Jackson. Thank you, and good morning, Chairman Lujan, 
Ranking Member Thune, and all of the members of the Committee. 
Thank you for the opportunity to appear here before you today. 
My name, as you said, is Rose Jackson. I am the Director of the 
Democracy and Tech Initiative at the Atlantic Council's Digital 
Forensic Research Lab.
    My work focuses on knitting together siloed conversations 
around tech governance, and that matters because the way that 
we fund tech, design it, govern it, fundamentally impacts 
democracy at home and abroad. Today's Internet is an outgrowth 
of incentives of global growth and financial gain. The Internet 
has connected the world and brought tremendous opportunity to 
billions. But the incentives that underpin it have also given 
rise to models that harm individuals and our democracy itself.
    To successfully address these tensions, whether to mitigate 
the impact of algorithms, limit consolidation or anti-
competitive behavior, protect human rights, or otherwise make 
the online world safer and fairer, we must recognize and 
address these incentives, as well as the complex and 
interconnected nature of the challenge. It is essential to 
acknowledge that the United States is not alone in this 
conversation. With our allies and our adversaries are actively 
trying to dictate the future of the global network, our allies 
are exploring sometimes conflicting solutions for how to 
reconcile new network challenges with deeply held democratic 
values. Meanwhile, authoritarian countries like Russia and 
China are taking an entirely different approach.
    They are driving an alternative vision for the Internet 
that is incompatible with our constitutional democracy and 
universal human rights. If they can drive a wedge between the 
world's democracies, they can reinforce their control at home, 
while limiting civic space online for everyone. And because the 
Internet is systemic, the rights that Americans are insured 
offline cannot exist in the online world if we don't continue 
to bolster the human rights based model internationally.
    Doing so requires us to answer our own tech policy gaps. 
And in that vein, Congress has an essential role in 
reestablishing U.S. leadership on these issues and 
incentivizing a safer and more accountable online experience 
for all Americans. Today, our digital world has grown to 
mediate nearly every aspect of our lives, and much of this 
ecosystem is made up of private companies, the majority of 
which are headquartered here in the United States.
    That means this committee in particular, can play a 
decisive role. It is hard to scroll through the news today 
without seeing stories about the latest tech scandal. Many of 
these harms stem from the same root cause succinctly identified 
by Mark Zuckerberg when he appeared before this very committee 
3 years ago. When he was asked about how Facebook sustains a 
business model in which users don't pay for services, 
Zuckerberg famously responded, Senator, we run ads. And it is 
the heart of the issue. It is not advertisements per se, but a 
particular business model of data extraction, user 
segmentation, targeted personalized, value maximized 
advertising.
    This model, which has powered much of the explosive growth 
of the internet's most successful and profitable firms, 
including social media companies, has an entire industry set on 
a race to gather as much information about us as possible, 
tracking us wherever we go, maximizing our time on their 
platforms, and feeding us any number of diversions to keep us 
clicking away.
    An extremely profitable intermediary market cleans, 
repackages, aggregates, and labels your data for anyone who 
wants to buy it, and that is either a shoe company or even a 
foreign intelligence agency. In the absence of meaningful 
privacy legislation, all of this is legal and very profitable.
    Given that engagement metrics and advertising revenue are 
some of the primary indicators of stock valuations, one could 
argue that in the status quo, many companies have a fiduciary 
responsibility to pursue data collection and engagement at all 
costs. But those costs are paid by people. And that can impact 
everything from someone's job prospects to their physical 
safety.
    The erosion of privacy, the algorithmic amplification of 
harmful content, and the compromise of information integrity 
all pose serious risks to our democracy and global stability. 
We must bring together all actors, including legislators like 
yourselves, industry leaders, and civil society experts to 
solve our problems together.
    And this means breaking down silos in Government and policy 
practices, improving our understanding of information 
ecosystems and the platforms that operate within it, 
strengthening our Government's ability to set and enforce the 
rules, and iterating our approaches as technology innovates.
    I provided a more expansive set of recommendations in my 
written testimony, but I wish to leave you with five key 
recommendations. First, Congress should work urgently to pass 
comprehensive Federal data protection and privacy legislation. 
Second, in the meantime, it should create and resource a new 
bureau at the FCC to address online privacy, data security, and 
other online abuses, as has been recommended by many in this 
committee and referenced by the Chairman just now and written 
into the House version of the Build Back Better Act.
    Third, Congress should also advance legislation to foster 
greater transparency and accountability around online harms, 
and address information asymmetries between companies and 
everyone else. Fourth, the Biden-Harris Administration should 
identify a lead at the White House to set a unified tech 
policy, bridging foreign and domestic priorities, and 
coordinating a more cohesive rights respecting approach across 
Government.
    And finally, the Administration also should integrate and 
elevate tech policy and U.S. diplomatic engagement to ensure 
the United States is a strong partner for the free, open, 
secure, and interoperable digital world. I am encouraged by the 
work of this committee and honored to be asked to testify. I 
submit the remainder of my testimony for the record, and I look 
forward to your questions. Thank you.
    [The prepared statement of Ms. Jackson follows:]

 Prepared Statement of Rose Jackson, Director of the Democracy + Tech 
      Initiative, Digital Forensic Research Lab, Atlantic Council
                           EXECUTIVE SUMMARY
    Good morning, Chairman Lujan, Ranking Member Thune, and all the 
members of the committee. Thank you for the opportunity to appear 
before you today. It is an honor to testify on this important topic.
    My name is Rose Jackson, and I'm the Director of the Democracy + 
Tech Initiative at the Atlantic Council's Digital Forensic Research 
Lab. My work focuses on knitting together the often siloed 
conversations around tech governance through the lens of democracy and 
human rights. This is relevant to this committee because the way that 
tech is funded, designed, and governed fundamentally impacts democracy 
at home and abroad.
    The internet, and today's dominant platforms, are an outgrowth of 
incentives-sometimes competing, and sometimes complementary-of global 
growth and financial gain. Together, we have built a wildly successful 
network that has connected the world and brought tremendous opportunity 
to billions. But these same incentives have also given rise to business 
models that exploit and extract value in ways that harm individuals and 
our democracy itself.
    We all share a goal of maximizing the best parts of our connected 
world, while minimizing its harms. If we are to successfully address 
these tensions, whether to mitigate the impact of algorithms, limit 
consolidation and anti-competitive behavior, protect human rights, or 
otherwise make the online world safer and fairer, we must start with an 
approach that recognizes and addresses these underlying incentives, and 
the complex and interconnected nature of the challenge.
    It is essential to acknowledge that the United States is not alone 
in this conversation. Other countries-from our allies to our 
adversaries-are actively regulating, legislating, and dictating the 
future of the global network. Allies like Canada, the United Kingdom, 
and the European Union are all exploring different, and sometimes 
conflicting, approaches to these same problems and platforms, seeking 
solutions for how to reconcile new networked challenges with deeply 
held democratic values.
    Meanwhile, authoritarian countries like China and Russia are taking 
an entirely different approach. They are proactively driving an 
alternative vision for the Internet that is incompatible with our 
constitutional democracy and universal rights. The authoritarian 
strategy benefits from a fragmented approach to the internet. If they 
can drive a wedge between the world's democracies, they can reinforce 
their control at home while limiting civic space online for everyone.
    And because the Internet is systemic, the rights Americans are 
ensured offline can't exist in the online world if we don't continue to 
promote and bolster the human rights-based model internationally. Doing 
so requires us to answer our own tech policy gaps. If we don't, the 
world will decide these things for us.
    Congress has an essential role to play in re-establishing U.S. 
leadership on these issues and incentivizing a safer and more 
accountable online experience for all Americans. The stakes couldn't be 
higher. Our failure to act threatens our national security, democracy, 
and global competitiveness.
    Today, our digital world has grown to mediate nearly every aspect 
of our lives, from the conduct of business to the exercise of our basic 
rights. Much of this essential ecosystem is made up of global private 
companies, the majority of which were founded and remain headquartered 
here in the United States. That means that this committee in particular 
plays a decisive role in shaping the incentives and ensuring 
protections for the platforms, providers, creators, and individuals 
that make up the digital ecosystem.
    It's hard to scroll through the news today without seeing stories 
about the latest tech scandal, whether whistleblower disclosures, 
hacking revelations, or the dystopian use of facial recognition tools. 
As the world grows disillusioned with an increasingly toxic digital 
experience,\1\ we find many of these harms stem from the same root 
cause, succinctly identified by Mark Zuckerberg when he appeared before 
this committee three years ago. When asked about how Facebook sustains 
a business model in which users don't pay for its services, Zuckerberg 
famously responded, ``Senator, we run ads.'' \2\
---------------------------------------------------------------------------
    \1\ Brooke Auxier, ``64 percent of Americans Say Social Media Have 
a Mostly Negative Effect on the Way Things Are Going in the U.S. 
Today,'' Pew Research Center, October 15, 2020. https://
www.pewresearch.org/fact-tank/2020/10/15/64-of-americans-say-social-
media-have-a-mostly-negative-effect-on-the-way-things-are-going-in-the-
u-s-today/.
    \2\ The Washington Post.''Transcript of Mark Zuckerberg's Senate 
Hearing,'' April 11, 2018, https://www.washingtonpost.com/news/the-
switch/wp/2018/04/10/transcript-of-mark-zucker
bergs-senate-hearing/.
---------------------------------------------------------------------------
    This is the heart of the issue. Not advertisements, per se, but a 
particular business model of data extraction, user segmentation, and 
targeted, personalized, value-maximized advertising.
    This model, which has powered much of the explosive growth of the 
internet's most successful and profitable platforms, has set an entire 
industry on a race to gather as much information about us as possible, 
tracking us wherever we go, maximizing our time on their platforms, and 
feeding us any number of diversions to keep us clicking away.
    And it's not just social media companies in on the game. An 
extremely profitable intermediary market, consisting of thousands of 
companies you've likely never heard of, cleans, repackages, aggregates, 
and labels your data for anyone who wants to buy it--whether a shoe 
company or a foreign intelligence agency.
    In the absence of meaningful privacy legislation, all of this is 
legal and highly lucrative. Given that engagement metrics and 
advertising revenue are the primary indicators that determine many 
companies' investment and stock valuations, one could argue that, in 
the status quo, these companies have a fiduciary responsibility to 
pursue data collection and engagement at all costs.
    But those costs are paid by people and can have serious 
ramifications on everything from someone's job prospects to physical 
safety. The erosion of privacy, algorithmic amplification of harmful 
content, and the compromise of information integrity all pose serious 
risks to our democracy and global stability.
    The issues are challenging, but the same ingenuity that built the 
Internet can be harnessed for the right solutions. We must bring 
together all actors, including legislators such as yourselves, industry 
leaders, and civil society experts, to solve our problems together. 
This means breaking down silos in government and policy practices; 
improving our understanding of our information ecosystem and the 
platforms within it; strengthening our government's ability to set and 
enforce the rules; and iterating our approaches as technology moves 
forward.
    I have provided a more expansive set of recommendations in the 
written testimony which I have submitted for the record. But I wish to 
leave you with five immediate, actionable things that would make a real 
difference in addressing the issues we are gathered to discuss today.

  1.  For Congress: Pass comprehensive Federal data protection and 
        privacy legislation.

  2.  For Congress: Create and resource a new bureau at the FTC to 
        address online privacy, data security, and other online abuses, 
        as provided for in the Chairwoman's Consumer Online Privacy 
        Rights Act (COPRA) and the House version of the Build Back 
        Better Act.

  3.  For Congress: Advance legislation to foster greater transparency 
        and accountability around online harms, and address information 
        asymmetries between companies and everyone else.

  4.  For the Administration: Identify a lead at the White House to set 
        a unified tech policy, bridging foreign and domestic 
        priorities, and coordinate a more cohesive rights-respecting 
        approach across the government.

  5.  For the Administration: Integrate and elevate tech policy in U.S. 
        diplomatic engagement to ensure the United States is a strong 
        partner for the free, open, secure, and interoperable digital 
        world.

    The goal of these recommendations is not to remove all bad things 
from the Internet or expect human nature itself to change. But rather, 
to provide the mechanisms necessary in a democracy for the public, 
government, and civil society to play their roles, in establishing 
clear expectations and rules, developing shared information, and 
relying on avenues for recourse to address harms when they happen.
    I am encouraged by the work of this committee and honored to be 
asked to testify. I submit the remainder of my testimony for the record 
and look forward to your questions.
                     GETTING TO THE ROOT OF THINGS
The Business Model
    The Internet has become essential to modern life. We use it to 
conduct business. We use it to learn. We use it to stay connected with 
our family and friends. And we use it to access government services and 
exercise our basic democratic freedoms.
    We often think of the Internet as an abstract idea, but its 
existence depends on a complex system of hardware and software. And 
today, most of those components from servers to websites, are built and 
run by private companies, each with their own business models. These 
models can take a range of forms, from your Internet service provider 
charging you a monthly fee to a marketplace taking a percentage from 
the sales price of those shoes you bought.
    Of course, platform advertising is a business model, too. The 
particular personalized, targeted advertising business model that has 
powered much of the explosive growth of the Internet has set companies 
on a race to gather as much information about us as possible, including 
by maximizing our time on their platforms.
The Incentives
    We produce an incredible amount of personal data points each day. 
The websites you visit and what you do on them. The details of your 
husband's grocery store run. The cell towers your phone pinged. What 
you typed in that search bar. When tracked and combined, it provides a 
shockingly accurate picture of a person's interests and behavior--a 
picture worth a lot of money to a lot of different potential buyers. 
According to a recent Aspen Institute report, in 2020 alone, the 
digital advertising market accounted for approximately $356 billion, 
with Google taking nearly 29 percent of the market, Facebook 25 
percent, and Amazon 10 percent.\3\
---------------------------------------------------------------------------
    \3\ Aspen Digital, ``Commission on Information Disorder Final 
Report,'' Aspen Institute, November 2021, https://
www.aspeninstitute.org/wp-content/uploads/2021/11/Aspen-Institute_Com
mission-on-Information-Disorder_Final-Report.pdf.
---------------------------------------------------------------------------
    With personal data worth so much, and with so few rules limiting 
its collection and use, almost everyone has an incentive to track and 
make money off of what we do online. Platforms have an incentive to 
supercharge your use of their services, because the more you use them, 
the more they know about you. An extremely profitable intermediary 
market, consisting of thousands of companies you've likely never heard 
of, cleans, repackages, aggregates, and labels your data for anyone who 
wants to buy it--whether a shoe company or a foreign intelligence 
agency. All of this is legal and highly lucrative.
    The platform design features and policies (such as recommendation 
algorithms or content moderation policies) are inevitably shaped by 
these monetary incentives. Given that engagement metrics and 
advertising revenue are the primary indicators that determine social 
media companies' investment and stock valuations, one could argue that, 
in the absence of any new legislation, these companies have a fiduciary 
responsibility to pursue engagement at all costs.
    Those costs are not paid by technology platforms. They are paid by 
us, American citizens. And their toll is pervasive, from the types of 
jobs we find to the mortgage rates we're offered. They are often nearly 
invisible, as well. We aren't informed about what opportunities are 
withheld from us because of the decision of an algorithm. We can't know 
what our better future might have been. This shapes our ability to 
choose the path our lives travel and exposes us to real risks.
                        UNDERSTANDING THE HARMS
    In a democratic society, the government has a duty to protect our 
rights and companies have a responsibility to respect our rights. 
However, that social contract is not reflected online in the United 
States today. While there are numerous harms stemming from our lack of 
tech sector regulation, in my testimony today, I'll focus on three 
broad categories: the erosion of privacy, algorithmic amplification of 
harmful content, and the compromise of information integrity.
    Individual privacy is a bedrock American right. It is also 
foundational to democracy but currently in peril. As citizens, we 
expect to have the ability to express ourselves, assemble, and 
otherwise conduct our lives without surveillance. While we pay a great 
deal of attention to limiting the powers of the state in this regard, 
when it comes to commercially driven privacy protections online, the 
United States is on an island by itself.
    Most countries have placed some restrictions on how an individual's 
data--whether personal details or behavior on and offline--can be used. 
But the lack of U.S. protections means the business of tracking 
Americans is quite lucrative. This financial upside incentivizes 
product approaches (e.g., certain algorithms) and policy decisions 
(e.g., what is and isn't allowed on a platform), designed to increase 
your engagement and generate more, and more valuable, data.
    We are most often unwitting participants in this exchange. For 
example, a company may produce an app that millions of people like to 
use to make funny videos. But that app's primary business could be 
sourcing faces to train algorithms or to track your location to sell to 
a third party, all without you knowing.
    Other platforms aren't just tracking you on their own apps, they 
follow you onto other sites, on your phone, or in some cases into the 
physical world. They combine the sum total of these data points to 
enable everyone from advertisers to political parties to better target 
you. But the network of data brokers I mentioned before means that this 
highly personalized information can be purchased by almost anyone for 
almost any purpose. This includes the U.S. and foreign governments.\4\
---------------------------------------------------------------------------
    \4\ Justin Sherman, ``Data Brokers Know Where You Are--and Want to 
Sell That Intel,'' Wired, August 8, 2021, accessed December 7, 2021, 
https://www.wired.com/story/opinion-data-brokers-know-where-you-are-
and-want-to-sell-that-intel/.
---------------------------------------------------------------------------
    What's more, if a company fails to secure your data and a hacker 
steals it, there's little to nothing you can do. Like me, you've 
probably received a dozen or more e-mails about private information 
like your credit card data, social security number, and purchase 
history being leaked in a data breach. But given the state of our laws 
today, there are few, if any, consequences for the companies you 
entrust your data to. There's no Federal law to require them to protect 
your data, and the FTC can only really go after companies if they were 
deceptive about the security promises they made to you.
    Data I don't even know exists about me can be bought and sold 
without my knowledge, and then it can be used for everything from 
convincing me to buy new clothes to determining if my health insurance 
premiums should go up.
    The privacy invasions here get worse when combined with algorithmic 
decision-making. Imagine that you have a fitness tracking app that 
shows that you recently stopped exercising as much. At the same time, 
you join your local bar's loyalty program that shows that you've 
started drinking more. Despite the fact that you may have just lost 
your Apple Watch or bought rounds of drinks for your team at work, a 
bank's algorithm might combine these data points and label you as a 
depressed alcoholic, leading you to be denied for a loan and hurting 
your credit score in the process.\5\
---------------------------------------------------------------------------
    \5\ Martin Tisne, ``It's Time for a Bill of Data Rights,'' MIT 
Technology Review, December 14, 2018, accessed December 7, 2021, 
https://www.technologyreview.com/2018/12/14/138615/its-time-for-a-bill-
of-data-rights/.
---------------------------------------------------------------------------
    In more authoritarian contexts, the data we generate just by using 
the Internet to click on posts and ``like'' things can be used to 
identify activists and members of vulnerable communities. Corporate-
held data is the fuel of the state surveillance machine, and it's 
relatively trivial for government actors around the world to track the 
movements of religious minorities, opposition parties, LGBTQ 
individuals, and human rights defenders. For every harm you can think 
of on the internet, the people most impacted are almost always the most 
vulnerable in society.
    Another harm is the erosion of the shared set of facts that 
democracy depends on. Camille Francois, a Lecturer at Columbia's School 
of International and Public Affairs, has used the term ``viral 
deception'' to focus conversations about the fragility of the 
information ecosystem on the manipulative actors, deceptive behaviors, 
and harmful content that combine to undermine democracy and target 
individuals.\6\
---------------------------------------------------------------------------
    \6\ Camille Francois, Briefing for the United States House of 
Representatives Committee on Science Space and Technology, 
Investigations and Oversight Subcommittee Hearing on Online Imposters 
and Disinformation, Graphika, September 26, 2019, https://
science.house.gov/imo/media/doc/Francois%20Testimony.pdf.
---------------------------------------------------------------------------
    While people often think of harms in this category as related to 
speech, solely focusing on content misses significant pieces of the 
puzzle. Microtargeting tools and recommendation engines are amplifying 
these dangerous messages and delivering them to those most susceptible.
    As researcher Renee DiResta has argued, speech and reach are not 
the same things.\7\ We must consider the harms amplified, 
microtargeted, or otherwise orchestrated through particular platform 
features or products. And further think of them in a multi-platform and 
multi-mode ecosystem.
---------------------------------------------------------------------------
    \7\ Renee DiResta, ``Free Speech Is Not the Same as Free Reach,'' 
Wired, August 30, 2018, https://www.wired.com/story/free-speech-is-not-
the-same-as-free-reach/.
---------------------------------------------------------------------------
    This includes everything from the ability to stoke public panic-as 
the Russian Internet Research Agency (IRA) did through coordinated 
messaging to convince a parish in Louisiana it was under imminent 
threat from a nonexistent chemical incident-to the deceptive 
advertising of harmful products, or the facilitation of outright 
illegal activities such as child trafficking or drug sales.\8\ It also 
includes the broader breakdown of public trust and open dialogue 
required for a functioning democracy. Of course, the events in Myanmar-
where the military and a group of extremist monks leveraged Facebook to 
unleash genocidal violence against the Rohingya people-reminds us of 
how severe the potential harms can be.\9\
---------------------------------------------------------------------------
    \8\ Camille Francois, Briefing for the United States House of 
Representatives Committee on Science Space and Technology, 
Investigations and Oversight Subcommittee Hearing on Online Imposters 
and Disinformation, Graphika, September 26, 2019, https://
science.house.gov/imo/media/doc/Francois%20Testimony.pdf.
    \9\ Human Rights Council, ``Report of the Independent International 
Fact-Finding Mission on Myanmar,'' September 2018, https://
www.ohchr.org/Documents/HRBodies/HRCouncil/FFM-Myanmar/A_HRC_39_64.pdf.
---------------------------------------------------------------------------
    To add to these dynamics, with the way real-time bidding for online 
ads works, many if not most advertisers don't even know what content 
their ads appear alongside, leading to sources of misinformation and 
radicalization directly profiting from this ecosystem as well. Indeed, 
a large part of the misinformation ecosystem would not exist without 
the ad ecosystem adding fuel to its fire. And this is independent of 
the social media platform systems, because even if a business, 
individual, or group is deplatformed, their websites can continue to 
make money from advertising, bolstered by the audience they built 
through social media.
    As study after study has shown, because content that outrages leads 
people to keep watching and clicking, companies will keep showing 
harmful content to users regardless of the consequences for the users 
or our society.\10\,\11\,\12\ That applies 
whether it be self-harm videos, sexualized images of children, calls 
for violence, or falsehoods about a politician.
---------------------------------------------------------------------------
    \10\ William J. Brady, Killian McLoughlin, Tuan N. Doan, and Molly 
J. Crockett, ``How Social Learning Amplifies Moral Outrage Expression 
in Online Social Networks,'' Science Advances 7 (33): eabe5641, 2021, 
https://doi.org/10.1126/sciadv.abe5641.
    \11\ Kate Starbird, Ahmer Arif, and Tom Wilso,. ``Disinformation as 
Collaborative Work,'' Proceedings of the ACM on Human-Computer 
Interaction 3 (CSCW): 1-26, 2019, https://doi.org/10.1145/3359229.
    \12\ Dag Wollebaek, Rune Karlsen, Kari Steen-Johnsen, and Bernard 
Enjolras, ``Anger, Fear, and Echo Chambers: The Emotional Basis for 
Online Behavior,'' Social Media + Society 5 (2): 205630511982985, 2019, 
https://doi.org/10.1177/2056305119829859.
---------------------------------------------------------------------------
    These challenges are inherently linked. The invasion of our privacy 
at the hands of the internet's advertising-driven business model leads 
to the microtargeting and amplification of harmful content, which in 
turn compromises the integrity of information online, all raising 
important questions about what the appropriate rules of the road should 
be for companies operating in this space.
    We have to avoid the trap of siloing our conversations to single 
harms, jurisdictions, or policy practices that end up missing the 
interrelated nature of business models, platform design, policy 
architecture, and user behavior.
                   BARRIERS TO A HEALTHIER ECOSYSTEM
    While there are certainly no silver bullets to these interrelated 
issues, this section is about approaches that will lead to better 
policy outcomes across the space. Two challenges in particular have 
hampered effective responses.
Policy and Issue Silos
    The first challenge is that, for the most part, we discuss the 
Internet and issues with it in defined policy silos. For instance, this 
Committee might examine everything from business regulation and trade 
to privacy and content moderation. The Judiciary Committees focus on 
antitrust and competition issues; the Foreign Relations Committees on 
geopolitical tech competition with countries like China; the Homeland 
Security Committees on cyber-and infrastructure security. And the Intel 
Committees focus on online drivers of radicalization or foreign 
influence operations. This dynamic is replicated in the executive 
branch. Each committee or agency focuses on the same platforms, 
ecosystems, and issues, but they are often doing so in isolation.
    This hearing is focused on how to address online harms exacerbated 
or created by the technologies, policies, and designs of various 
platforms. The truth is that each one of the above policy issues is 
interconnected and essential in some form to address the harms we are 
focusing on in this hearing. Siloing our conversations to single harms, 
jurisdictions, or policy practices ends up missing the interrelated 
nature of business models, platform design, policy architecture, and 
user behavior.
    Laying out the complexity of this ecosystem isn't meant to cause 
policy paralysis or inaction for fear of focusing on the wrong thing. 
It's meant to encourage a more holistic approach that unlocks broader 
coalitions and better solutions.
The Lack of Information and Explainability
    The second challenge is our limited understanding of the 
information ecosystem and how it operates within societies. The open-
source research community has been able to increasingly track networks 
seeking to manipulate platforms and populations. But it is undeniable 
that companies have more information on how harms are perpetrated on 
their platforms than anyone else. They do not, however, have cross-
platform visibility and are often siloed internally, meaning they may 
not track or examine the ways different business lines incentivize 
harms in another area of the company. Further, their own terms of 
service sometimes outright prohibit the kinds of open research that can 
add valuable context to our understanding of the information 
ecosystem.\13\
---------------------------------------------------------------------------
    \13\ Marshall Erwin, ``Getting Serious about Political Ad 
Transparency with Ad Analysis for Facebook--Open Policy & Advocacy,'' 
Open Policy & Advocacy, October 18, 2018, https://blog.mozilla.org/
netpolicy/2018/10/18/getting-serious-about-political-ad-transparency-
with-ad-analysis-for-facebook/.
---------------------------------------------------------------------------
    Even with the best of intentions, platforms alone cannot solve 
these problems. But for civil society, researchers, journalists, and 
governments to play their role, there need to be better mechanisms for 
understanding how companies operate, what their technology is doing, 
and how people are using or misusing their tools in ways that drive 
harm for society and individuals.
    ``Transparency'' has become the watchword of the day, perhaps the 
single greatest point of agreement in the tech policy community. But 
few people can articulate exactly what transparency should mean. Is it 
the disclosure of unlimited data? The release of company-scoped 
transparency reports or impact assessments? Reporting on enforcement? 
Or something else entirely? Transparency is a means, not an end. So 
while working to advance improved information and data access, it's 
important to be clear what we want from that information, which may be 
case specific, but should always be focused on enabling accountability, 
the protection of individual rights and privacy, and prevention of 
harms.
    For example, if a goal is explainability, we might then be focused 
on questions around how algorithms are intended to work, how they are 
trained, and how they influence the content people see. This is a more 
specific end than unscoped disclosure.
    There are of course a number of other barriers to action, including 
the lack of empowered and resourced regulators. But I want to focus now 
on the broader context in which this hearing takes place.
                           THE GLOBAL CONTEXT
    As a country we are behind on setting the rules needed to protect 
rights online. Our companies already comply with a myriad of foreign 
national and U.S. state laws on everything from privacy to content 
moderation. That patchwork of laws sometimes brings de facto 
improvements and protections to U.S. citizens, and other times 
undermines their basic human rights. As countries around the world 
grapple with how to address many of the same concerns we will discuss 
today, it's helpful to learn from their experimentation and consider 
U.S. action in the context of an urgent and existential global 
competition.
    Europe has been setting the digital rules of the road for years, 
starting most notably with the General Data Protection Regulation 
(GDPR), which set the global standard leading everyone from Brazil to 
Kenya to legislate basic data protections for its citizens. The United 
States is an outlier on this issue, and that leaves Americans uniquely 
vulnerable. Most U.S. companies already comply with the GDPR and other 
data protection legislation, they just don't generally make those 
protections available to Americans.
    Most recently though, Europe has jumped into a major, once-in-a-
generation rewriting of the rules of the digital economy in the form of 
the Digital Services and Digital Markets Acts (DSA and 
DMA).\14\,\15\ These two bills amount to the most 
significant rights-respecting effort at comprehensive tech regulation 
in history. Broadly, they cover everything from content moderation and 
the algorithms that shape our online experience to ads and commercial 
partnerships, as well as the data implications of mergers and 
acquisitions. For users, the outcomes will determine what they see 
online, how they can be tracked, and by whom. The bills will likely 
pass within the next year and set the global standard for determining 
which platforms meet criteria for regulation, content moderation norms, 
and requirements for transparency, reporting, and enforcement. With 
regard to algorithmic amplification specifically, the DSA would require 
covered platforms to make clear to users what targeting parameters are 
used to determine what they see through recommender systems, as well as 
provide an opt-out option for any personal data-based 
recommendations.\16\
---------------------------------------------------------------------------
    \14\ Proposal for a Regulation of the European Parliament and 
Council on a Single Market For Digital Services (Digital Services Act) 
and amending Directive 2000/31/EC, COM/2020/825 final https://eur-
lex.europa.eu/legal-content/en/TXT/?uri=COM%3A2020%3A825%3AFIN.
    \15\ Proposal for a Regulation of the European Parliament and 
Council on contestable and fair markets in the digital sector (Digital 
Markets Act) COM(2020)0842, https://www.europarl
.europa.eu/RegData/docs_autres_institutions/commission_europeenne/com/
2020/0842/COM_
COM(2020)0842_EN.pdf.
    \16\ Proposal for a Regulation of the European Parliament and 
Council on a Single Market For Digital Services (Digital Services Act) 
and amending Directive 2000/31/EC, COM/2020/825 final https://eur-
lex.europa.eu/legal-content/en/TXT/?uri=COM%3A2020%3A825%3AFIN.
---------------------------------------------------------------------------
    Of course these bills are not perfect, and as they move through the 
European Parliament, a number of troubling provisions are under 
consideration that hew closely to those in other allied nations 
struggling to get it right. The United Kingdom and Canada are currently 
considering bills that would require companies to proactively monitor 
and remove ill-defined categories of ``harmful'' 
speech.\17\,\18\ Australia has already passed a bill that 
requires this and mandates takedowns of such content within 24 hours of 
it being flagged. Australia takes things a step further and empowers a 
government authority to demand certain content be removed, without 
recourse for companies or users. Laws like these are proven to result 
in companies over-policing speech to avoid liability, while relying on 
algorithmic content moderation systems that do little to reduce real 
world harm.\19\
---------------------------------------------------------------------------
    \17\ Draft Online Safety Bill, Presented to Parliament by the 
Minister of State for Digital and Culture by Command of Her Majesty, 
May 2021, https://assets.publishing.service.gov.uk/government/uploads/
system/uploads/attachment_data/file/985033/Draft_Online_SafetyBill_Book
marked.pdf.
    \18\ ``The Government's Proposed Approach to Address Harmful 
Content Online,'' July 29, 2021, https://www.canada.ca/en/canadian-
heritage/campaigns/harmful-online-content.html.
    \19\ Daphne Keller, ``Empirical Evidence of Over-Removal by 
Internet Companies under Intermediary Liability Laws: An Updated 
List,'' Cyberlaw.stanford.edu, February 8, 2021, accessed December 8, 
2021, http://cyberlaw.stanford.edu/blog/2021/02/empirical-evidence-
over-removal-internet-companies-under-intermediary-liability-laws.
---------------------------------------------------------------------------
    India, once a leader in democratic online governance, has recently 
implemented draconian rules that require platforms to take down content 
when requested by a government ministry, and to assign in-country staff 
dedicated to respond to these requests who are personally criminally 
liable if their companies refuse to do so.\20\ This Indian law in 
particular is reminiscent of provisions recently leveraged in Russia 
threatening Google and Apple's in-country staff to compel the companies 
to remove an opposition party app from their platforms.\21\
---------------------------------------------------------------------------
    \20\ Jochai Ben-Avie, ``India Should Look to Europe as Its Model 
for Data Privacy,'' Financial Times, March 4, 2019, https://www.ft.com/
content/56ec37c8-39c0-11e9-9988-28303f70fcff.
    \21\ ``Google, Apple Remove Navalny App from Stores as Russian 
Elections Begin,'' Reuters, 
September 17, 2021, https://www.reuters.com/world/europe/google-apple-
remove-navalny-app-stores-russian-elections-begin-2021-09-17/.
---------------------------------------------------------------------------
    These discrepancies in approach and embrace of undemocratic 
regulations bolster the efforts of China and other authoritarian 
countries proactively advancing an alternative vision for the internet. 
The consensus on a rights-based world order helped drive the adoption 
of the early principles stating that the Internet should be global, 
free, open, secure, and interoperable. Those principles helped spur the 
U.S. tech industry, and they still largely stand today. But they are 
being actively challenged by authoritarians who want a splinternet 
where they can be as repressive and controlling online as they are in 
the physical world, while leveraging the digital world to facilitate 
their centralized political control and sustained power.
    The interests of authoritarians, from Russia to Iran, are well 
served by initiatives like China's Digital Silk Road, which is pouring 
billions of dollars into digital infrastructure. With nothing similar 
emerging from the world's democracies, many amongst the half of 
humanity currently unconnected to the Internet will take their first 
steps online into a state-owned, authoritarian-inclined digital 
landscape. At each step in their digital journey--from the fiber they 
connect through to the hardware and software they use to connect--their 
data will feed state-owned or backed enterprises that have shown few 
qualms supporting authoritarian agendas.
    These countries are simultaneously leveraging the international 
system to advance their vision, seeking influence and control over the 
venues where norms and treaties around Internet policy are set. As we 
speak, countries are debating these issues in Krakow at the Internet 
Governance Forum, the main UN body for global, multi-stakeholder 
conversations on Internet policy. A Russian and an American are 
competing for the 2022 presidency of the International 
Telecommunication Union, a global treaty body which sets standards for 
global telecommunications, including the internet.\22\ At the 
International Organization for Standardization, a Huawei senior 
director chairs the committee that sets global standards on AI and 
other technologies.\23\ While authoritarian countries advance their 
preferences through the ``rules-based order,'' democratic countries are 
falling behind.
---------------------------------------------------------------------------
    \22\ Aaron Schaffer, ``Analysis | U.S. And Russian Candidates Both 
Want to Lead the U.N.'s Telecom Arm,'' The Washington Post, October 12, 
2021, accessed December 8, 2021, https://www.washingtonpost.com/
politics/2021/10/12/us-russian-candidates-both-want-lead-un-tele
com-arm/.
    \23\ Justus Baron and Olia Kanevskaia Whitaker, Global Competition 
for Leadership Positions in Standards Development Organizations, March 
31, 2021, http://dx.doi.org/10.2139/ssrn.38
18143.
---------------------------------------------------------------------------
    Our companies will compete in this landscape and, as they have for 
years, respond to the regulations that are required of them. Each of 
these laws and actions are shaping our options. All of these foreign 
countries are deciding for us what the answer to these challenges will 
be.
    And because the Internet is systemic, there isn't a world in which 
the rights Americans are ensured offline exist in the online world if 
we don't continue to promote and bolster the human rights-based model 
internationally. Doing so requires us to answer our own policy gaps on 
these issues.
    To say it clearly, our lack of clear regulatory frameworks at home 
is a foreign policy weakness and a national security threat to our 
country.
                            RECOMMENDATIONS
    The good news is that there are a number of things that this 
committee, other parts of Congress, the executive branch, industry, and 
the American public can do. Addressing these issues will require 
collective action-each of the following recommendations will be 
strengthened by a ``multi-stakeholder approach;'' that is, designing 
our policies to incentivize industry, empower civil society, center the 
American public and users, and strengthen the government's ability to 
set and enforce the rules. Regulation alone will not solve these 
problems, but the absence of it will make it hard for everyone else to 
play their roles.
    We also must consider this policymaking in the global context and 
work closely with our allies to ensure we are not regulating at cross 
purposes. Perhaps more importantly though, U.S. leadership is necessary 
to help drive better global outcomes and ensure we have the ability to 
make these decisions for ourselves. U.S. foreign policy should 
prioritize rights-based tech governance, and, to do so, we need to 
center it in the highest levels of our diplomatic engagements. To that 
end, I am encouraged by Secretary of State Blinken's recent 
announcement creating a new bureau focused on tech diplomacy, 
reorganizing resources and expertise within the Department, and 
bringing in new people and skills to bolster the agency's capacity on 
these issues. Congress should ensure that this new bureau receives the 
resources and authorities it needs to fill this essential and unique 
role.\24\
---------------------------------------------------------------------------
    \24\ ``Office of the Coordinator for Cyber Issues,'' United States 
Department of State, accessed December 8, 2021, https://www.state.gov/
bureaus-offices/secretary-of-state/office-of-the-coordinator-for-cyber-
issues/.
---------------------------------------------------------------------------
Government Capacity to Set and Implement Tech Policy
    But here there is an urgent need for the U.S. government to have a 
unified approach to tech policy, knitting together an overarching 
strategy for its foreign policy, trade priorities, and domestic 
imperatives, with the human-rights frame at its core. There are plenty 
of qualified officials spread throughout the government working on 
pieces of this broader issue set. The Biden-Harris Administration 
should name a Tech Policy lead, with a joint National Security Council, 
National Economic Council, Domestic Policy Council, and Office of 
Science and Technology Policy mandate, and formalize a working group to 
articulate clear areas of lead and support. This official should be 
tasked with reviewing the existing equities across the wide range of 
agencies and offices that touch these issues, work to develop a unified 
policy, and ensure the full force of U.S. power is moving in a common, 
rights-respecting direction.
    For Congress, prioritizing action around protecting privacy, 
enabling transparency and accountability, and ensuring the U.S. 
government is appropriately staffed and resourced for our digital age 
would provide a meaningful foundation for many of the specific laws and 
rulemaking under consideration.
    It's not just the State Department that needs the staff and mandate 
to drive better outcomes on technology. As the primary regulator of the 
tech industry, the FTC has increasing demands on it to oversee an 
industry that has grown exponentially. And yet, it has fewer total 
staff than it did in the 1970s and, in particular, woefully 
insufficient tech and data-specific staff to keep 
apace.\25\,\26\
---------------------------------------------------------------------------
    \25\ Group Letter in Support of FTC Privacy Funding, September 21, 
2021, https://www.accessnow.org/cms/assets/uploads/2021/09/Group-
letter-in-support-of-FTC-privacy-funding.pdf.
    \26\ The FTC is vastly underfunded and understaffed, particularly 
in comparison to the large, well-funded entities that it is tasked with 
regulating. Currently, the FTC only has 1,100 full-time employees 
(FTEs) to pursue both its competition and consumer protection missions. 
This number has been roughly flat over the past twelve years, and 
represents a substantial decrease from 1,746 FTEs in 1979. Put another 
way, since that time, the economy has grown nearly three times while 
the FTC's capacity has decreased 37 percent. In contrast, in 2020, 
Facebook alone had total revenues of nearly $86 billion and nearly 
60,000 employees.
---------------------------------------------------------------------------
    The House version of the Build Back Better Act borrows from this 
committee's ideas and leadership and features significant resources to 
create and staff a new bureau at the FTC to address online privacy, 
data security, and other online abuses.\27\ Advancing those resources 
through the Senate would be a significant step forward on privacy and 
tech governance more broadly. Other efforts, like Chairman Lujan's 
Technologists Act, could also help ensure the FTC has the capacity it 
needs to fill its essential role.\28\ But the task of rulemaking and 
enforcement cannot sit with the FTC alone. The Federal Communications 
Commission (FCC), Consumer Financial Protection Bureau (CFPB), and 
others have significant roles to play. Congress should use the tools it 
has to help bolster staffing and expertise within those organizations, 
while also using comprehensive reform efforts to better define their 
responsibilities.
---------------------------------------------------------------------------
    \27\ Congress.gov. ``Text--H.R. 5376--117th Congress (2021-2022): 
Build Back Better Act.'' November 19, 2021. https://www.congress.gov/
bill/117th-congress/house-bill/5376/text.
    \28\ Congress.gov, ``S.3187--117th Congress (2021-2022): Federal 
Trade Commission Technologists Act of 2021,'' November 4, 2021, https:/
/www.congress.gov/bill/117th-congress/senate-bill/3187.
---------------------------------------------------------------------------
    The same is true across the Federal government. Investments in the 
appropriate structure and staff to drive a cogent approach forward, 
whether in the Departments of Homeland Security, Defense, or Health and 
Human Services, will impact our ability to lead. Programs like 
Congress's own ``Tech Congress'' demonstrate the value of well-placed 
technologists in these roles. It also behooves Capitol Hill to find 
ways to break the legislative silos that exist around technology 
policy. Working across committee lines and jurisdictions needs to 
become the norm when approaching these questions.
    To bring the point home, earlier this week, on December 7, the 
Senate Finance Subcommittee on Fiscal Responsibility and Economic 
Growth held a hearing on data brokers.\29\ The witnesses all focused on 
the urgency of privacy legislation and the impact that the data broker 
market has on the issues under discussion at this hearing.
---------------------------------------------------------------------------
    \29\ ``Promoting Competition, Growth, and Privacy Protection in the 
Technology Sector | the United States Senate Committee on Finance,'' 
December 8, 2021. www.finance.senate.gov, accessed December 8, 2021, 
https://www.finance.senate.gov/hearings/promoting-competition-growth-
and-privacy-protection-in-the-technology-sector.
---------------------------------------------------------------------------
    Congress could formalize or incentivize collaboration across the 
various Committees that share remit or responsibilities for tech policy 
and oversight. For instance, a caucus among Members or an informal 
forum for legislative staff to work together could serve an important 
purpose in generating ideas or deconflicting areas of lead and support 
among committees. This step would make it easier for experts to engage 
with Congress, as opposed to member by member or committee by 
committee.
Privacy
    Substantively, the United States needs a comprehensive Federal data 
protection and privacy law. U.S. companies already comply with the GDPR 
in Europe and strong state laws in California and Illinois. Extending 
these protections to all Americans would help to shift the incentives 
currently driving many of the harms we discussed today and provide a 
base set of protections that make it easier to begin to address other 
tech governance challenges, from transparency to advertising. The 
bottom line is, every American should be able to control what data of 
theirs is collected, how it is used, and who has access to it.
    In addition to pursuing this comprehensive approach, the committee 
should consider bolstering regulators' mandates for addressing data and 
privacy violations. For the FTC, section 5 is too limited as currently 
conceived. Further, with regulatory and enforcement jurisdiction on 
this issue spanning the FTC, FCC, CFPB, and DOJ, we need to better 
coordinate and ensure clearer understanding of how each body can 
approach its role on tech oversight in general. The only beneficiaries 
of a patchwork approach are special interest and the largest companies.
Transparency
    It seems that almost everyone agrees that establishing meaningful 
transparency and data-sharing standards is a foundational requirement 
to address an array of digital issues. What such transparency should 
include, for which purposes, and what information is shared with which 
people are big questions we need to answer to ensure we don't legislate 
transparency for transparency's sake. With that said, Congress can and 
should take action to address the information asymmetry that exists 
between what companies know about their corner of the information 
ecosystem and what everyone else knows. This requires figuring out the 
mechanisms through which information can be securely and appropriately 
shared. Doing so can also help shape the questions we as a society are 
focused on in ensuring our technology reinforces our democracy.
    To that end, I am encouraged by a slate of new legislative 
proposals in the U.S. Congress focused on this question, alongside 
major provisions in the European Union's DSA that mandate data sharing 
with regulatory authorities and qualified researchers, alongside 
potential risk-assessments and human rights impact reports.\30\ 
Congress would be wise to build on the momentum created by this EU 
action and shape the direction both the EU and our own bodies take in 
incentivizing better information sharing and unlocking improved 
research.
---------------------------------------------------------------------------
    \30\ Proposal for a Regulation of the European Parliament and 
Council on a Single Market For Digital Services (Digital Services Act) 
and amending Directive 2000/31/EC COM/2020/825, https://eur-
lex.europa.eu/legal-content/en/TXT/?uri=COM%3A2020%3A825%3AFIN
---------------------------------------------------------------------------
    It's worth noting that you would not be acting alone. In addition 
to the EU, the Organization for Economic Co-operation and Development 
and the United Nations are also advancing transparency-focused 
initiatives.\31\,\32\ And I'm excited to announce a 
coalition my organization is jointly forming with the Global Network 
Initiative, the Institute for Strategic Dialogue, the Partnership for 
Countering Influence Operations, and the Center for Democracy and 
Technology, among others, that seeks to set common definitions, to 
better articulate the trade-offs, and to ensure each of the regulatory 
and policy efforts underway have a common grounding and language. There 
is much to work out in the details, but I am confident that this is an 
area where a great deal of progress can be made over the next year, and 
I hope our coalition's efforts, alongside these other initiatives, can 
inform the work of this committee and Congress more broadly.
---------------------------------------------------------------------------
    \31\ J. Llanos, Transparency reporting: Considerations for the 
review of the privacy guidelines,'' OECD Digital Economy Papers, No. 
309, OECD Publishing, 2021, https://doi.org/10.1787/e90c11b6-en.
    \32\ The United Nations, ``Our Common Agenda--Report of the 
Secretary-General,'' September 2021, https://www.un.org/en/content/
common-agenda-report/assets/pdf/Common_Agenda_Re
port_English.pdf.
---------------------------------------------------------------------------
    With regard to social media algorithms specifically, proposals like 
those in the DSA, that provide some explainability and transparency to 
the user as to why they are seeing what they are seeing could be an 
important step in this larger conversation. However, doing so without 
giving the user the ability to change or remove those display 
algorithms would amount to transparency without control or 
accountability.
    Which brings me to an important caution in the focus on 
transparency, which is that failure to tie it to accountability can 
result in chasing a bottomless pit of data, and what researcher Charley 
Johnson refers to as an endless transparency feedback loop, 
illuminating individual problems but not systemic drivers of those 
problems.\33\
---------------------------------------------------------------------------
    \33\ Charley Johnson, ``Some Unsatisfying Solutions for Facebook,'' 
Untangled, December 5, 2021. https://untangled/p/-some-unsatisfying-
solutions-for.
---------------------------------------------------------------------------
Other Remedies
    As we stay focused on remedies that address the underlying 
incentives driving online harms, it's worth calling out a number of 
proposals worthy of attention. The first is efforts to address the ad-
based model driving many of these business incentives. The Honest Ads 
Act is a common sense option for migrating standards on political 
advertising that we already have on TV and radio to the digital 
sphere.\34\ We should also consider proposals that require 
comprehensive disclosure of ads to researchers, such as the House 
Social Media DATA Act proposed by Representative Trahan.\35\
---------------------------------------------------------------------------
    \34\ Congress.gov, ``S.1356--116th Congress (2019-2020): Honest Ads 
Act,'' May 7, 2019, https://www.congress.gov/bill/116th-congress/
senate-bill/1356.
    \35\ Congress.gov, ``H.R. 3451--117th Congress (2021-2022): Social 
Media DATA Act,'' May 31, 2021, https://www.congress.gov/bill/117th-
congress/house-bill/3451.
---------------------------------------------------------------------------
    Efforts targeting virality as a unique characteristic of the 
problem could also have an impact--whether finding ways to introduce 
friction into sharing or prioritizing human review of content once it 
shows markers of potential viral spread.
    If this list of recommendations sounds expansive, it's because it 
is. As I said at the beginning of this testimony, single interventions 
will not work in addressing the underlying incentives driving online 
harms. The menu of options available for action are all interconnected.
    However, even if each of these things happened tomorrow, our 
digital world would still be fraught. The goal of these recommendations 
is not to remove all bad things from the Internet or expect human 
nature itself to change. But rather, to provide the mechanisms 
necessary in a democracy for the public, government, and civil society 
to play their roles, in establishing clear expectations and rules, 
developing shared information, and relying on avenues for recourse to 
address harms when they happen.
    It's worth noting that the issues we're discussing do not sit with 
platforms and the technology they build alone. We won't solve all of 
society's ills through Internet governance. But we can work to ensure 
that the way platforms are funded, built, and governed at the very 
least does not exacerbate the harms we are discussing today.
    I appreciate the Committee's leadership on these urgent issues and 
look forward to working together to address them.
Bibliography
    Brooke Auxier, ``64 percent of Americans Say Social Media Have a 
Mostly Negative Effect on the Way Things Are Going in the U.S. Today,'' 
Pew Research Center, October 15, 2020, https://www.pewresearch.org/
fact-tank/2020/10/15/64-of-americans-say-social-media-have-a-mostly-
negative-effect-on-the-way-things-are-going-in-the-u-s-today/.
    Justus Baron and Olia Kanevskaia Whitaker, ``Global Competition for 
Leadership Positions in Standards Development Organizations,'' March 
31, 2021, http://dx.doi.org/10.2139/ssrn.3818143
    Jochai Ben-Avie, ``India Should Look to Europe as Its Model for 
Data Privacy,'' Financial Times, March 4, 2019, https://www.ft.com/
content/56ec37c8-39c0-11e9-9988-28303f70fcff.
    William J. Brady, Killian McLoughlin, Tuan N. Doan, and Molly J. 
Crockett, ``How Social Learning Amplifies Moral Outrage Expression in 
Online Social Networks,'' Science Advances 7 (33): eabe5641, 2021, 
https://doi.org/10.1126/sciadv.abe5641.
    Aspen Digital, ``Commission on Information Disorder Final Report,'' 
Aspen Institute, November 2021, https://www.aspeninstitute.org/wp-
content/uploads/2021/11/Aspen-Institute_Com
mission-on-Information-Disorder_Final-Report.pdf.
    Congress.gov, ``Text--H.R. 5376--117th Congress (2021-2022): Build 
Back Better Act,'' November 19, 2021, https://www.congress.gov/bill/
117th-congress/house-bill/5376/text.
    Congress.gov, ``S.3187--117th Congress (2021-2022): Federal Trade 
Commission Technologists Act of 2021,'' November 4, 2021, https://
www.congress.gov/bill/117th-congress/senate-bill/3187.
    Congress.gov, ``S.1356--116th Congress (2019-2020): Honest Ads 
Act,'' May 7, 2019, https://www.congress.gov/bill/116th-congress/
senate-bill/1356.
    Congress.gov, ``H.R. 3451--117th Congress (2021-2022): Social Media 
DATA Act,'' May 31, 2021, https://www.congress.gov/bill/117th-congress/
house-bill/3451.
    Renee DiResta, ``Free Speech Is Not the Same as Free Reach,'' 
Wired, August 30, 2018, https://www.wired.com/story/free-speech-is-not-
the-same-as-free-reach/.
    Charles Duhigg, ``How Companies Learn Your Secrets,'' The New York 
Times, February 16, 2012, https://www.nytimes.com/2012/02/19/magazine/
shopping-habits.html.
    Marshall Erwin, ``Getting Serious about Political Ad Transparency 
with Ad Analysis for Facebook--Open Policy & Advocacy,'' Open Policy & 
Advocacy, October 18, 2018, https://blog.mozilla.org/netpolicy/2018/10/
18/getting-serious-about-political-ad-transparency-with-ad-analysis-
for-facebook/.
    Camille Francois, Briefing for the United States House of 
Representatives Committee on Science Space and Technology, 
Investigations and Oversight Subcommittee Hearing on Online Imposters 
and Disinformation, Graphika, September 26, 2019, https://
science.house.gov/imo/media/doc/Francois%20Testimony.pdf.
    ``Google, Apple Remove Navalny App from Stores as Russian Elections 
Begin,'' Reuters, September 17, 2021, https://www.reuters.com/world/
europe/google-apple-remove-navalny-app-stores-russian-elections-begin-
2021-09-17/.
    ``The Government's Proposed Approach to Address Harmful Content 
Online,'' July 29, 2021, https://www.canada.ca/en/canadian-heritage/
campaigns/harmful-online-content.html.
    Group Letter in Support of FTC Privacy Funding, September 21, 2021, 
https://www.accessnow.org/cms/assets/uploads/2021/09/Group-letter-in-
support-of-FTC-privacy-funding.pdf.
    Human Rights Council, ``Report of the Independent International 
Fact-Finding Mission on Myanmar,'' September 2018, https://
www.ohchr.org/Documents/HRBodies/HRCouncil/FFM-Myanmar/A_HRC_39_64.pdf.
    Charley Johnson, ``Some Unsatisfying Solutions for Facebook,'' 
Untangled, December 5, 2021, https://untangled.substack.com/p/-some-
unsatisfying-solutions-for.
    Daphne Keller, ``Empirical Evidence of Over-Removal by Internet 
Companies under Intermediary Liability Laws: An Updated List,'' 
Cyberlaw.stanford.edu, February 8, 2021,. accessed December 8, 2021, 
http://cyberlaw.stanford.edu/blog/2021/02/empirical-evidence-over-
removal-internet-companies-under-intermediary-liability-laws.
    J. Llanos, Transparency reporting: Considerations for the review of 
the privacy guidelines,'' OECD Digital Economy Papers, No. 309, OECD 
Publishing, 2021, https://doi.org/10.1787/e90c11b6-en.
    ``Office of the Coordinator for Cyber Issues,'' United States 
Department of State, accessed December 8, 2021, https://www.state.gov/
bureaus-offices/secretary-of-state/office-of-the-coordinator-for-cyber-
issues/.
    ``Promoting Competition, Growth, and Privacy Protection in the 
Technology Sector | the United States Senate Committee on Finance,'' 
December 8, 2021, accessed December 8, 2021, www.finance.senate.gov.
    ``Proposal for a Regulation of the European Parliament and Council 
on contestable and fair markets in the digital sector (Digital Markets 
Act),'' COM(2020)0842, https://www.euro
parl.europa.eu/RegData/docs_autres_institutions/commission_europeenne/
com/2020/0842/COM_COM(2020)0842_EN.pdf.
    ``Proposal for a Regulation of the European Parliament and Council 
on a Single Market For Digital Services (Digital Services Act) and 
amending Directive 2000/31/EC COM/2020/825,'' https://eur-
lex.europa.eu/legal-content/en/TXT/?uri=COM%3A2020%3A825%3AFIN.
    Aaron Schaffer, ``Analysis | U.S. And Russian Candidates Both Want 
to Lead the U.N.'s Telecom Arm,'' The Washington Post, October 12, 
2021, accessed December 8, 2021, https://www.washingtonpost.com/
politics/2021/10/12/us-russian-candidates-both-want-lead-un-telecom-
arm/.
    Justin Sherman, ``Data Brokers Know Where You Are--and Want to Sell 
That Intel,'' Wired, August 8, 2021, accessed December 7, 2021, https:/
/www.wired.com/story/opinion-data-brokers-know-where-you-are-and-want-
to-sell-that-intel/.
    Kate Starbird, Ahmer Arif, and Tom Wilson, ``Disinformation as 
Collaborative Work,'' Proceedings of the ACM on Human-Computer 
Interaction 3 (CSCW): 1-26, https://doi.org/10.1145/3359229.
    Martin Tisne, ``It's Time for a Bill of Data Rights,'' MIT 
Technology Review, December 14, 2018, accessed December 7, 2021, 
https://www.technologyreview.com/2018/12/14/138615/its-time-for-a-bill-
of-data-rights/.
    The Washington Post. 2018. ``Transcript of Mark Zuckerberg's Senate 
Hearing,'' The Washington Post. April 11, 2018,. https://
www.washingtonpost.com/news/the-switch/wp/2018/04/10/transcript-of-
mark-zuckerbergs-senate-hearing/.
    The United Nations, ``Closing Remarks 15th Annual Internet 
Governance Forum Internet Governance in the Age of Uncertainty,'' 
November 17, 2020, https://www.un.org/en/desa/closing-remarks-15th-
annual-internet-governance-forum-internet-governance-age-uncertainty.
    The United Nations, ``Our Common Agenda--Report of the Secretary-
General,''. Published by the United Nations. September 2021,. https://
www.un.org/en/content/common-agenda-report/assets/pdf/
Common_Agenda_Report_English.pdf.
    Dag Wollebaek, Rune Karlsen, Kari Steen-Johnsen, and Bernard 
Enjolras, ``Anger, Fear, and Echo Chambers: The Emotional Basis for 
Online Behavior,'' Social Media + Society 5 (2): 205630511982985, 2019, 
https://doi.org/10.1177/2056305119829859.

    Senator Lujan. Thank you, Ms. Jackson. Next, we will hear 
from Ms. Gonzalez, the Co-CEO of Free Press. Jessica.

 STATEMENT OF JESSICA J. GONZALEZ, CO-CHIEF EXECUTIVE OFFICER, 
                       FREE PRESS ACTION

    Ms. Gonzalez. Thank you, Chairman Lujan, Ranking Member 
Thune, members of the Subcommittee. Thank you for inviting me 
to testify and please accept my condolences on the passing of 
your former colleague, Senator Dole.
    Today's hearing calls us to address the harms of 
``persuasive technology.'' Of course, harmful and persuasive 
media is nothing new. As Free Press's media 2070 Project has 
documented, there is a long history of the U.S. media system 
spreading racist propaganda.
    For instance, what is unique about the digital age is how 
online platforms collect troves of private data about us and 
then use it to target ads, recommendations, and other content 
based on our perceived interests and vulnerabilities. Sometimes 
this system produces innocuous or even beneficial outcomes.
    But as whistleblowers, journalists, researchers, and 
activists have revealed, too often these systems cause grave 
harm, all while companies have little to no accountability to 
the American people. Indeed, they have failed outright at 
fulfilling even the most basic requests for transparency about 
how their systems work. They have rolled out expensive PR 
campaigns while continuing to profit handsomely from amplifying 
conspiracy theories and bigotry and enabling discrimination.
    Tech firms are unwilling and unfit to effectively self-
govern. They have amplified lies about COVID and vaccines, 
served up housing and employment ads in discriminatory ways, 
amplified voter suppression campaigns launched by foreign State 
actors, and discouraged participation in the U.S. census. They 
targeted abuse at people of color and have done a particularly 
poor job at solving these problems in languages other than 
English.
    Facebook long has known the extent to which its products 
harm people of color, other minority groups, and our society. 
Its top executives have brazenly and routinely lied to or 
withheld the full truth from civil rights leaders, researchers, 
the U.S. Congress, and the American people. This simply can't 
stand. This week, Free Press action and allies and the Disinfo 
Defense League released a policy platform designed to rein in 
these abuses.
    Our main proposal for Congress is to adopt a comprehensive 
privacy and civil rights bill, which ideally would enshrine 
nine key concepts. Congress should limit text collection and 
use of our personal data, establish people's rights to control 
their own data, enhance data transparency, prevent 
discrimination by algorithms, increase platforms transparency 
about the known impacts of their business models, protect 
whistleblowers and external researchers, expand Federal Trade 
Commission oversight, encourage collaboration across agencies 
that hold specialized expertise, and set a Federal floor for 
consumer protection, not a ceiling. Fortunately, many of these 
elements are included in Senator Markey's Algorithmic Justice 
and Online Transparency Act, which prohibits algorithms that 
discriminate based on protected characteristics, and 
establishes safety and effectiveness standards.
    And in the Consumer Online Privacy Rights Act, sponsored by 
Senators Cantwell, Schatz, Klobuchar, and Markey, which 
penalizes platforms that abuse personal data, allows people to 
see the information companies collect on them, and preserves a 
Federal private right of action. Free Press Action endorses 
both bills. But in addition to privacy and civil rights 
safeguards, we must invest in the media system we need to 
support a 21st century democracy.
    Senator Cantwell's 2020 journalism report found local 
journalism is essential for healthy communities, competitive 
marketplaces, and a thriving democracy. I agree. Free Press 
Action and the Disinfo Defense League have proposed that 
Congress pass legislation to tax digital advertising and direct 
those monies to support high quality, noncommercial, and local 
journalism, including journalism by and serving people of 
color, non-English speakers, and other minority groups.
    Let us make no mistake, tech companies are undermining 
public health, safety, civil and human rights, and our 
democracy, and they are doing so by extracting and exploiting 
our personal data. It is time for Congress and the FTC to act. 
Thank you, and I look forward to your questions.
    [The prepared statement of Ms. Gonzalez follows:]

Prepared Statement of Jessica J. Gonzalez, Co-Chief Executive Officer, 
                           Free Press Action
    Chairman Lujan and Ranking Member Thune, Chairwoman Cantwell and 
Ranking Member Wicker, and other esteemed members of the Subcommittee: 
Thank you for inviting me to appear before you, and for seeking Free 
Press Action's views on how to rein in online platforms' harmful 
practices.
    Today's hearing calls us to address the harms of ``persuasive 
technology.'' Of course persuasive media is nothing new. For instance, 
as Free Press' Media 2070 project has documented, there is a long 
history of the U.S. media system being used to spread racist propaganda 
that legitimized and advanced the subjugation of Black people.\1\ Yet 
the U.S. media system has been used for persuasion in the public 
interest as well, like for suicide prevention, tobacco warnings and 
other public-service announcements that support public health and 
safety.
---------------------------------------------------------------------------
    \1\ See Joseph Torres, et al., ``Media 2070: An Invitation to Dream 
Up Media Reparations,'' Free Press (Oct. 6, 2020), https://
mediareparations.org/wp-content/uploads/2020/10/media-2070
.pdf.
---------------------------------------------------------------------------
    What's unique about media persuasion in the digital age is how 
online platforms collect troves of private demographic and behavioral 
data about us all and then use it to target us with ads, 
recommendations, and other content based on our perceived interests and 
vulnerabilities. That data collection is pervasive, and may be invasive 
and extractive. Years of research, investigative journalism, activism 
and whistleblower revelations have illuminated that the companies 
collecting that data have little to no accountability to the American 
people. Indeed they have failed outright at fulfilling even the most 
basic requests for transparency about how their systems work and what 
they know about us. They haven't even attempted to redress the harms 
caused by their business models, which promote content to maximize 
engagement and profit handsomely from amplifying conspiracy theories 
and bigotry.\2\
---------------------------------------------------------------------------
    \2\ Jeremy B. Merrill and Will Oremus, ``Five points for anger, one 
for a `like': How Facebook's formula fostered rage and 
misinformation,'' The Washington Post (Oct. 26, 2021), https://
www.washingtonpost.com/technology/2021/10/26/facebook-angry-emoji-
algorithm/.
---------------------------------------------------------------------------
    Another novel aspect here is that, in part due to the sheer volume 
of clickbait pushed to users and the decimation of quality local 
journalism, people are consuming content not in ecosystems but largely 
in their own echo chambers. We each are seeing vastly different 
things--predicated on online platforms' abusive data practices. The 
divide between us all in what we see (and when we are largely unaware 
of what others see) means that we simply don't have the ability to 
fully ascertain who is being persuaded by what. Only platforms have the 
data about who sees what information and how much they engage with it. 
That gross imbalance is even more in focus now because platforms like 
Facebook have been shown to be aware of the negative impacts of their 
design and have failed to adapt it to prevent and mitigate harms.
    Thus legislative and regulatory interventions ought to focus not on 
disrupting algorithms or shuttering persuasive technology per se but 
instead on mitigating its abuses and deterring the harms that companies 
employing it knowingly and negligently cause. This means prohibiting 
harmful data collection and use, including discriminatory algorithmic 
targeting, that runs afoul of civil and human rights, privacy rules, 
democratic norms, and public health and safety. Moreover, given how 
online platforms have polluted our information ecosystem, legislative 
approaches to invigorate local, independent journalism are ripe for 
examination.
Tech Companies Are Undermining Public Health, Safety, Civil and Human 
        Rights and Our Democracy--And They're Doing So By Extracting 
        and Exploiting Our Private Data
    Nearly every day we learn about another tech company creating or 
exacerbating deep societal harms. Facebook's prioritization of profit 
at the expense of human rights and public safety--not to mention its 
now well-documented efforts to cover up its misdeeds--is too extensive 
to summarize here. The past few months of blockbuster revelations have 
confirmed, however, that Facebook is unwilling to self-govern 
effectively. And Facebook is not alone. Many other prominent and more 
obscure tech companies are similarly compromised by their own self-
interests as they prioritize their profits over societal health.
    To date, we know that tech companies have facilitated, profited 
from and sometimes even participated in activities that harm our 
democracy and voting rights, public health and safety, and other civil 
and human rights. To cite a few of the most prominent examples:

   Dangerous COVID-19 and vaccine conspiracy theories have 
        proliferated over social media, exacerbating the public health 
        crisis, impacting our hospitals and leading to far more death 
        and serious illness than might otherwise have occurred. On 
        Facebook, African Americans, Native Americans, Latinx people 
        and other people of color were less likely to be shown credible 
        public health information than White people.\3\
---------------------------------------------------------------------------
    \3\ Corin Faife and Dara Kerr, ``Official Information about COVID-
19 Is Reaching Fewer Black People on Facebook,'' The Markup (Mar. 4, 
2021), https://themarkup.org/citizen-browser/2021/03/04/official-
information-about-covid-19-is-reaching-fewer-black-people-on-facebook.

   Facebook served discriminatory ads to Facebook users in 
        violation of the Fair Housing Act, prompting the U.S. 
        Department of Housing and Urban Development to bring a lawsuit. 
        Documents show that Facebook allowed advertising to target 
        users based on location and other identity markers that could 
        be proxies for protected categories, resulting in Black users 
        seeing less or no ads for affected housing. Research indicates 
        that Facebook continued to run such ads on its platform more 
        than a year after settling the initial complaint.\4\
---------------------------------------------------------------------------
    \4\ Tracy Jan and Elizabeth Dwoskin, ``HUD is reviewing Twitter's 
and Google's ad practices as part of housing discrimination probe,'' 
The Washington Post (Mar 28, 2019), https://www.washingtonpost.com/
business/2019/03/28/hud-charges-facebook-with-housing-discrimination/; 
Makena Kelly, ``Facebook still runs discriminatory ads, new report 
finds,'' The Verge (Aug. 20, 2020), https://www.theverge.com/2020/8/26/
21403025/facebook-discriminatory-ads-housing-job-credit-hud.

   Twitter and Facebook have consistently failed to remove or 
        flag as false content that discourages people from voting.\5\ 
        This content includes deception (lying about the time, place, 
        and manner of voting); calls for boycott from individuals with 
        alleged sponsorship ties to foreign state actors; and voter 
        intimidation or threats, such as claims that people will show 
        up to polling locations with guns.\6\
---------------------------------------------------------------------------
    \5\ See, e.g., Young Mie Kim, ``Voter Suppression Has Gone 
Digital,'' The Brennan Center for Justice (Nov. 20, 2018), https://
www.brennancenter.org/our-work/analysis-opinion/voter-suppression-has-
gone-digital; see also Ian Vandewalker, ``Digital Disinformation and 
Vote Suppression,'' The Brennan Center for Justice (Sep. 2, 2020), 
https://www.brennancenter.org/our-work/research-reports/digital-
disinformation-and-vote-suppression.
    \6\ See Kim, supra n.5.

   Voter suppression campaigns over social media included 
        surgical efforts to dissuade Black, Indigenous, and Latinx 
        voters from turning out to the polls. False election and 
        polling messages in non-English languages were much less likely 
        to be taken down or flagged than similar messages in 
        English.\7\ Non-partisan organization Protect Democracy found 
        that ``[s]ocial media platforms were plagued by false content 
        about various candidates for office, patently untrue 
        information about electoral processes, systematic efforts to 
        amplify bogus claims about voter fraud, and coercive political 
        messaging tied to COVID-19 conspiracy theories. A great deal of 
        this content targeted marginalized communities and, in 
        particular, communities of color.'' \8\
---------------------------------------------------------------------------
    \7\ Samuel Woolley and Mark Kumleben, ``At The Epicenter: Electoral 
Propaganda in Targeted Communities of Color,'' Protect Democracy (Nov. 
2021), https://protectdemocracy.org/project/understanding-
disinformation-targeting-communities-of-color/#section-1 (``In Georgia, 
African Americans and Hispanic Americans were on the receiving end of 
sophisticated microtargeting efforts erroneously claiming that then-
Senate candidate Raphael Warnock ``celebrated'' Fidel Castro. In 
Arizona, Hispanic American and Native American communities faced a 
cascade of untrue digital messaging over Twitter about the voting 
process. In Wisconsin, multiple communities of color from Madison to 
Milwaukee were targeted with lies about mail-in ballot fraud and ballot 
dumping.'' (internal citations omitted)).
    \8\ Id.

   Internal Facebook research from 2019 brought to light by 
        whistleblower Frances Haugen found that there was a concerted 
        effort to discourage Latinx people from participating in the 
        U.S. Census.\9\ The company's research summarized posts 
        ``telling Hispanic[s] to not fill out the form; telling 
        Hispanics not to participate in answering questions about 
        citizenship; saying that people would be in danger of being 
        deported if they participated; implying the government would 
        `get' immigrants who participated; and discouraging ethnic 
        groups from participating.'' \10\
---------------------------------------------------------------------------
    \9\ Brian Contreras and Maloy Moore, ``What Facebook knew about its 
Latino-aimed disinformation problem,'' LA Times (Nov. 16, 2021), 
https://www.latimes.com/business/technology/story/2021-11-16/facebook-
struggled-with-disinformation-targeted-at-latinos-leaked-documents-
show.
    \10\ Id. (internal quotation marks omitted).

   Facebook has allowed alleged discriminatory employment ad 
        targeting on the basis of gender and age.\11\
---------------------------------------------------------------------------
    \11\ Noam Scheiber, ``Facebook Accused of Allowing Bias Against 
Women in Job Ads,'' New York Times (Sep. 18, 2018), https://
www.nytimes.com/2018/09/18/business/economy/facebook-job-ads.html.

   Google's algorithms drive discriminatory search results, 
        pushing users to image search results that under-represent 
        women and women of color.\12\ Research from 2018 showed its 
        search terms related to Black girls mostly led to pornography, 
        even when terms like `` `porn,' `pornography,' or `sex' were 
        not included in the search box.'' \13\
---------------------------------------------------------------------------
    \12\ Xavier Harding, ``Breaking Bias: Search Engine Discrimination? 
Sounds About White,'' Mozilla Foundation (Sep. 28, 2021), https://
foundation.mozilla.org/en/blog/breaking-bias-search-engine-
discrimination-sounds-about-white/.
    \13\ Dr. Safiya Noble, ``Google Has a Striking Bias Against Black 
Girls,'' TIME (Mar. 26, 2018), https://time.com/5209144/google-search-
engine-algorithm-bias-racism/.

   In Free Press Action's statement for the record for this 
        subcommittee's hearing ``Shot of Truth: Communicating Trusted 
        Vaccine Information,'' we documented how online platforms and 
        broadcasters have both shirked their responsibility to the 
        public and failed to mitigate harms for their roles in 
        spreading dangerous disinformation that is designed to target 
        Black, Latinx, AAPI, Indigenous, and other communities of 
        color, as well as non-English speakers.\14\
---------------------------------------------------------------------------
    \14\ See Carmen Scurato and Jessica J. Gonzalez, Written Testimony 
for the Record, ``Shot of Truth: Communicating Trusted Vaccine 
Information,'' U.S. Senate Committee on Commerce (Apr. 15, 2021), 
https://www.freepress.net/sites/default/files/2021-04/
free_press_action_written
_testimony_shot_of_truth_hearing.pdf.

    As startling as these examples are, Free Press Action finds just as 
troubling the recent revelations that Facebook and other tech companies 
have long known exactly the extent to which their products were harming 
specific users and groups, and the lengths to which they have gone to 
cover up these harms rather than acknowledging and addressing them. We 
now know that Facebook refused to heed countless warnings from its own 
employees.\15\ Facebook's top decision makers have brazenly and 
routinely lied to or withheld the full truth from me and other civil 
rights leaders, in meetings between company executives and leaders from 
the Stop Hate for Profit campaign, Change the Terms coalition, and 
Spanish-Language Disinformation coalition. Facebook has done the same 
to researchers, the U.S. Congress, and the American public. Facebook 
spent vast amounts of time and money running public relations campaigns 
to obfuscate the truth and pretend to fix its problems, instead of just 
investing to improve the integrity of its systems. And Facebook is far 
from the only bad actor here. In a meeting last year with Change the 
Terms coalition leaders, many YouTube executives showed an appallingly 
shallow grasp of civil and human rights.
---------------------------------------------------------------------------
    \15\ For example, the L.A. Times reported that the

    leaked documents reveal substantial disagreement among staff about 
all sorts of issues plaguing the firm, with misinformation prominent 
among them.

    The 2020 product risk assessment indicates one such area of 
dissent. After noting that Spanish-language misinformation detection 
remains ``very low-performance,'' the report offers this 
recommendation: ``Just keep trying to improve. Addition of resources 
will not help.''

    Not everyone was satisfied with that answer. ``For misinfo this 
doesn't seem right . . . curious why we're saying addition of resources 
will not help?,'' one employee asked in a comment. ``My understanding 
is we have 1 part time [software engineer] dedicated on [Instagram] 
detection right now.''

    A second comment added that targeted misinformation ``is a big gap. 
. . . Flagging that we have zero resources available right now to 
support any work that may be needed here.'' (Redactions make it 
impossible to tell whether the same employee was behind both comments.)

    In communications with the outside world, including lawmakers, the 
company has stressed the strength of its Spanish-language content 
moderation rather than the concerns raised by its own employees.

    See Contreras & Moore, supra n.10.
---------------------------------------------------------------------------
Where Do We Go From Here?
    Given the substantial challenges outlined above, how do we reset 
our information ecosystem to advance democracy, to prevent 
discrimination and abuse, and protect public health and safety? We need 
a comprehensive approach. Fortunately, there are solutions to the 
challenges we collectively face, and at least three of them have been 
introduced in legislation before this body already: The Fourth 
Amendment is Not for Sale Act (S. 1265), the Algorithmic Justice and 
Online Transparency Act (S. 1896), and the Consumer Online Privacy 
Rights Act (S. 3195).
Data Privacy and Civil Rights Legislation
    Two days ago, in collaboration with three dozen other non-profit 
organizations--including Access Now, Common Cause, Demos, Cybersecurity 
for Democracy, and MediaJustice--that comprise the Disinformation 
Defense League, Free Press Action released a policy platform \16\ 
designed to rein in technology companies' abuses. Our main proposal for 
Congress is to adopt a comprehensive privacy and civil rights bill,\17\ 
which ideally would enshrine nine key concepts.
---------------------------------------------------------------------------
    \16\ Disinfo Defense League, Policy Platform (Dec. 7, 2021), 
https://www.disinfodefense
league.org/policy-platform. The Disinfo Defense League (DDL) is a 
distributed national network of organizers, researchers and 
disinformation experts disrupting online racialized disinformation 
infrastructure and campaigns that deliberately target Black, Latinx, 
Asian American/Pacific Islander and other communities of color. DDL was 
created by and for these communities.
    \17\ See Free Press Action and Lawyer's Committee for Civil Rights 
Under Law, ``The Online Civil Rights and Privacy Act of 2019'' (Mar. 
14, 2019), https://www.freepress.net/sites/default/files/2019-03/
online_civil_rights_and_privacy_act_of_2019.pdf.
---------------------------------------------------------------------------
    1. Congress should limit online platforms' and data brokers' 
collection and use of our personal data. Users should be able to 
control how apps use our data. We may want to share our data to receive 
services we sign up for, but apps should be prohibited from collecting 
more information than they need from us and from surreptitiously 
tracking us across the web. For example, the information we hand over 
for one reason--like providing a phone number for security purposes--
shouldn't be shared or sold to third-party companies.
    2. Congress should establish individuals' rights to control our own 
data. Everyone should have rights to easily access, correct, delete or 
download their personal information and take it with them when they 
leave an online service. Making data portable by law would let people 
free themselves from a corporate walled garden and easily use other 
services. These rights should apply equally to users across languages.
    3. Congress should enhance data transparency. We deserve to know 
what kinds of information companies and data brokers are collecting 
about us, and there need to be strict safeguards on what is off limits. 
Data brokers gather incredibly private details like individuals' sex, 
age, gender, geolocation, and health information; they can also collect 
internet-search histories that reveal even more sensitive information 
like a visit to a mental-health facility or house of worship. Companies 
need to disclose not just what information they collect, but where they 
get the information; who shares data with them, and with whom they 
share data; how they analyze data to profile us; how else they use our 
information; how they make decisions about what content, goods or 
services to offer us; and how they secure our data.
    Congress should close loopholes in existing privacy law by banning 
law enforcement from purchasing this information from data brokers 
without a warrant,\18\ and companies should conduct routine audits for 
bias, including opportunities for independent analysis of algorithmic 
bias, as well as privacy assessments to determine the risks of their 
data collection. And companies should be required to convey all of this 
information in two different ways: in an easy-to-understand format 
proactively notifying users, and in a detailed format for regulators, 
advocates and researchers for regular review.
---------------------------------------------------------------------------
    \18\ Passing The Fourth Amendment Is Not for Sale Act would be an 
excellent start on preventing platforms and data brokers from selling 
people's personal information to law enforcement and intelligence 
agencies without a warrant or any other court oversight. This 
bipartisan bill led by Sens. Wyden and Paul was co-sponsored by 
eighteen Senators at introduction in April, including members of this 
subcommittee Sens. Lee, Markey, Baldwin, Schatz, Tester, Blumenthal, 
and Commerce Committee Chairwoman Cantwell.
---------------------------------------------------------------------------
    4. Congress should prevent discrimination by algorithms. Everything 
we do online generates data, and every bit of that data can be tracked 
and used--no matter how innocuous it may appear in isolation--to create 
dangerous and invasive online profiles. Data feeds powerful algorithms 
to deliver personalized ads, recommendations and other services. There 
are some beneficial and harmless uses of these mechanisms, especially 
when robust transparency and user control are present. But Congress 
should ban algorithms that profile users and target content in ways 
that constitute unlawful discrimination in employment, housing, 
lending, and e-commerce on the basis of age, race, sex and other 
protected categories. Congress and relevant Federal agencies should 
investigate voting and other civil rights violations that flow from 
abusive data practices too.
    5. Congress should increase platform transparency about known 
harmful impacts of their business models. Reporting over the past 
several years has demonstrated that--just as the tobacco companies knew 
that their products were killing people long before the public was made 
aware--social-media companies knew how their business models were 
harming people and communities long before the details came to light. 
Companies should be required to provide access to researchers and to 
immediately disclose the harm when they learn that a platform's 
algorithms are being used to discriminate against or otherwise harm 
people; and the companies should actively and in an ongoing manner 
mitigate those harms and be held accountable for any persisting harms.
    6. Congress should protect whistleblowers and external researchers. 
We must protect whistleblowers who come forward to expose unethical, 
immoral, illegal and discriminatory behaviors, algorithms and practices 
inside of tech companies. Protecting whistleblowers from retaliation, 
labor law violations, baseless lawsuits, and targeted harassment is 
critical. We must also set out explicit protections for external 
researcher access to platform data to guard against what is now a 
documented pattern of targeted efforts by platforms to deny external 
researchers the opportunity to investigate platform practices.\19\
---------------------------------------------------------------------------
    \19\ Platform efforts to cut external researcher access to their 
data have been well-documented. Documentation includes recent testimony 
by NYU's Cybersecurity for Democracy initiative before the Subcommittee 
on Investigations and Oversight, U.S. House Science, Space, and 
Technology Committee. See ``Testimony of Laura Edelson, NYU 
Cybersecurity for Democracy,'' Hearing on ``The Disinformation Black 
Box: Researching Social Media Data,'' Sept. 28, 2021. (``Unfortunately, 
researchers have been severely hampered not just by lack of such data, 
but also outright hostility from platforms toward our research. Indeed, 
this summer, Facebook cut off my team's access to their data. We used 
that very data to support the finding in our recent study that posts 
from misinformation sources on Facebook got six times more engagement 
than factual news during the 2020 elections, to identify multiple 
security and privacy vulnerabilities that we have reported to Facebook, 
and to audit Facebook's own, public-facing Ad Library for political 
ads.''), https://medium.com/cybersecurity-for-democracy/testimony-of-
laura-edelson-nyu-cyberse
curity-for-democracy-e0b7e046eb8.
---------------------------------------------------------------------------
    7. Congress should expand Federal Trade Commission oversight. The 
FTC should have the power and resources to conduct rulemakings and 
effectively enforce against and prevent data abuses and other unfair or 
deceptive practices. Congress cannot anticipate and legislate against 
all future uses and abuses of data that companies may engage in, so 
lawmakers should enable the FTC to oversee and respond to future 
violations. For instance, users shouldn't have to waive our privacy, 
quality of service, or other rights by surrendering unnecessary data 
just to access a given service when there's no need for that extraneous 
data to deliver the promised service.
    8. Congress should encourage collaboration across agencies that 
hold specialized expertise. Federal agencies such as the Consumer 
Financial Protection Bureau, Department of Education, Department of 
Labor, Department of Justice and Department of Veterans Affairs, among 
others, should study how personal information is used in their fields, 
identify disparities and risks for discrimination, and issue public 
reports to Congress on a regular basis with special focus on the 
discriminatory effects on communities of color and non-English speaking 
groups.
    9. Congress should set a floor for consumer protection, not a 
ceiling. A Federal law must refrain from pre-empting the work that 
states are doing to build their own consumer-protection or privacy 
regimes. Many state consumer-protection laws are used to protect 
marginalized communities. A Federal data-privacy law that broadly 
preempts state laws and weakens these kinds of protections would 
jeopardize civil rights.
    Fortunately, two solid legislative vehicles that address many of 
these core concepts have already been introduced. Free Press Action 
endorses S. 1896, Senator Markey's Algorithmic Justice and Online 
Transparency Act. We're particularly interested in this approach 
because it:

   Defines as places of public accommodation ``any commercial 
        entity that offers goods or services through the Internet to 
        the general public,'' then prohibits algorithms that 
        discriminate based on race, age, gender, disability status and 
        other protected characteristics.

   Establishes safety and effectiveness standards for 
        algorithms.

   Mandates that platforms provide simple explanations of 
        algorithmic processes and the data they collect to power them.

   Requires that platforms submit detailed records about their 
        algorithmic processes for review by the FTC to ensure 
        compliance with key privacy practices.

   Compels platforms to publish transparency reports about 
        their content moderation efforts.

   And creates an inter-agency task force to investigate 
        algorithmic discrimination across sectors.

    Free Press Action also endorses S. 3195, the Consumer Online 
Privacy Rights Act, introduced by Senator Cantwell and co-sponsored by 
Senators Schatz, Klobuchar and Markey. We're supportive of this 
approach because it:

   Penalizes platforms that abuse people's personal data.

   Restricts Internet companies and data brokers from 
        processing or transferring data based on an individual's or 
        group's ``actual or perceived race, color, ethnicity, religion, 
        national origin, sex, gender, gender identity, sexual 
        orientation, familial status, biometric information, lawful 
        source of income, or disability'' in ways that violate people's 
        civil rights and access to economic opportunities.

   Allows people to see the personal information that companies 
        have collected on them.

   Establishes a Federal private right of action.

   And sets a floor, not a ceiling, for consumer protection.
Reforms to Section 230, If Any, Should Be Narrow and Targeted
    While Senators on both sides of the aisle have proposed changes to 
Section 230 as a way to reform platforms' harms both real and 
perceived, Free Press Action believes that any changes to this law 
should be narrow and targeted, ideally focused on clarifying the 
portions of the statute that courts have interpreted too broadly to 
preclude consideration of platform liability even for their own bad 
conduct.
    The Senate bill that feels closest to hitting the mark in our view 
is S. 797, Senators Schatz and Thune's PACT Act. The transparency 
provisions in that bill are especially strong. Important too is the 
bill's language suggesting that platforms could be liable for the harms 
that their continued distribution of harmful material causes once they 
have actual knowledge of those harms, even if we preserve as we should 
these online intermediaries' initial protection against being treated 
as publishers for making user-generated content available to the public 
in the first instance.
    As Free Press Action testified in a hearing before the U.S. House 
of Representatives last week, lawmakers should account for six precepts 
in any legislative discussions about Section 230:

  1.  Any Section 230 reforms should strike a balance by preserving low 
        barriers to the distribution of benign and beneficial content, 
        yet allowing platforms to be held accountable for their own bad 
        acts, such as knowing distribution of content adjudicated to be 
        unlawful or otherwise actionable, as well as other conduct, 
        content, or defective design and distribution choices of the 
        platform's own making.

  2.  Congress should reject any suggestion of a full repeal or 
        effective evisceration of Section 230. A repeal would raise 
        barriers to speech and chill expression by promoting excessive 
        takedowns, possibly shuttering entire sites and services, and 
        disproportionately shutting out people of color and other 
        already-marginalized speakers.

  3.  Section 230 reforms should apply across the board, not just to 
        ``big tech'' companies, because much harmful and abusive 
        activity happens on smaller platforms too.

  4.  The best path for Section 230 reform, in our view, would be to 
        clarify that the plain text of Section 230 does not immunize 
        ``interactive computer services'' for their own actions beyond 
        liability for ``publishing'' information others provide, or 
        removing or restricting access to such information that the 
        platform considers objectionable. Platforms' use of algorithms 
        to distribute content when they have knowledge of its harms, or 
        their monetization of engagement with that content, would be 
        factors in determining ultimate knowledge and liability but 
        would not automatically turn off 230s protections.

  5.  Even if Congress significantly altered Section 230, much of the 
        speech the members are (legitimately) concerned about would 
        still be protected by the First Amendment and otherwise 
        unactionable on the basis of existing tort law.

  6.  That means Section 230 reform is not the only or even the most 
        effective way to stem the tide of harm that online platforms 
        are facilitating, and the types of privacy law enhancements and 
        enforcement of existing laws described above, and efforts to 
        rebuild our media system described below, are essential 
        components for holding big tech accountable and combatting the 
        harms that platforms pose to society.\20\
---------------------------------------------------------------------------
    \20\ See Matthew F. Wood, Written Testimony for the Record, 
``Holding Big Tech Accountable: Targeted Reforms to Tech's Legal 
Immunity'' (Dec. 1, 2021), https://www.freepress.net/sites/default/
files/2021-12/
matthew_wood_written_testimony_holding_big_tech_accountable.pdf.
---------------------------------------------------------------------------
Building The Media System We Need To Protect Democracy In A 21st 
        Century Society
    Privacy and civil rights legislation is a critical part of 
mitigating the harms caused and perpetuated by social media platforms' 
failures. But we also must invest in alternatives to the current media 
system that create healthy pathways for online users to access 
information. Local news can be a powerful antidote to digital harms 
such as disinformation, further reinforcing the need to build up local, 
noncommercial media to correct course on these systemic and compounding 
crises. Research spanning the last decade underscores that local 
journalism is essential, a public good in fact, for a thriving 
democracy.\21\ And as the presence of local journalism withers in 
communities, research has shown an increase in corporate malfeasance 
and a decrease in civic engagement and voting. Meanwhile, Americans are 
desperate for local media. In the COVID-19 era, people in the U.S. have 
turned to local news and local TV more than ever. Twenty-three percent 
of Americans prefer local media over national,\22\ and most American 
adults trust local media more than national outlets.\23\
---------------------------------------------------------------------------
    \21\ See, e.g., PEN America, ``Losing the News: The Decimation of 
Local Journalism and the Search for Solutions,'' (Nov. 20, 2019) 
(``According to one analysis, the act of reading a newspaper alone can 
effectively encourage 13 percent of non-voters to vote.''); see also 
Meghan E. Rubado and Jay T. Jennings, ``Political Consequences of the 
Endangered Local Watchdog: Newspaper Decline and Mayoral Elections in 
the United States,'' 56 Urban Affairs Review 1327 (2019), doi.org/
10.1177/1078087419838058; Danny Hayes and Jennifer L. Lawless, ``As 
Local News Goes, So Goes Citizen Engagement: Media, Knowledge, and 
Participation in U.S. House Elections,'' 77 The Journal of Politics 447 
(2015), dx.doi.org/10.1086/679749.
    \22\ Pew Research Center, ``Local news is playing an important role 
for Americans during COVID-19 outbreak'' (July 2, 2020), https://
www.pewresearch.org/fact-tank/2020/07/02/local-news-is-playing-an-
important-role-for-americans-during-covid-19-outbreak/.
    \23\ Pew Research Center, ``For Local News, Americans Embrace 
Digital but Still Want Strong Community Connection'' (March 26, 2019), 
https://www.pewresearch.org/journalism/2019/03/26/for-local-news-
americans-embrace-digital-but-still-want-strong-community-connection/.
---------------------------------------------------------------------------
    Senator Cantwell's 2020 report, Local Journalism: America's Most 
Trusted News Sources Threatened, found that ``local journalism is 
essential for healthy communities, competitive marketplaces, and a 
thriving democracy.'' I wholeheartedly concur. Yet, we find ourselves 
in a local journalism crisis. For instance, in the past fifteen years 
the newspaper industry has lost 55 percent of its reporting jobs, 
leaving us with a ``reporting gap.'' \24\
---------------------------------------------------------------------------
    \24\ See S. Derek Turner, ``How Big is the Reporting Gap?,'' Free 
Press (June 2020), https://www.freepress.net/sites/default/files/2020-
06/free_press_reporting_gap_analysis_report.pdf.
---------------------------------------------------------------------------
    Yet even in its heyday the corporate media structure never 
sufficiently met the information needs of all people in the United 
States. The industry has a long and sordid history of racism and 
exclusion. Indeed in 1969 the DOJ's Community Relations Service 
identified that ``[f]ew American institutions have so completely 
excluded minority group members from influence and control as have the 
news media. This failure is reflected by general insensitivity and 
indifference and is verified by ownership, management, and employment 
statistics.'' \25\ As Free Press' Media 2070 project has documented, 
the ``white dominant press has used the power of racist narratives to 
subjugate, punish and control Black bodies and perpetuate white 
supremacy--both intentionally and unintentionally. Controlling 
narrative is about maintaining power. And that power has been wielded 
against Black and other Indigenous and colonized people to launch 
disinformation media campaigns from colonial times to the present.'' 
\26\ Today many of these problems persist. This past summer, Reps. 
Jamaal Bowman, Yvette Clarke, Brenda Lawrence, and twenty-two other 
members of the House of Representatives wrote in a letter to FCC 
Chairwoman Jessica Rosenworcel that
---------------------------------------------------------------------------
    \25\ 1969 Annual Report of the Community Relations Service, 
Department of Justice, U.S. Government Printing Office, at 22 (1969).
    \26\ See, e.g., Torres, et al., supra n.1, at 24.

        Today, people of color own and control just 6 percent of our 
        Nation's full-power TV stations, 7 percent of commercial FM 
        radio stations and 12 percent of commercial AM radio stations 
        despite making up more than 40 percent of the U.S. population. 
        As of 2017, Black Americans owned or controlled less than 1 
        percent of television stations. Although many journalists and 
        artists of color have used their talent to ensure critical 
        stories about their communities are being told, our Nation's 
        big media companies nevertheless continue to stereotypically 
        depict people of color as being a threat or a burden to 
        society. Historic Federal policies are a primary reason why 
        structural inequities exist in our Nation's media and 
        telecommunication systems today. FCC policies, license 
        decisions and inaction have had the result of effectively 
        excluding people of color from media ownership opportunities. 
        Our nation's first radio and TV licenses were awarded by the 
        Federal Radio Commission and then its successor, the FCC, 
        during an era of Jim Crow segregation. The previous 
        administration's efforts to consolidate the media marketplace 
        limited ownership opportunities for people of color and 
        women.\27\
---------------------------------------------------------------------------
    \27\ Letter from Rep. Jamaal Bowman et al. to Acting FCC Chair 
Jessica Rosenworcel (June 28, 2021), https://bowman.house.gov/_cache/
files/6/5/65b9a1a7-3553-4d5b-9d69-b8a92c0e628d/
3290E56EAC603E81B4CFE58A5DAEBEF1.0628-congressional-letter-on-fcc-
racial-equity-assessment-final.pdf.

    In sum, there are multifaceted causes for this reporting gap, and 
traditional media and now social media alike have been used to exclude, 
marginalize, and harm people of color in particular. But while these 
problems are not new to platforms nor wholly traceable to them, Free 
Press Action has proposed, and the Disinformation Defense League has 
endorsed, that Congress pass legislation to tax digital advertising and 
direct those monies to support high-quality noncommercial and local 
journalism. To fund local journalism, Congress could levy a small tax 
on the online-advertising revenues of large online platforms. For 
example, a 2 percent tax would yield more than $2 billion annually for 
a national endowment to support local news and information, including 
journalism by and serving people of color, non-English speakers, and 
other minority groups.\28\
---------------------------------------------------------------------------
    \28\ Free Press's Beyond Fixing Facebook paper offers ideas on how 
Congress could institute an online-advertising tax to support local 
journalism.
---------------------------------------------------------------------------
    Thank you and I look forward to your questions.

    Senator Lujan. Thank you, Ms. Gonzalez. Next, we will hear 
from Dr. Poulos, the Executive Director of the American Mind, 
Claremont Institute.

  STATEMENT OF JAMES POULOS, EXECUTIVE EDITOR OF THE AMERICAN 
                   MIND, CLAREMONT INSTITUTE

    Mr. Poulos. Good morning, Chairman Lujan and Ranking Member 
Thune, members of the Subcommittee. I am James Poulos, 
Executive Editor of the American Mind at the Claremont 
Institute, where my research focuses on preserving our shared 
way of life and form of Government in a digital age. It is an 
honor to speak today about how Congress can act. I will start 
by putting algorithmic harm in broader context and then make a 
few recommendations. The memes that algorithmic harm results 
from greedy CEOs hacking our minds fails to grasp the true 
nature of the digital crisis roiling America.
    As with digital programs and data centers, today the main 
purpose of algorithms is not to make money or influence 
thoughts, but to control people in a direct and alien way 
hostile to our core beliefs and principles. The digital medium 
is unlike prior media such as print or television. Those 
technologies also reshaped minds and built fortunes. But 
Americans always felt at home with them. There was a comfort 
level, a compatibility with our life ways and a regime that is 
absent regarding digital tech, which most Americans feel unable 
to understand, much less to master.
    As I show in my new book Human Forever, this morose 
incompetence is the product of a public, private partnership 
between unelected and unaccountable leaders across America's 
major institutions and our security and intelligence state.
    Almost all digital tech ordinary Americans use is the 
product of innovations from military and spy agency research, 
spun off into consumer entertainment and corporate craft by 
tech companies key to our national strategic infrastructure. 
These leaders whom citizens and even elected officials are 
functionally unable to remove from power, have moved so much of 
our political and social life into their technological 
ecosystem that they now make and enforce fundamental decisions 
about what we can and must think, say, and do.
    This lockstep reconfiguration of American life outside the 
reach of the democratic process has plunged us into a nascent 
social credit system. Social media's algorithmic harm is real, 
but social media is true to form a screen that obscures the 
depths below, where leaders like Eric Schmidt and Jeff Bezos 
are working to refound America as a control system built on 
innumerable swarms of programs and devices in a vast network of 
data centers. Through machine learning and AI, digitized 
governance automates the behavior of those swarms and through 
them, the behavior of you and me.
    The new system's logic is driven by seeing digital tech as 
better and stronger than humanity. As one ex-Google executive 
puts it, we humans ``suck'' in comparison to the new ``god'' 
our engineers are building. This view of our given humanity is 
a pathetic curse and not a precious gift is spreading because 
our leaders betrayed the expectations they created. Many 
believes when they told us the tech would bring global peace 
and harmony. The TV age ideal that whoever dreamed biggest, and 
best would rightly rule the world led to shock and panic when 
populists used digital tech to fight technocratic globalism.
    Today, our techno ethical elite is religiously convinced 
they can invent a coding language so deterministic that they 
can take true control of the digital swarms, eliminating the 
need for politics as it has been known in the West since 
Aristotle. They expect, will soon merge fully, with our 
technology becoming as gods. To make us yearn to upload our 
consciousness to the cloud, these new cyborgs theocrats have 
begun by uploading our conscience. Many Americans now think the 
culture war demands a digital regime, reward the pure and 
obedient and crush the opposition, online and off.
    This is why the COVID crisis has morphed so quickly into a 
pitched battle over who gets to act as judge, jury, and 
executioner when it comes to defining, preventing, and 
punishing harm. Now, representatives face a fateful choice, 
restore citizen controls of technology or surrender to the 
cyborg theocracy. Americans need Congress to intervene against 
the emergent social credit system.
    Trust in digital competence on the Hill can be built with 
bipartisan steps protecting children from the worst online 
harms. And lawmakers can protect free association and 
expression by favoring policies like algorithmic choice over 
calls to legislatively overturn Force v. Facebook, which would 
entail cosmic Federal choices about the metaphysics of harm 
that amount to an establishment of a religion.
    But unless ordinary Americans regain a hands on mastery of 
our most powerful digital tools, we will become compliant post-
humans or ungovernable psychotics, sacrificing what is left of 
our civilization and our Nation to vengeful new gods. Congress 
can best blunt digital harm by passing what I and others 
compare to a Second Amendment for computing.
    As I argued recently in the New York Times, legislation 
should enshrine Americans rights to buy and use high powered 
GPUs and to mine and hold bitcoin. This tech puts computation 
into human service, building apps and institutions where users 
create and exchange valuable, memorable works of culture.
    To model this approach, I published my book, Human, 
Forever, onto the blockchain at the Bitcoin based platform 
Canonic.XYZ. Americans have the ability right now to restore 
their practical use of technology to defend and protect all 
they hold sacred from the social credit borg. By recognizing 
the free exercise of that ability as a fundamental right of the 
digital age, lawmakers can save America and Congress from 
technological oblivion. Thank you. I will be happy to take 
questions.
    [The prepared statement of Mr. Poulos follows:]

 Prepared Statement of James Poulos, Executive Editor of the American 
                       Mind, Claremont Institute
    Good morning Chairman Lujan, Ranking Member Thune, Members of the 
Subcommittee. I'm grateful to join you. My name is James Poulos. I'm 
the Executive Editor of the American Mind at the Claremont Institute, 
where my research responsibilities focus on preserving our shared way 
of life and form of government in a digital age.
    It's an honor to speak today about how Congress can act. I'll start 
my remarks by putting algorithmic harm in a broader context. Then I'll 
make a few recommendations.
    The meme that algorithmic harm results from greedy CEOs hacking our 
minds fails to grasp the true nature of the digital crisis roiling 
America. The main purpose of algorithms, like digital programs and 
datacenters more broadly, is not to make money or influence thoughts, 
but to control people--in a direct and alien way hostile to our core 
beliefs and principles.\1\
---------------------------------------------------------------------------
    \1\ The objection could be raised that algorithms, in a certain 
strengthening sense, actually mainly exist so that digital devices and 
entities can communicate with one another. One reason digital 
technology is so alien to us is its indifference to our feelings and 
our existence alike, and our shared sense that digital tech which 
operated even in part based on its awareness of our presence and 
attitudes would be dangerous or difficult to establish trust with. 
Perceptive digital entities which did communicate openly with us might 
still ``talk behind our backs'' amongst themselves. In June 2019 
testimony before the U.S. Senate Commerce Committee's Subcommittee on 
Communications, Technology, Information, and the Internet,'' Stephen 
Wolfram observed that ``if we want to seriously use the power of 
computation--and AI--then inevitably there won't be a `human-
explainable' story about what's happening inside . . . if you can't 
check what's happening inside the AI, what about putting constraints on 
what the AI does? Well, to do that, you have to say what you want. What 
rule for balance between opposing kinds of views do you want? How much 
do you allow people to be unsettled by what they see? And so on.'' As 
Norbert Wiener helps us understand in The Human Use of Human Beings: 
Cybernetics and Society, even if algorithms and content-selecting AIs--
or whole swarms of digital entities--communicate almost exclusively 
with one another in an ignorance of us they cannot describe, our 
commands are still the inputs, and however imperfectly they are 
executed, the outputs must inevitably have human significance intended 
to impact (e.g., use) human beings.
---------------------------------------------------------------------------
    The digital medium is unlike prior communications technologies such 
as the printing press or television. Those media also reshaped minds 
and built fortunes. But Americans always felt at home with them. There 
was a comfort level and compatibility with our lifeways and our regime 
that's absent regarding digital technology, which most Americans feel 
hopelessly unable to understand, much less master.
    As I show in my new book Human, Forever, this morose incompetence 
is the product of a public-private partnership between unelected and 
unaccountable leaders across America's major institutions and our 
security and intelligence state. As economists such as David P. Goldman 
and Maria Mazzucato have reminded us, almost all the digital technology 
ordinary Americans use is the product of innovations attained through 
military and spy agency research and spun off into consumer 
entertainment and corporate cruft by the tech companies crucial to our 
national strategic infrastructure.
    However well-intentioned, these leaders, whom citizens and even 
elected officials are functionally unable to remove from power, have 
moved so much of our political and social life into their technological 
ecosystem that they now make and enforce fundamental decisions about 
what we can and must think, say, and do.
    This lockstep reconfiguration of American life outside the reach of 
the democratic process has plunged us into a nascent social credit 
system. Social media's algorithmic harm is real, but social media is, 
true to form, a screen that obscures the depths below--where leaders 
like Eric Schmidt are working to re-found America as a control system 
built on innumerable swarms of programs and devices in a network of 
vast datacenters. Through machine learning and artificial intelligence, 
digitized governance aims to automate the behavior of those swarms and, 
through them, the behavior of us human beings.\2\
---------------------------------------------------------------------------
    \2\ Innovation is palpably shifting, even within social media, away 
from writing algorithms and toward conducting swarms. Analyzing a 
leaked document from TikTok revealing the app's inner workings, UC-San 
Diego computer science professor Julian McAuley recently told Ben Smith 
of The New York Times that TikTok's advantage marries machine learning 
to ``fantastic volumes of data, highly engaged users, and a setting 
where users are amenable to consuming algorithmically recommended 
content (think how few other settings have all of these 
characteristics!). Not some algorithmic magic.''
---------------------------------------------------------------------------
    The new system is driven by the logic of seeing technology as 
better and stronger than humanity. As Mo Gawdat, an ex-Google executive 
on publicity tour, recently told The Times, we humans ``suck'' in 
comparison to the new ``god'' our tech engineers are building. This 
view of our given humanity as a pathetic curse, not a precious gift, is 
spreading because our leaders betrayed the expectations they created. 
Many believed what they told us about how tech would bring global peace 
and harmony. The TV-age belief that whoever dreamed biggest and best 
would rule the world and deserved to led to shock and panic when 
populists used digital tech to fight technocratic globalism.
    Today, our technoethical elite is religiously convinced they can 
discover a mathematical coding language so deterministic that they can 
take true control of the digital swarms, eliminating the need for 
politics as it has been known in the West since Aristotle.\3\ 
Eventually, they believe, we will fully merge with our technology and 
become ``as gods.''
---------------------------------------------------------------------------
    \3\ The politics of determinacy extend well beyond the elite. 
Debates rage over whether digital technology is somehow ``neutral'' or 
can be made so by policy. It cannot be, in two senses. Pure neutrality 
cannot be achieved by or through algorithms, which, as instructions to 
produce a certain result, are always inherently ``biased'' by default. 
``Correcting'' an algorithm means giving it a new and different bias. 
On the level of the medium, interoperability is the form of digital 
technology that shapes all digital entities. While the bias of digital 
tech in this sense is toward interoperability, the human bias is toward 
incommensurability. We may enjoy temporarily joining the crowd, the 
mass, or even the mob, the feeling passes and we return to abide in the 
unique and particular personhood of our self. Digital entities do not 
share this dynamic. Unlike us, they are biased toward the collective 
identity of the swarm, an identity incompatible with our human one. The 
quest for the Holy Grail of perfect determinacy depends on the faith 
that mathematics itself is neutral and unbiased in the sense of 
ultimately being perfectly legible and comprehensible--without 
secrets--to rational human minds. Quantum physics and millennia of 
Western theology agree that the truth is more complicated. Devotion to 
mathematics as the perfect language of true explanation is necessarily 
``biased against'' mysteries, against the need for or permanence of 
mystery, and against the idea that the primal condition of reality 
involves phenomena inaccessible to human logic.
---------------------------------------------------------------------------
    Rather than trying to upload our consciousness to the cloud, these 
new cyborg theocrats have begun by uploading our conscience. Many 
Americans now think the culture war must be fought and won through a 
digital regime that rewards the ethically pure and obedient and crushes 
the opposition online and off. This is why the covid crisis has morphed 
so quickly into a pitched battle over who gets to act as judge, jury, 
and executioner when it comes to defining, preventing, and punishing 
harm.
    Now, the people's elected representatives face a fateful choice: 
restore citizen controls of technology or surrender to the cyborg 
theocracy. Americans need Congress to intervene against the emergent 
social credit system. Trust in digital competence on the Hill can be 
built with bipartisan steps protecting children from the worst online 
harms.\4\ And lawmakers can protect Americans' free association and 
expression by favoring policies like algorithmic choice over calls to 
legislatively overturn Force v. Facebook, which would entail cosmic 
Federal choices about the metaphysics of harm that amount to the 
establishment of a religion.\5\
---------------------------------------------------------------------------
    \4\ Legislators should be prepared to discover that algorithms and 
human users often share joint responsibility for what emerges over time 
as accumulated harm. Writing about Instagram's algorithms in The 
Atlantic, Jonathan Haidt observes that ``the toxicity comes from the 
very nature of a platform that girls use to post photographs of 
themselves and await the public judgments of others,'' specifically, we 
should recognize, other girls posting photographs. Social media is a 
hotbed of mimesis, the reflexive behavior of imitating one's real and 
imagined rivals that social theorists from Rousseau to Girard have 
recognized as fundamental to our human identity. The strongest and most 
prudent legislative intervention against the experience of having 
suffered harm from engaging in habits reinforced by algorithm would be 
to legally protect and defend Americans' fruitful use of digital 
technologies, such as Bitcoin, which do not inflict algorithmic or 
mimetic harm in the manner of social media platforms because they are 
not social media platforms.
    \5\ The House Energy and Commerce Committee, for instance, recently 
discussed the ``Justice Against Malicious Algorithms Act,'' a bill that 
would amend Section 230 by allowing users to sue platform companies for 
inflicting ``severe emotional injury,'' but which did not define 
emotional injury. The ``Protecting Americans from Dangerous 
Algorithms'' bill introduced last session would, as Cato Institute 
policy analyst Will Duffield noted last year at Techdirt, ``have grave 
consequences for legitimate speech and organization . . . an 
omnipresent corrective authority''--spiritual and temporal together, 
like Hobbes' Leviathan--``would foreclose the sense of privileged 
access necessary to the development of a self.''
---------------------------------------------------------------------------
    But unless ordinary Americans regain a hands-on mastery of our most 
powerful digital tools, we will become compliant posthumans or 
ungovernable psychotics, sacrificing what is left of our civilization 
and nation to vengeful new gods. Congress can save our humanity, our 
country, and our form of government from digital harm by passing what I 
and others compare to a Second Amendment for Compute. Legislation 
should enshrine Americans' rights to buy and use high-powered GPUS and 
to mine, hold, and use Bitcoin. This tech puts computation into human 
service building apps and institutions where users create and exchange 
valuable, memorable works of culture.\6\
---------------------------------------------------------------------------
    \6\ Bitcoin is deeply resonant with American civilization. In no 
other country, especially leading country, has interest and activity in 
Bitcoin been so immediate, sustained, and powerful. Some countries, 
including China, have cracked down on Bitcoin or banned it outright. 
Given that digital technology's world dominance makes us reconsider 
venerable theological matters by causing us to question our identity 
and purpose, it seems important that the Bitcoin blockchain relies for 
the legitimacy of its architecture and operations on the deeply 
Protestant concept of Proof of Work. To activate the consensus that 
allows new blocks of information to be added on chain, Bitcoin miners 
must compete to solve a math problem. The achievement satisfying Proof 
of Work is not to have ``cracked the code'' but to have evinced the 
input of the most computational labor. Allegations that Bitcoin is 
therefore energy-intensive enough to represent an unjust harm to the 
natural environment fail on several fronts, including the relative 
energy consumed over a given period by China or the petrodollar, but 
especially on the ultimately theological basis of the idea of fair play 
through competitive labor that is a cornerstone of American 
civilization. Of course, many of those who insist we must leave 
theology behind in assessing the value of political or economic 
measures still retain, due to their own theological inheritance, an 
idea that fair play through the labor of competitive reasoning is a 
thoroughly secular standard of justice. Whether inflected in this more 
secular or the more theological key, the inner logic and structure of 
Bitcoin is at home in America, where the common sense is still that the 
unceasing labor of building and maintaining culture is the price of 
flourishing in freedom.
---------------------------------------------------------------------------
    To model this approach I published my book, Human, Forever, onto 
the blockchain, at the Bitcoin-based platform Canonic.xyz. Americans 
have the ability right now to restore their practical use of technology 
to defend and protect all they hold sacred from the maw of the social 
credit borg.\7\ By recognizing the free exercise of that ability as a 
fundamental right of the digital age, lawmakers can save Congress--and 
America--from technological oblivion.
---------------------------------------------------------------------------
    \7\ Notably, the hardware associated with mining and building on 
Bitcoin gives users the ability to freely generate algorithmic markets, 
which guide people within a technological ecosystem based on voluntary 
agency and not, as in a social credit system, mandatory compliance.

    Senator Lujan. Thank you very much. Dr. Eckles, our 
Associate Professor of Marketing at MIT Sloan School of 
Management. Dr. Eckles.

STATEMENT OF DR. DEAN ECKLES, ASSOCIATE PROFESSOR OF MARKETING, 
                 MIT SLOAN SCHOOL OF MANAGEMENT

    Mr. Eckles. Chairman Lujan, and Ranking Member Thune, and 
members of the Subcommittee, thank you for asking me to appear 
here today to discuss these important topics. I typically don't 
work on public policy directly, but I have been studying social 
media for over a decade, currently at MIT. I also previously 
worked in industry, including as a scientist at Facebook. I 
have a wide view of current practice and fundamental challenges 
in the context of social media.
    Social media have dramatically lowered the costs for anyone 
to share information and media with many people. It is so easy 
to propagate content that people often share things that on 
further consideration they would realize are misinformation. 
And various actors from legitimate marketers and political 
campaigns to spammers, scammers, and foreign disinformation 
efforts seek to maximize their reach and influence on these 
platforms. What role do algorithms have to play here? Yes, the 
algorithms show us content we are predicted to click on, to 
watch, and to propagate. But importantly, not just this.
    As sophisticated platforms, algorithmic ranking involves 
using many signals. This includes negative signals, signals 
that reveal we will regret engaging with the content and 
attitudinal signals like predictions of whether we would say 
the content is informative, important, or fun, if we were 
asked. These can serve as an important way of balancing against 
solely short run engagement signals.
    Quantifying the impacts of algorithmic ranking is quite 
difficult. The platforms themselves often struggle, despite 
their data, their randomized trials, and their expertise, to 
quantify some of the effects of the decisions they must make. 
This isn't just because of technical complexity, but due to 
people's complex and often strategic responses to changes in 
algorithms.
    First, everyday users adjust their behavior in response to 
algorithms. Say you are connected on social media to a relative 
who post a lot of political content that you disagree with. Due 
to algorithmic ranking, you don't see a lot of these posts. The 
platform accurately predicts you will not engage, and you 
wouldn't say they help you be informed. But you do see most of 
that person's posts about fishing.
    Now, this might seem like a clear case of a filter bubble, 
whereby the algorithm is causing you to be exposed to less 
cross-cutting content than you would under a simple 
chronological ranking. But consider this, under chronological 
ranking, you might initially see so many undesirable political 
posts from this person that you choose to unfollow them.
    So this could result in then seeing less crosscutting 
content and having fewer cross-partisan ties than you would 
under the algorithmic ranking. So was it really creating a 
filter bubble? What was really being amplified here? Second, 
more professional and strategic actors, marketers, politicians, 
publishers, et cetera, they have tuned what they post and when 
they post it to the status quo algorithm. If things change 
dramatically, they will adjust dramatically. Accounting for 
this is key to assessing what is truly being amplified by an 
algorithm.
    Overall, we lack clear evidence about broader aggregate 
harms or benefits of algorithmic ranking. Nonetheless, far from 
there being clear benefits from simple rankings like 
chronological or overall popularity rankings, it can make some 
forms of undesirable strategic behavior easier and more 
successful.
    I have seen some of this in my work on cross-platform 
coordinated campaigns to game Twitter trends during the 2019 
Indian elections. So what can policymakers do? Some of what I 
said suggests that some of the easy pass here won't work. For 
example, simple definitions of algorithmic amplification can be 
turned out to be hugely misleading.
    One basic thing policymakers can do, though, is to help 
scientists, journalists, and the public better understand these 
systems. First, they can protect the ability of external 
researchers to probe these systems, not unlike what staff here 
have done by creating accounts on Instagram and TikTok, but 
more systematically. While some court rulings have been 
encouraging, there has been substantial legal risk for 
researchers probing platforms to ask questions of broad 
interest. This can have chilling effects.
    Second, policymakers can provide clear paths for platforms 
to retain and share data in privacy preserving ways with 
researchers. There remains a great deal of uncertainty about 
what methods for anonymization of data satisfy some 
international privacy regulations like GDPR, likely 
discouraging sharing data with researchers.
    My understanding is that some platforms are not even 
retaining critical data for their own internal research, 
including on algorithmic impacts, because of such regulations. 
This suggests the value of care and clarity in crafting these 
regulations. I am encouraged by this committee's work in this 
area, and I look forward to your questions.
    [The prepared statement of Mr. Eckles follows:]

     Prepared Statement of Dr. Dean Eckles, Associate Professor of 
                              Marketing, 
                     MIT Sloan School of Management

 Algorithmic transparency and assessing effects of algorithmic ranking

    Contemporary communication technologies have dramatically lowered 
monetary and practical costs of broadcasting information and media to 
many people--and of consuming others' broadcasts. They have created new 
ways for people to share their own thoughts, experiences, and 
creations, to consume and react to those shared by others, and--quite 
importantly--to rapidly propagate them. It is so easy to propagate 
content that people often share information that, on further 
consideration, they themselves would realize is misinformation.\1\ 
Interactions on social media affect commerce, culture, politics, and 
public health,2-6 thereby reasonably attracting scientific, 
public, and regulatory attention.
    What role do algorithms play in all of this? Algorithms are 
unavoid-able here. Even sorting posts by friends in chronological 
order* or videos by overall popularity is algorithmic; and 
often it is unclear there is a single, simple baseline algorithm. What 
much of the public conversation about algorithms has in mind are 
particular kinds of more complex, potentially more opaque, algorithms--
typically based on statistical machine learning--that are adaptive to, 
e.g., features of the content and each person's history of consumption. 
In the context of social media, these algorithms typically first 
present items that are predicted to be something the consumer will take 
desirable actions on or would say is important, interesting, or fun.
---------------------------------------------------------------------------
    \*\ For simplicity I use ``chronological'' to refer to a ranking 
that shows items ordered by recency--perhaps more precisely called 
reverse chronological.
---------------------------------------------------------------------------
    Are these algorithms better or worse--for individual consumers and 
for society--than simpler alternatives that would, e.g., present 
everything from the accounts someone follows in chronological order? 
Can we straightforwardly specify what a given algorithm ``amplifies'' 
in a way suitable for assigning moral or legal responsibility?
    Here I briefly summarize some key points.

  1.  At established platforms, algorithmic ranking and recommendation 
        involve using many signals and are typically not aimed at 
        simply maximizing short-run engagement.

  2.  Quantifying the impacts of algorithmic ranking is quite 
        difficult, even with access to proprietary data. This is not 
        only because of the complexity of these technical systems, but 
        due to people's complex and often strategic responses to 
        changes in algorithms.

  3.  We lack clear evidence about broader benefits or harms of 
        algorithmic ranking. Nonetheless, simple rankings and 
        recommendations (e.g., chronological, overall popularity) can 
        make some forms of undesirable strategic behavior easier.

  4.  Policymakers can protect the ability of external researchers to 
        probe these systems, and they can provide clear paths for 
        platforms to retain and share data in privacy-preserving ways.

    The rest of this statement is organized as follows. First, I 
characterize current practice in algorithmic ranking and recommendation 
of content in social media. Second, I elaborate on how we can learn--
and what we already know--about effects of algorithms in social media. 
Third, I conclude by stating some policy implications.
Current practice
    Before considering assessing their effects or crafting policy, it 
can be useful to understand the state-of-the-art in algorithmic ranking 
and recommendation in social media. This involves choosing from a large 
collection of items (i.e., content, stories, posts, activity) that a 
user is eligible to see and determining which of those items are 
displayed in what order--and also how each item is displayed, as there 
are often multiple variations available. I start with a prototypical 
case, presenting a somewhat simplified solution, and then discuss a few 
relevant variations.
    Consider the problem of choosing which of a large set of photos or 
videos shared by accounts a user follows to display and in what order. 
This is a version of the problem faced by Instagram and Snapchat in 
their feeds (that is, setting aside their other channels for now). 
People do not typically spend enough time using these services to see 
all--or even the majority--of what is shared by the accounts they 
follow.7,8
    These platforms have multiple goals in mind when doing this 
ranking. They then attempt to quantify these goals in metrics, usually 
defined at the level of each user. These could include the fraction of 
days that users log in, the number of photos they post, their time 
spent using the service, what they would say in response to a survey 
question asking about the service, the revenue from them viewing or 
clicking on ads, etc. These may be combined into a single evaluation 
criterion,\9\ or managers may decide to try to maximize one metric 
while ensuring that they do not have substantial negative effects on 
others.
    It is difficult to directly optimize what items to show to maximize 
such metrics, since these metrics are defined at the level of users and 
there are many possible rankings of items.* So typically 
platforms simplify the problem by defining a small number of 
constituent scores for each item, defining a combined score that, e.g., 
sums up these scores, and then ranking items by that combined score 
(Figure 1).10,11
---------------------------------------------------------------------------
    \*\ For a single user, there often are more possible rankings of 
available items than there are atoms in the universe: say there are 100 
items available, then there are over 10156 possible 
orderings of those items.
---------------------------------------------------------------------------
    These constituent scores are typically predictions of actions the 
viewing user might take on or related to the item. For example, there 
could be predictions of the probability the viewer will ``like'' the 
item. Similar predictions can be made for many other 
item-level actions, such as commenting on it or viewing it for at least 
x seconds. Usually ``negative'' actions, such as unfollowing or 
unfriending the poster or hiding the item, are similarly predicted and 
given negative weights in the combined score.* Data 
scientists and managers try to identify new actions to predict that 
provide additional relevant signals (e.g., sequences of actions 
indicative of ``regretted'' clicks).
---------------------------------------------------------------------------
    \\ Ranking engineers at multiple firms can be overheard 
referring to ``p-like''.\12\
    \*\ That is, the combined score for an item will look something 
like score = wlike x P(like) + wcomment x 
P(comment) + . . . + whide x P(hide), where the ws are the 
(positive or negative) weights given to the predictions of different 
actions. Recent reporting makes clear TikTok's ranking likewise follows 
this pattern,\12\ though it apparently lacks both any negative signals 
or any signals from attitudinal (e.g., survey) data.
---------------------------------------------------------------------------
    Figure 1: Schematic representation of inputs to scores used to rank 
items in social media feed. Numerous characteristics of the item are 
used to predict the probability of the viewer taking various actions 
(e.g., ``liking'' the item) based on historical data. These predictions 
together determine a combined score; for example, the combined score 
might add up these predictions, each with a different weight.

[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]

    The same approach is often used to predict other, less direct, 
actions. For example, the platform may produce predictions of whether 
the viewer will follow or friend someone involved in the item (e.g., 
someone tagged in the photo). The general approach can also incorporate 
non-behavioral signals that are only available for a small sample of 
users, such as data from surveys. For example, if the platform has 
contractors or everyday users rating content in their 
feeds, they can predict those ratings; that is, the 
platform can produce predictions of whether, if asked, the viewer would 
say this item is important, informative, funny, or makes them feel 
connected.
---------------------------------------------------------------------------
    \\ Facebook, for example, created a paid ``feed quality 
panel'' to meticulously rate items in their feeds, and has combined 
this with data from users who respond to prompts to rate their 
feeds.\13\ These ratings can be used both to evaluate a prospective 
ranking algorithm and in the algorithm itself via predictions, as 
described here.
---------------------------------------------------------------------------
    How are the weights on these different constituent scores chosen? 
In some cases these weights may be selected through managerial 
judgement alone, but often sophisticated A/B tests= are used 
to compare how different choices of weights affect 
metrics.10,14,15 For example, the weights might be selected 
to maximize a metric designed to (ambitiously) measure people's level 
of ``meaningful social interaction'', subject to a constraint on 
revenue. While it is most straightforward to assess effects on metrics 
for the viewing user, many of the immediate consequences of ranking 
apply to other people (e.g., by showing this item to this viewer, the 
poster of that item may receive an additional comment).\16\ Platforms 
often try to incorporate these consequences into their ranking as well.
---------------------------------------------------------------------------
    \=\ A/B tests (i.e., randomized experiments, randomized controlled 
trials) involve randomly assigning users to different variations on a 
service. Here this would be assigning users to rankings using different 
weights on the predicted actions. As in medicine and economics, these 
randomized trials are considered the ``gold standard'' for evidence and 
decision-making.
---------------------------------------------------------------------------
    Recall the constituent scores are the probabilities of various 
actions; how do platforms predict these actions? They use very large 
statistical machine learning (or, if one prefers, artificial 
intelligence) models trained on historical data about which items 
people were shown, what the characteristics of those items where, and 
what the viewing user did.* For example, the historical data 
would generally reveal that if the viewer frequently comments on items 
involving a particular person (e.g., as the poster of the photo, as 
someone tagged in the photo), then they are more likely to do so for 
this new item. More recently, platforms have more thoroughly 
incorporated techniques from natural language processing and computer 
vision into these predictive models; the actual contents on the photo 
are used to predict what actions the viewer would take if shown 
it. Furthermore, the history of a user's 
interactions and connections in the social network is sometimes used to 
learn some numeric representation of their preferences and 
dispositions, which can also be used in these predictions; these can be 
regarded as latent (i.e., unobserved) positions of each user.
---------------------------------------------------------------------------
    \*\ Usefully, this historical data typically involves users seeing 
stories in an order other than the status quo because they are in an A/
B test or because the scores for items have had a small amount of 
random noise added to them.
    \\ That Facebook uses such signals can be seen in that ad 
delivery is immediately imbalanced by gender depending on whether the 
ad image alone includes stereo-typically male- or female-relevant 
themes.\17\
---------------------------------------------------------------------------
    Thus, the algorithmic ranking of social media feeds typically 
depends on numerous inputs. Some of these inputs themselves are the 
results of prior statistical machine learning (e.g., learned 
representations of the objects in photos). However, decision-makers 
within platforms are typically less focused on these inputs, but rather 
on understanding and effectively making tradeoffs between various 
metrics--defined not solely at the level of individual items to be 
ranked, but for individual users or for the entire service. They often 
make these tradeoffs by differently weighting predictions about how a 
viewer will act on or evaluate an item.
Further variations and complications
    There are some variations on and complications of the above problem 
and solution.

  1.  When deciding what item to show in the second position, it can be 
        useful to account for what the first item is; more generally, 
        it can make sense to consider the full ranked set of items so 
        as to reflect, e.g., demand for variety of topics or sources. 
        While it is typically computationally difficult to directly 
        optimize rankings in this way, platforms often implement 
        various heuristics to improve the final ranking. For example, 
        they may have rules that prevent a ranking from being highly 
        repetitive in having several items from the same poster, only 
        videos, etc.*
---------------------------------------------------------------------------
    \*\ This is sometimes described as the ``slate recommendation'' 
problem, including in research from YouTube and others.\18\ Facebook 
managers describe this as part of a ``contextual pass'' that, among 
other things, implements poster and content-type diversity rules.\11\

  2.  There are sometimes multiple ways to display an item. For 
        example, multiple photos posted by the same person could all be 
        shown smaller and grouped together, or some or all could be 
        displayed at a larger size and shown individually. Some or all 
        existing comments on an item could be displayed by default.\16\ 
        Thus, similar algorithms are also often used to make these 
---------------------------------------------------------------------------
        decisions.

  3.  What comprises the inventory of items to be ranked is not always 
        obvious, and it is often not limited to other accounts 
        broadcasting new content (e.g., posting a photo), but can 
        include other activity (e.g., a friend being tagged in a public 
        photo, a friend commenting on a photo by another friend). 
        Typically, many of these items score poorly or are removed 
        according to rules to avoid repetitiveness (e.g., several items 
        about friends commenting on the same photo), but some may score 
        well.

  4.  Even in the same platform there are numerous channels for 
        algorithmic ranking and recommendation. Public discourse often 
        has in mind a single feed (e.g., Facebook News Feed, TikTok's 
        series of videos), but these same items might be delivered to 
        users via multiple feeds and via e-mail or 
        mobile push notifications, with these likewise subject to 
        optimization.\19\ Established platforms are often ranking and 
        recommending many types of items in many different formats. For 
        example, many platforms suggest accounts to follow or friend 
        based on numerous signals and with the aim of getting new users 
        to make, e.g., engaging and varied connections.\20\ Many 
        platforms also have personalized search functionality in many 
        places, including some that might seem mundane or invisible 
        (e.g., autocompletion of friends' names when tagging them in 
        photos).
---------------------------------------------------------------------------
    \\ For some time starting in 2011, Facebook had both its 
News Feed and a second, chronological feed (Ticker) that updated in 
real-time; both were visible simultaneously when using Facebook on a 
computer.
---------------------------------------------------------------------------
Algorithmic transparency and impact: Evidence and challenges
    There is substantial interest in characterizing ``algorithmic 
amplification''--what content is given greater reach than it would have 
with some baseline algorithm--and the broader impact of algorithms. The 
title of this hearing refers to ``dangerous algorithms'' and posits 
associated harms. I find it quite plausible that, in particular cases, 
algorithmic ranking in social media is the proximal cause of specific 
harms; likewise, in particular cases, algorithmic ranking is the 
proximal cause of specific benefits and the absence specific harms. But 
how would we know whether aggregate effects of algorithmic ranking and 
recommendation in social media are positive or negative? To be clear, I 
do not think there is a social-scientific consensus here in favor of 
some simple baseline ranking (e.g., chronological) over the status quo.
    Ideally, we would like to have rigorous, quantitative evidence 
about the effects of algorithmic ranking. While there presently is not 
a large body of evidence, one hope is that algorithmic transparency and 
other efforts would enable expanding this evidence. Here I review ways 
we can learn about algorithmic ranking and recommendation, in the pro-
cess summarizing what limited evidence* we do have and 
highlighting some of the challenges in using simplistic comparisons of 
algorithms.\22\
---------------------------------------------------------------------------
    \*\ Further afield, there are detailed studies of predictive 
systems making biased and harmful decisions in, e.g., health care,\21\ 
a setting where both quite different regulatory considerations apply 
and some of the challenges described above as less severe.
---------------------------------------------------------------------------
    I see three common ways we can learn about the effects of 
algorithms: (a) querying the algorithm with different inputs, (b) 
comparing outputs (i.e., rankings) of different algorithms on the same 
inputs (i.e., content inventory), and (c) conducting randomized trials 
assigning users (whether content producers or consumers) to different 
algorithms.
    First, and most minimally, we may be able to see how the algorithm 
ranks different items; we can provide a variety of items (perhaps 
systematically varying some of their characteristics) and see the 
output. To some degree, this is commonly done by marketers (whether in 
commercial, public interest, or political campaigns) probing social 
media platforms in attempts to optimize their own reach by trying 
numerous variations on the content and timing of what they post. This 
is also an approach that has been used successfully to identify, e.g., 
disparate error rates in commercial computer vision 
systems.\23\ One challenge here is that one needs 
relevant samples of items--and perhaps the ability to generate 
systematic variations on them--to run through the algorithm. In the 
context of computer vision, there are available corpuses of images and 
creating a new sample of images is possible since they can be provided 
to the algorithm in standard formats. However, in the context of social 
media, external researchers may have little access to a distribution of 
items and many different signals are used in the ranking (i.e., the 
left column of Figure 1), many of which have only some proprietary 
format.
---------------------------------------------------------------------------
    \\ A variation on this approach would also use access to 
the algorithms' code itself. This might seem like a substantial 
advantage, but often the complete algorithm is complex enough--having 
potentially billions of parameters--that this may neither facilitate 
human understanding (even by the engineers building them) or readily 
enable external researchers or auditors to run a variation on the 
algorithm.
---------------------------------------------------------------------------
    Second, we may be able to compare how different algorithms rank the 
same inventory for the same user. Academic researchers have, for 
example, set up Twitter accounts that emulated some archetypal real 
users and compared how news content is displayed in the ``Home'' 
(algorithmically ranked) versus ``Latest Tweets'' (approximately 
chronological) view; one study with eight artificial accounts found 
that the algorithmic ranking resulted in less exposure to external 
links as a whole, including links to news.\24\*
---------------------------------------------------------------------------
    \*\ These studies are often limited--by platform policies and 
enforcement--in the number of artificial accounts they can use to probe 
these algorithms.
---------------------------------------------------------------------------
    While these kinds of comparisons are an important tool, they 
typically do not tell us about ``algorithmic amplification'' writ 
large. These comparisons tell us about how items would be ranked if--
for a moment and for one account only--the algorithm was changed; they 
do not typically tell us about what would happen if the algorithm were 
changed for a longer period of time and for many people or everyone.
    Consider an example. Say you are connected on social media (e.g., 
are friends on Facebook, follow them on Twitter) to a relative with 
political views very different than your own and who posts a lot of 
political content. Due to algorithmic ranking and recommendation you 
might not see a lot of those posts, as the platform accurately predicts 
you will not engage with them nor would you say, if asked, that they 
helped you be informed or feel connected; rather you see a few of those 
posts, but you do see most of their posts about family and fishing. 
This might seem like a clear case of a ``filter bubble'' whereby the 
algorithm is causing you to be exposed to less cross-cutting content 
than you would under a simple chronological ranking. 
However, the truth can be a bit more complex than that. Under 
chronological ranking you might initially see so many undesirable 
political posts from this person that you choose to unfollow or 
unfriend them. This could result in eventually seeing less cross-
cutting content than you would under the algorithmic 
ranking.=
---------------------------------------------------------------------------
    \\ While studies by researchers at Facebook\25\ document a 
reduction in cross-cutting political content from the set available 
from friends to that displayed in the feed, these researchers 
cautioned\26\ that this was not immediately informative about what 
would happen in a counterfactual world with a chronological feed.
    \=\ Perhaps in contrast to conventional wisdom, whether such 
reduced exposure to cross-cutting content is harmful for society is not 
always so clear,26-28 further highlighting that the 
conclusions here are not so obvious. I am far from alone in arguing 
that some common claims associated with the ``filter bubble'' idea lack 
evidence.29-31
---------------------------------------------------------------------------
    More generally we can note that some of the effects of algorithmic 
ranking and recommendation of content occur by causing people to change 
their behavior including the formation, maintenance, and dissolution of 
their social network ties.\32\ It can also affect the timing and 
duration of use of social media. Both of these then determine the 
longer-run consequences for what items they see. This process is 
illustrated in Figure 2.
    Third, platforms can conduct randomized trials comparing different 
ranking algorithms. This is a key tool by which they optimize these 
algorithms with respect to their goals, and it is also used to gain 
further understanding of these complex systems.33-35
    Randomized trials show that ranking choices can indeed matter. In 
the lead up to the 2012 U.S. presidential election, a routine ranking 
experiment at Facebook randomly assigned over 1 million Americans to an 
algorithm that boosted the ranking of ``hard news'' from established 
news outlets; a preliminary analysis concluded that this increased 
political knowledge, altered policy preferences, and increased voter 
turnout compared with the status quo ranking.\36\ Note, however, that 
the baseline here is not chronological ranking; in fact, the novel 
algorithm only moderately differed from the status quo, as it only 
affected a small fraction of items.\37\Sec.  Thus, while 
this study provides evidence that, in a broad sense, ranking matters, 
it says less about how something approximately like the social media 
status quo compares with, e.g., a chronological ranking.
---------------------------------------------------------------------------
    \Sec. \ With respect to some of the challenges described above and 
below, this is actually an advantage of this study, as we might expect 
less dramatic adjustments by users and publishers in response to such a 
change.
---------------------------------------------------------------------------
    Figure 2: A simplified feedback loop in algorithmic ranking in 
social media. The content available to a user (i.e., their inventory) 
depends on what accounts they follow. Even in the case of chronological 
ranking, what they see depends on their behavior (via, e.g., timing of 
use). Users change their behavior when the content they are shown 
changes (e.g., unfollowing other accounts, spending more or less time 
on platform).

[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]

    There are a small number of published studies directly comparing 
algorithmic ranking with simple heuristic ranking by chronology or 
overall popularity. These are largely in settings (e.g., music and 
podcast recommendations) removed from much of the public discussion 
about social media.* However, recently leaked documents from 
Facebook include a preliminary analysis of a test in which some users 
were randomly assigned to chronological ranking.39,40 While 
we lack important details about this test, and a single such A/B test 
is not definitive, this reveals how, in the absence of ranking, metrics 
measuring exposure to likely ``bad'' content (e.g., spam, items people 
propogate and then delete) increased dramatically--as did these users' 
time spent on Facebook and the revenue attributed to 
them. Users in chronological ranking also did not 
seem to like what they were seeing, as they hid many more items. As 
noted by the author of the internal report,\40\ as with the comparisons 
described above, it is possible some of these effects are only short-
run and they might reverse over the longer run as users remove friends, 
leave groups, or lower their expectations for Facebook and visit less.
---------------------------------------------------------------------------
    \*\ For example, Spotify conducted a test of their podcast 
recommendations in which some users received recommendations 
personalized based on their listening, while others received 
recommendations personalized based only on their demographics.\38\ The 
more personalized group listened to somewhat more podcasts, but on 
average listened to a lower variety of categories of podcasts.
    \\ This highlights that the status quo ranking was clearly 
not optimized to simply maximize short-run time spent or revenue.
---------------------------------------------------------------------------
    A recent article reports on a similar, longer-running experiment at 
Twitter, in which some users were randomly assigned to chronological 
ranking.\41\ These authors summarize their results as indicating that 
politicians in general receive ``algorithmic amplification'' in that 
they get more reach among users in the status quo ranking compared with 
chronological ranking; across several countries, right-leaning parties 
are said to receive more amplification. While this long-running 
experiment avoids some of the challenges described above (e.g., these 
users have had time to adjust who they follow), it is still not 
obviously informative about what would happen in the counterfactual 
world where everyone has chronological ranking.*
---------------------------------------------------------------------------
    \*\ The Twitter researchers refer to these challenges briefly, 
citing some of my work\42\ and noting that the experiment does not 
``provide unbiased estimates of causal quantities of interest''.\41\
---------------------------------------------------------------------------
    To see an important remaining challenge here, consider how 
politicians or their social media managers would respond if all Twitter 
users have chronological ranking. Presently, they have tuned their 
Twitter activity to the status quo, but if suddenly everything was 
chronological, they would change their behavior. 
Accounting for this kind of response from users--and especially from 
strategic, professional users of a platform--is key to assessing what 
is truly being amplified by an algorithm. Highly-transparent rankings 
and recommendations can be easier for various actors to 
``game''.= I would not be surprised if key results from 
these experiments might disappear or reverse when accounting for these 
responses.
---------------------------------------------------------------------------
    \\ Sophisticated social media managers would be optimizing 
this quantitatively, but even individual users (including politicians) 
would receive feedback via the amount of likes and retweets their 
different posts get or may view basic aggregate reach statistics 
provided by Twitter.
    \=\ Rankings by recent overall popularity are also often subject to 
strategic behavior as well. For example, we observed groups coordinate 
their Twitter activity to make dozens of political hashtags appear on 
Twitter's trending topics list during the 2019 Indian general 
election.\43\
---------------------------------------------------------------------------
    Platforms clearly regard these feedback loops, spillovers across 
users, and adaptive, strategic behaviors as important, with several 
industry and industry-academic teams working on methods to better 
quantify these effects.35,38,44 Thus, assessing what a 
ranking algorithm is amplifying is not a trivial task that platforms 
have simply neglected. And we may be substantially mislead by 
assessments of algorithmic amplification that simply compare two 
rankings of the same content or even compare consumption by users 
randomized to different algorithms.
Qualitative evidence
    In the absence of evidence from quantitative studies, we might look 
to more qualitative evidence. Commentators frequently point to cases 
where algorithmic recommendation of, e.g., content or groups is an 
antecedent of some harm. But it is unclear how much of these outcomes 
to attribute to specific features of social media, and algorithmic 
ranking in particular. One useful comparison is that clearly 
inflammatory content and misinformation can very much go viral in the 
absence of such ranking, with similarly terrible immediately 
consequences. For example, viral rumors spread on the group messaging 
service WhatsAppSec. --which lacks algorithmic ranking--have 
been the proximal cause of lynching and other violence in India.\46\ 
This hardly proves that algorithmic ranking does not matter, but it 
certainly highlights the difficulty of credibly attributing harms to 
algorithmic ranking based on these kinds of observations alone.
---------------------------------------------------------------------------
    \Sec. \ In our study of images shared in politics-related WhatsApp 
groups in India, images identified as misinformation by journalists 
made up around 13 percent of all image shares.\45\
---------------------------------------------------------------------------
    This absence of general, credible evidence about effects of 
algorithmic ranking and recommendation partially reflects the limited 
ability of external researchers, journalists, and others to probe these 
systems. Thus, we may expect that efforts to promote algorithmic 
transparency and to enable producing reports on algorithmic impact 
using internal, proprietary data could facilitate the creation of 
better evidence. But this absence also reflects some fundamental 
challenges, some of which I have elaborated on here, to learning about 
effects of interventions in these complex environments--even with 
access to all of the proprietary data.
Policy implications
    It is beyond the scope of my testimony and my expertise to make 
comprehensive policy recommendations. Instead I just highlight a few 
implications of the preceding discussion.

   Policy-making should not presume that requiring or promoting 
        the use of largely unfiltered chronological feeds in social 
        media will be beneficial. There could be benefits from 
        resulting transparency (e.g., requiring platforms to offer 
        choice may make it easier for external researchers to learn 
        about the algorithms), and perhaps to individual consumers. But 
        there may be other substantial costs, including substantial 
        increases in spam and misinformation without corresponding 
        direct improvements for consumers.

   Minimal versions of algorithmic transparency--such as 
        requiring disclosure of ranking algorithms--may not be 
        sufficiently informative to yield many benefits. Depending on 
        the values of policy-makers, this might suggest these efforts 
        are not worthwhile, or that a broader and more substantial 
        version of transparency, involving disclosure of substantially 
        more data, is needed.

   Algorithmic ranking and recommendation are ubiquitous, often 
        appearing in many distinct channels on the same platform. 
        Policymakers may want to consider whether a particular 
        regulation should apply to all of these channels, especially if 
        compliance requires substantial effort and user-facing controls 
        for each channel (e.g., a prominent icon) and if some channels 
        are already close substitutes for each other.

   Algorithmic ranking and recommendation are not limited to 
        the decision of whether to display and item and in what order, 
        as items can be aggregated or displayed more or less 
        prominently. Expanding this repertoire of user interfaces can 
        be an important site of innovation. Transparency mandates that 
        neglect this may (a) fail to be informative about important 
        decisions by platforms and/or (b) impede innovation in new ways 
        of presenting information and media.

   There are existing (e.g., TikTok, YouTube) and potential 
        social media platforms where there is no obvious ``default'' 
        ranking. Policies that would enshrine comparisons to a 
        particular baseline when defining ``algorithmic amplification'' 
        may lead to absurd conclusions for many platforms.

   Naive definitions of ``algorithmic amplification'' can be 
        misleading because they do not account for the reactions of 
        consumers and (often quite strategic) producers, which can be 
        dramatic. Alongside other problems, this may complicate efforts 
        to use such definitions to establish moral or legal 
        responsibility.\47\

   External researchers (whether academics, journalists,\48\ or 
        government officials) depend on the ability to probe these 
        systems. Various data-collection methods often opposed by 
        platforms (including scraping public content, participants 
        consenting to automated contribution of their 
        data,49,50 confirming participants have stopped 
        using a service,51,52 and sophisticated audit 
        studies24,53-55) remain central to their work. If 
        policy-makers want rigorous external research--whether on 
        foreign interference in U.S. elections\56\ or on effects of 
        algorithms--to be possible, they can protect the right to 
        employ these methods, rather than, e.g., the Computer Fraud and 
        Abuse Act discouraging such work.\57\

   Some of the best evidence I have drawn on comes from A/B 
        tests (i.e., randomized trials) conducted by platforms. These 
        are regarded as a ``gold standard'' for learning about cause-
        and-effect relationships, and they are a key tool for 
        understanding effects of changes to algorithmic ranking, as 
        well as many other key decisions. Policies and rhetoric 
        discouraging their use could leave scientists and the public 
        less informed--as well as make it harder for the platforms 
        themselves to make good decisions.\58\

   Arrangements whereby platforms share data with external 
        researchers while protecting people's privacy can be beneficial 
        for science and the public. However, there remains substantial 
        uncertainty about what methods for privacy protection satisfy 
        some international privacy regulations,\59\ perhaps 
        discouraging more of such sharing. In some cases, privacy 
        regulations have apparently discouraged platforms from even 
        retaining detailed data that would be useful for assessing 
        algorithmic impact.\56\
About me
    I am a researcher and educator working at the intersection of 
communication technologies and topics of broad societal interest, 
including political participation, public health, and commerce. In 
addition to doing empirical research, I work on methodology for 
learning about cause and effect relationships (i.e., causal inference), 
particularly when people are affected by others' behaviors, as in much 
of social life, but especially in social media.
    I am faculty at the Massachusetts Institute of Technology (MIT), 
where I am currently the Mitsubishi Career Development Professor, an 
associate professor in the Marketing group of the Sloan School of 
Management, and affiliated faculty at the Institute for Data, Systems & 
Society in the Schwarzman College of Computing, including its Center 
for Statistics and Data Science. I teach analytics in our professional 
degree programs and research methodology to PhD students in multiple 
programs. My academic research has been published in peer-reviewed 
journals and proceedings spanning several fields. I studied at Stanford 
University, resulting in five degree including my PhD.
    While currently an academic, I have broad knowledge of the Internet 
industry and state-of-the-art practices there. I was previously a 
scientist and consultant at Facebook, where I worked on feed, 
messaging, advertising, survey methods, and tools for randomized 
experiments. Before that I worked at Nokia and Yahoo, where I likewise 
worked in research, studying and designing social media. I co-organize 
an annual conference that involves substantial participation by data 
scientists and managers from the Internet industry, and I frequently 
have industry experts as guest speakers in my classes at MIT.
Acknowledgements
    I am grateful to comments from several people, including Jennifer 
Allen, Daniel Bjorkegren, Tom Cunningham, Andrey Fradkin, Marta Franco, 
David Rand, and Johan Ugander. Thanks to Alex Kantrowitz for allowing 
me to review a key document\40\ used in his reporting.\39\ 
Responsibility for this document is mine alone.
Disclosures
    Here I note some relevant potentially competing interests. My full 
set of potentially competing interests are listed on my website. I am a 
consultant to Twitter. My research has been funded by Amazon, Boston 
Globe Media, Facebook, IBM, the U.S. Air Force (via a subcontract from 
Lincoln Laboratory), and The World Bank. A conference I co-organize has 
been sponsored by Amazon, Facebook, and Netflix. I conduct some of my 
research via both data use agreements with unnamed firms (e.g., 
retailers, Internet companies, and fitness companies) and by advising 
research by collaborators, often students, who are employees or 
consultants of other firms (e.g., Facebook). I have significant 
financial interests in Amazon, GoFundMe, Google, and Salesforce.
References
    1. Pennycook, G., Epstein, Z., Mosleh, M., Arechar, A. A., Eckles, 
D., & Rand, D. G. (2021). Shifting attention to accuracy can reduce 
misinformation online. Nature, 592(7855), 590- 595.
    2. Bond, R. M., Fariss, C. J., Jones, J. J., Kramer, A. D. I., 
Marlow, C., Settle, J. E., & Fowler, J. H. (2012). A 61-million-person 
experiment in social influence and political mobilization. Nature, 
489(7415), 295-298. doi:10.1038/nature11421
    3. Jones, J. J., Bond, R. M., Bakshy, E., Eckles, D., & Fowler, J. 
H. (2017). Social influence and political mobilization: Further 
evidence from a randomized experiment in the 2012 U.S. Presidential 
Election. PLOS ONE, 12(4), e0173851.
    4. Aral, S. & Nicolaides, C. (2017). Exercise contagion in a global 
social network. Nature Com-munications, 8(1), 1-8.
    5. Holtz, D., Zhao, M., Benzell, S. G., Cao, C. Y., Rahimian, M. 
A., Yang, J., . . . Sowrirajan, T. et al., (2020). Interdependence and 
the cost of uncoordinated responses to COVID-19. Proceedings of the 
National Academy of Sciences, 117 (33), 19837-19843.
    6. Aral, S. (2021). The Hype Machine: How Social Media Disrupts Our 
Elections, Our Economy, and Our Health--and How We Must Adapt. 
Currency.
    7. Bernstein, M. S., Bakshy, E., Burke, M., & Karrer, B. (2013). 
Quantifying the invisible audience in social networks. In Proceedings 
of the SIGCHI conference on human factors in computing systems (pp. 21-
30).
    8. Mosseri, A. (2021). Shedding more light on how Instagram works. 
Instagram. Retrieved from https://about.instagram.com/blog/
announcements/shedding-more-light-on-how-instagram-works
    9. Kohavi, R., Tang, D., & Xu, Y. (2020). Trustworthy online 
controlled experiments: A practical guide to A/B testing. Cambridge 
University Press.
    10. Backstrom, L. (2016). Serving a billion personalized news 
feeds. In Proceedings of the ninth ACM International Conference on Web 
Search and Data Mining (pp. 469-469). Retrieved from https://
www.youtube.com/watch?v=Xpx5RYNTQvg
    11. Lada, A., Wang, M., & Yan, T. (2021). How does news feed 
predict what you want to see? Facebook Newsroom. Retrieved from https:/
/about.fb.com/news/2021/01/how-does-news-feed-predict-what-you-want-to-
see/
    12. Smith, B. (2021). How TikTok reads your mind. Retrieved from 
https://www.nytimes.com/2021/12/05/business/media/tiktok-algorithm.html
    13. Oremus, W. (2016). Who controls your Facebook feed. Slate. 
Retrieved from http://www.slate.com/articles/technology/cover_story/
2016/01/how_facebook_s_news_feed_algorithm_
works.html
    14. Letham, B., Karrer, B., Ottoni, G., & Bakshy, E. (2019). 
Constrained Bayesian optimization with noisy experiments. Bayesian 
Analysis, 14(2), 495-519.
    15. Obeng, A. & Bakshy, E. (2020). Preference learning for real-
world multi-objective decision making. In ICML 2020 Workshop on Real 
World Experiment Design and Active Learning.
    16. Eckles, D., Kizilcec, R. F., & Bakshy, E. (2016). Estimating 
peer effects in networks with peer encouragement designs. Proceedings 
of the National Academy of Sciences, 113(27), 7316-7322.
    17. Ali, M., Sapiezynski, P., Bogen, M., Korolova, A., Mislove, A., 
& Rieke, A. (2019). Discrimination through optimization: How Facebook's 
ad delivery can lead to biased outcomes. Proceedings of the ACM on 
Human-Computer Interaction, 3(CSCW), 1-30.
    18. Wilhelm, M., Ramanathan, A., Bonomo, A., Jain, S., Chi, E. H., 
& Gillenwater, J. (2018). Practical diversified recommendations on 
youtube with determinantal point processes. In Proceedings of the 27th 
ACM International Conference on Information and Knowledge Management 
(pp. 2165-2173).
    19. Zhao, B., Narita, K., Orten, B., & Egan, J. (2018). 
Notification volume control and optimization system at Pinterest. In 
Proceedings of the 24th ACM SIGKDD International Conference on 
Knowledge Discovery & Data Mining (pp. 1012-1020).
    20. Su, J., Kamath, K., Sharma, A., Ugander, J., & Goel, S. (2020). 
An experimental study of structural diversity in social networks. In 
Proceedings of the International AAAI Conference on Web and Social 
Media (Vol. 14, pp. 661-670).
    21. Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. 
(2019). Dissecting racial bias in an algorithm used to manage the 
health of populations. Science, 366(6464), 447-453.
    22. Rahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, 
J.-F., Breazeal, C., . . . Jackson, M. O. et al., (2019). Machine 
behaviour. Nature, 568(7753), 477-486.
    23. Buolamwini, J. & Gebru, T. (2018). Gender shades: 
Intersectional accuracy disparities in commercial gender 
classification. In Proceedings of the 1st Conference on Fairness, 
Accountability and Transparency (pp. 77-91). PMLR.
    24. Bandy, J. & Diakopoulos, N. (2021). Curating quality? How 
Twitter's timeline algorithm treats different types of news. Social 
Media + Society, 7 (3), 20563051211041648.
    25. Bakshy, E., Messing, S., & Adamic, L. A. (2015). Exposure to 
ideologically diverse news and opinion on Facebook. Science, 348(6239), 
1130-1132.
    26. Bakshy, E. & Messing, S. (2015). Exposure to ideologically 
diverse news and opinion, future research. Retrieved from https://
solomonmg.github.io/post/exposure-to-ideologically-diverse-response/
    27. Bail, C. A., Argyle, L. P., Brown, T. W., Bumpus, J. P., Chen, 
H., Hunzaker, M. F., . . . Volfovsky, A. (2018). Exposure to opposing 
views on social media can increase political polarization. Proceedings 
of the National Academy of Sciences, 115(37), 9216-9221.
    28. Levy, R. (2021). Social media, news consumption, and 
polarization: Evidence from a field experiment. American Economic 
Review, 111(3), 831-70.
    29. Guess, A., Nyhan, B., Lyons, B., & Reifler, J. (2018). Avoiding 
the echo chamber about echo chambers. Knight Foundation.
    30. Hosanagar, K. & Miller, A. P. (2020). Who do we blame for the 
filter bubble? On the roles of math, data, and people in algorithmic 
social systems. In K. Werbach (Ed.), After the Digital Tornado: 
Networks, Algorithms, Humanity (pp. 103-121). Cambridge University 
Press.
    31. Bail, C. (2021). Breaking the Social Media Prism. Princeton 
University Press.
    32. Berman, R. & Katona, Z. (2020). Curation algorithms and filter 
bubbles in social networks. Marketing Science, 39(2), 296-316.
    33. Bakshy, E., Eckles, D., & Bernstein, M. S. (2014). Designing 
and deploying online field experiments. In Proceedings of the 23rd 
international conference on World Wide Web (pp. 283-292).
    34. Peysakhovich, A. & Eckles, D. (2018). Learning causal effects 
from many randomized experiments using regularized instrumental 
variables. In Proceedings of the 2018 World Wide Web Conference (pp. 
699-707).
    35. Gupta, S., Kohavi, R., Tang, D., Xu, Y., Andersen, R., Bakshy, 
E., . . . Coey, D. et al., (2019). Top challenges from the first 
practical online controlled experiments summit. ACM SIGKDD Explorations 
Newsletter, 21(1), 20-35.
    36. Messing, S. (2013). Friends that matter: How social 
transmission of elite discourse shapes political knowledge, attitudes, 
and behavior. Doctoral dissertation, Chapter 7. Stanford University.
    37. Allen, J., Howland, B., Mobius, M., Rothschild, D., & Watts, D. 
J. (2020). Evaluating the fake news problem at the scale of the 
information ecosystem. Science Advances, 6(14), eaay3539.
    38. Holtz, D., Lobel, R., Liskovich, I., & Aral, S. (2020). 
Reducing interference bias in online marketplace pricing experiments. 
arXiv preprint arXiv:2004.12489.
    39. Kantrowitz, A. (2021). Facebook removed the News Feed algorithm 
in an experiment. Then it gave up. Big Technology. Retrieved from 
https://bigtechnology.substack.com/p/facebook-removed-the-news-feed-
algorithm
    40. Anonymous. (2018). What happens if we delete ranked News Feed? 
Facebook. Part of ``The Facebook Files''.
    41. Huszar, F., Ktena, S. I., O'Brien, C., Belli, L., Schlaikjer, 
A., & Hardt, M. (2021). Algorithmic amplification of politics on 
Twitter. arXiv preprint arXiv:2110.11010.
    42. Eckles, D., Karrer, B., & Ugander, J. (2017). Design and 
analysis of experiments in networks: Reducing bias from interference. 
Journal of Causal Inference, 5(1).
    43. Jakesch, M., Garimella, K., Eckles, D., & Naaman, M. (2021). 
Trend alert: A cross-platform organization manipulated Twitter trends 
in the Indian general election. Proceedings of the ACM on Human-
Computer Interaction, 5(CSCW2). doi:10.1145/3479523
    44. Karrer, B., Shi, L., Bhole, M., Goldman, M., Palmer, T., 
Gelman, C., . . . Sun, F. (2021). Network experimentation at scale. In 
Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & 
Data Mining (pp. 3106-3116).
    45. Garimella, K. & Eckles, D. (2020). Images and misinformation in 
political groups: Evidence from Whatsapp in India. Harvard Kennedy 
School Misinformation Review.
    46. Dwoskin, E. & Gowen, A. (2018). On WhatsApp, fake news is 
fast--and can be fatal. Washington Post. Retrieved from https://
www.washingtonpost.com/business/economy/on-whats
app-fake-news-is-fast-and-can-be-fatal/2018/07/23/a2dd7112-8ebf-11e8-
bcd5-9d911c784c38_
story.html
    47. Keller, D. (2021). Amplification and its discontents: Why 
regulating the reach of online content is hard. Journal of Free Speech 
Law, 1, 227-268.
    48. Matt, F., Stern, J., Barry, R., West, J., & Wells, G. (2021). 
How TikTok's algorithm figures out your deepest desires. Wall Street 
Journal. Retrieved from https://www.wsj.com/video/series/inside-
tiktoks-highly-secretive-algorithm/investigation-how-tiktok-algorithm-
figures-out-your
-deepest-desires/6C0C2040-FF25-4827-8528-2BD6612E3796
    49. NYU researchers were studying disinformation on Facebook. the 
company cut them off. (2021). NPR. Retrieved from https://www.npr.org/
2021/08/04/1024791053/facebook-boots-nyu-disinformation-researchers-
off-its-platform-and-critics-cry-f
    50. Brinberg, M., Ram, N., Yang, X., Cho, M.-J., Sundar, S. S., 
Robinson, T. N., & Reeves, B. (2021). The idiosyncrasies of everyday 
digital lives: Using the human screenome project to study user behavior 
on smartphones. Computers in Human Behavior, 114, 106570.
    51. Brynjolfsson, E., Collis, A., & Eggers, F. (2019). Using 
massive online choice experiments to measure changes in well-being. 
Proceedings of the National Academy of Sciences, 116(15), 7250-7255.
    52. Allcott, H., Braghieri, L., Eichmeyer, S., & Gentzkow, M. 
(2020). The welfare effects of social media. American Economic Review, 
110(3), 629-76.
    53. Edelman, B., Luca, M., & Svirsky, D. (2017). Racial 
discrimination in the sharing economy: Evidence from a field 
experiment. American Economic Journal: Applied Economics, 9(2), 1-22.
    54. Metaxa, D., Park, J. S., Robertson, R. E., Karahalios, K., 
Wilson, C., Hancock, J., Sandvig, C. et al., (2021). Auditing 
algorithms: Understanding algorithmic systems from the outside in. 
Foundations and Trends in Human-Computer Interaction, 14(4), 272-344.
    55. Chen, W., Pacheco, D., Yang, K.-C., & Menczer, F. (2021). 
Neutral bots probe political bias on social media. Nature 
Communications, 12(1), 1-10.
    56. Aral, S. & Eckles, D. (2019). Protecting elections from social 
media manipulation. Science, 365(6456), 858-861.
    57. Sandvig v. Sessions. (2018). District Court, District of 
Columbia.
    58. Meyer, M. N. (2015). Two cheers for corporate experimentation: 
The A/B illusion and the virtues of data-driven innovation. Colorado 
Technology Law Journal, 13, 273.
    59. Altman, M., Cohen, A., Nissim, K., & Wood, A. (2021). What a 
hybrid legal-technical analysis teaches us about privacy regulation: 
The case of singling out. Boston University Journal of Science and 
Technology Law, 27, 1.

    Senator Lujan. Thank you, Dr. Eckles. I will recognize 
myself for questions. And starting a discussion as critical and 
complex as this one, it is important to start from common 
ground. And with that, I would like to start with the point 
where I believe there is broad agreement. Personal data used by 
online recommendation and targeting systems has come to include 
everything from personally identifiable information and 
preferences to individual behavior patterns.
    This data draws on the incredible ability for 
recommendation and engagement algorithms to target consumers at 
the optimal time with the optimal content to provoke a 
reaction. Unrestricted access sharing and consolidation of data 
allows users to be targeted in ways they do not agreed to. Many 
times they are not even aware that it is happening.
    As I said in my opening statement, giving consumers control 
over their own data is a shared priority. The vast majority of 
Americans believe Congress should act on comprehensive privacy 
legislation. Earlier this year, Morning Consult found that 86 
percent of Democrats and 81 percent of Republicans want privacy 
to be an important priority for Congress.
    I know Chair Cantwell recognizes this, and I am grateful 
for her leadership on privacy reform and advocacy for increased 
resourcing the Federal Trade Commission so it can effectively 
do its job. Now, Ms. Gonzalez, how is data privacy a critical 
component for protecting consumers from the harms of 
algorithmic amplification?
    Ms. Gonzalez. Thank you for the question, Senator Lujan. 
You know, part of the way that harms happen online is because 
we are being targeted because in ways that companies know that 
we might be persuaded because of the data they are tracking. So 
take, for instance, targeted fraud against elders. This is not 
a new problem. It is a problem we have dealt with as a society 
for a long time.
    But online, a person, an older adult may be targeted based 
on their perceived fears or interests. And so there is a 
heightened duty, I think, to protect consumers when they are 
being targeted for abuse, fraud, and other types of harm in 
these ways, based on the personal characteristics that they may 
or may not know they are even giving up when they searched the 
Internet or go on an app.
    Senator Lujan. And Ms. Jackson, online platforms take 
measures to ensure their platforms aren't actively promoting 
civil rights violations, terrorist recruitment, and other 
extremism. Unfortunately, too often these systems fail, and 
platforms instead do the opposite and actually contribute to 
extreme harms through their algorithmic recommendations. That 
is why I am proud again to have introduced the Protecting 
Americans from Dangerous Algorithms Act. Online platforms have 
real accountability when their algorithms are at fault for 
making recommendations that contribute to terrorism. So, Ms. 
Jackson, what principles are necessary to increase platform 
accountability when they amplify extremism online?
    Ms. Jackson. Thank you, Senator, and thank you for your 
leadership on this issue. I think what is important and I would 
commend the companion bill that you recently introduced for 
taking a similar approach is to focus on not content itself, 
but on the mechanisms of the platforms, and in particular how 
they spread across an ecosystem platform to platform.
    And so the more that we are looking at questions of the 
data itself and the incentives behind it, the algorithms that 
may interact with other incentives on a platform, the business 
incentives that are a completely different aspect of a 
platform, the better shape that we are in. And I think that 
gets back to all of the focus that I think each of the 
panelists have had on issues of transparency, privacy, 
protection, and having a better understanding of our ecosystem 
at large.
    [Technical problems.]
    Senator Lujan.--algorithmic ranking is complex and not just 
aimed at maximizing short run engagement. Oversight of the 
harms of algorithmic ranking is a challenge because these 
systems are complex, and users often respond strategically to 
try to game these systems. Three, we are still gathering 
evidence on the harms and benefits of algorithmic 
amplification. And four, Congress can act to protect external 
researchers and enable independent oversight and auditing of 
these systems in privacy preserving ways.
    Now, I agree that this is a challenge, but we have to rise 
to this moment. What can platforms do to reduce the 
effectiveness of these cross-platform manipulation campaigns?
    Ms. Gonzalez. Senator, I am not sure if that question was 
for me because the audio cut out on my end. Please excuse me.
    Senator Lujan. That was for Dr. Eckles. Thank you.
    Mr. Eckles. So in some of my own work, we have been able to 
track some of these cross-platform campaigns because of 
collecting data on multiple platforms, on WhatsApp and on 
Twitter. Individual platforms often have blind spots there, and 
if they start to coordinate more, that can raise questions of 
anti-competitive behavior or them acting as content cartels, as 
some legal scholars have suggested.
    And so I think getting clear guidance from regulators, 
whether from Congress or from the FTC, about what sorts of 
coordination between platforms and data sharing they want to 
encourage could be helpful. It could allow more of that 
coordination across platforms. Though in some ways I found it 
striking just how easy some of these algorithms seem to be to 
game. And those are some of the ones that are most transparent. 
What we observed in our research was the trending topics on 
Twitter being gamed, which is pretty much just a popularity 
ranking.
    And so these groups were able to coordinate looking at the 
current ranking and work to move their topics up the ranking, 
in part because everything was so simple and so transparent. 
And so in a lot of ways, I would like to see the platforms both 
exchanging more information, but also making some of these 
algorithms more complex in some ways to account for what they 
detect when there is this coordination happening. Sometimes 
this coordination doesn't rise to the level of manipulation, it 
is in a gray area between genuine political participation and 
astroturfing.
    Senator Lujan. Thank you, Dr. Eckles. I will now recognize 
the Ranking Member, Mr. Thune. Both, Mr. Thune, if you want to 
take your 5 minutes for an opening statement and then also for 
questions as well. In addition to that, sir.

                 STATEMENT OF HON. JOHN THUNE, 
                 U.S. SENATOR FROM SOUTH DAKOTA

    Senator Thune. Thank you, Mr. Chairman. And I won't take 
that for five minutes and I appreciate your indulgence. Some of 
us were over at the Capitol at a rotunda funeral there. I do 
want to ask that my full statement be included as a matter of 
the record and say how much I appreciate you convening today's 
hearing to continue the conversation on how these massively 
wealthy and powerful big tech platforms are using persuasive 
technology and how they are profoundly altering our future. 
There is also a growing bipartisan consensus that we need to 
shed greater light on the secretive content that the moderation 
processes that social media companies use.
    And it is time, in my view, to make big tech more 
transparent, more accountable, and I look forward to working 
with you and our colleagues on this committee to get 
legislation like the Filter Bubble Transparency Act and the 
PACT Act across the finish line.
    So I want to thank each of the witnesses for their 
participation in this important hearing, particularly as the 
Committee works to enact meaningful legislation to hold big 
tech accountable. And I again will submit my entire statement 
for the record.
    [The prepared statement of Senator Thune follows:]

 Prepared Statement of Hon. John Thune, U.S. Senator from South Dakota
    Thank you, Chairman Lujan.
    In 2019, six months before the onset of the COVID-19 pandemic, I 
served as chairman of this subcommittee, and convened a hearing on the 
use of algorithms and persuasive technology on Internet platforms.
    During that hearing, I noted that former Google Executive Chairman, 
Eric Schmidt, once said modern technology platforms ``are even more 
powerful than most people realize, and our future will be profoundly 
altered by their adoption and successfulness in societies everywhere.''
    Now we are near the end of 2021, and it seems like 2019--only two 
years ago--was a different age.
    Since that time, the pandemic has accelerated the use of technology 
platforms, and algorithms have become even more ubiquitous in every 
aspect of the technology we use each day, largely without us even 
realizing it.
    Indeed, while the pandemic has damaged many sectors of the economy, 
2021 has been a year for Big Tech's record books, even more so than was 
2020.
    Already in June of this year, the Financial Times observed that 
Apple, Microsoft, Google, Amazon, and Facebook were adding a combined 
$51 billion dollars of equity value a week.
    The current stock market value of these five companies, at over $9 
trillion dollars, is more than the value of the next 27 most valuable 
companies put together.
    As Eric Schmidt predicted, these technology companies have grown 
exponentially more wealthy and powerful, especially over the course of 
the pandemic.
    So Mr. Chairman, I appreciate you convening today's hearing to 
continue the conversation on how these massively wealthy and powerful 
big tech platforms are using persuasive technology, and how they are 
profoundly altering our future.
    There is no question that Internet platforms, and especially social 
media platforms, have transformed the way we communicate and interact 
with friends, family, and the public.
    So much of the content we see on social media is entertaining, 
educational, and can be beneficial to the public.
    However, I, along with several of my colleagues, and the public, 
have become more and more concerned about the influence and power of 
big tech within our society and the potentially damaging effect these 
platforms can have on individuals.
    We've heard testimony earlier this year about social media's 
damaging effects on consumers.
    Hearing after hearing demonstrated that users are not aware of how 
their data is being used by big tech to affect their behavior and 
influence certain outcomes.
    We know that a major problem with social media platforms, as well 
as search engines, is their use of secret algorithms--artificial 
intelligence developed by Silicon Valley software engineers that's 
designed to shape and manipulate users' experiences.
    The powerful AI behind these platforms serve as prediction engines, 
creating a unique universe of information for each user, a phenomenon 
that's often referred to as the ``filter bubble.''
    The filter bubble contributes to political polarization and social 
isolation.
    Perhaps the most important thing to understand is that most users 
don't make a conscious decision to enter the filter bubble.
    This can be particularly troubling for younger users.
    For example, a recent Wall Street Journal investigation described 
in detail how TikTok's algorithm serves up highly inappropriate videos 
to minors.
    As a general matter, the economic incentives are aligned for big 
tech to keep the filter bubble in place, without users' awareness, 
because platforms only make money by keeping eyeballs on their 
platforms as long as possible.
    Without congressional intervention, platforms have very little 
incentive to be more transparent about the existence of the filter 
bubble.
    The days are over when you logged into your favorite social media 
platform and consumed content that had been posted chronologically 
since your previous log-in.
    Now platforms like Facebook, Instagram, and TikTok--and other 
social media platforms, as well as search engines--use algorithms to 
shape your newsfeed and suggest additional, seemingly never-ending 
content, emphasizing posts the platforms think you'll be interested in 
and deemphasizing the ones they want you to scroll past.
    And looking to the future, at the cutting edge of the use of 
algorithms, machine learning, and artificial intelligence on Internet 
platforms, is the idea of the ``embodied internet,'' or ``metaverse,'' 
which Mark Zuckerberg described as ``the next evolution of social 
connection'' when he changed the name of his company to Meta this past 
October.
    The aspect of algorithms amplifying or suppressing content leads to 
increasing dissatisfaction and suspicion about bias being built in to 
the algorithms.
    Big tech platforms are certainly free to deploy algorithms that 
select content based on what will keep each user engaged.
    But the platforms should not be free to keep their users unaware of 
the fact that an algorithm is controlling which content each consumer 
sees on the platform.
    With some rare exceptions, people aren't given a choice to easily 
opt out of a black-box algorithm that secretly selects the content they 
see.
    We're learning more and more about what the problem is, and I have 
offered several proposals aimed at giving the public more transparency 
into these systems, and more control and accountability for consumers.
    I've introduced the bipartisan Filter Bubble Transparency Act, 
which would give consumers the privacy, choice, and transparency that 
has been absent on these platforms for too long.
    Specifically, large-scale Internet platforms would be required to 
notify users that their platform uses secret algorithms to select the 
content they see, what's often described as the ``filter bubble.''
    In addition, users would be given the choice to switch to a 
different version of the platform that is filter bubble-free.
    At the very least, users should have the option to engage on these 
platforms without being manipulated by secret algorithms.
    There's also a growing bipartisan consensus that we need to shed 
greater light on the secretive content moderation processes social 
media companies use.
    That's why I've introduced the bipartisan PACT Act, which, among 
other things, would require Internet platforms to make biannual 
transparency reports outlining material they've removed from their 
sites or chosen to deemphasize available to the public--and not just in 
intentionally complicated, hard-to-understand legalese. Sites would 
also be required to provide consumers with more due process and 
explanation when content is removed or otherwise moderated.
    It's time to make big tech more transparent and accountable, and I 
look forward to working with my colleagues to get these proposals 
across the finish line.
    Today, each of our witnesses has deep expertise regarding 
algorithms and artificial intelligence more broadly, as well as in the 
more narrow context of engagement, prediction, behavior modification, 
and persuasion, and brings valuable perspective about where we are 
today and what we can expect in the future on these matters.
    Your participation in this important hearing is appreciated, 
particularly as this Committee works to enact meaningful legislation to 
hold big tech accountable.
    Thank you, Mr. Chairman.

    Senator Thune. Mr. Eckels, there is a growing concern that 
large technology platforms may one day know every individual 
even better than we know ourselves. In your written remarks, 
you discussed the vast amount of data Internet platforms have 
on users and that these platforms are able to predict certain 
outcomes for every individual based on this data.
    Could you talk further about how Internet platforms use 
algorithms to make predictions about each user and how they 
influence the behavior of every individual, and perhaps how 
that might concern you?
    Mr. Eckles. Thank you, Senator Thune. I would say, while 
some of these tech giants have a lot of data about many of us, 
we are a little ways off from having that kind of a mastery. 
Sometimes it seems like the people most convinced of that 
mastery are the tech critics and marketing snake oil salesman 
like the people at Cambridge Analytica who were suggesting that 
they really could control people and they could predict them so 
well. So I think we have a bit of time in that sense. What I 
have been sometimes concerned about in the past is when this 
ability to predict people's behavior, how they will respond to 
messages, reaches across many domains.
    Maybe I have interacted with one firm where I bought some 
products. Can they also use that same data to know how to 
persuade me in politics? Could they sell that for that purpose? 
Could they use that same data to help me stick with my fitness 
goals, so there could be some benefits there as well? But in a 
lot of cases, when that data jumps from one domain to another, 
I think that is a potentially concerning area. And when you 
have single firms that reach into so many parts of our lives, 
we think they might be able to start doing that kind of cross 
domain prediction.
    So that is one of the aspects that I find concerning is 
that it may be hard for consumers to understand that this data 
about some purchases I made could also be used in conjunction 
with some other data about other people to predict my behavior 
in a totally different setting and to target totally different 
types of messages, maybe political messages, to me.
    So that is one aspect that I find concerning because there 
is this opacity, and it is a little hard for consumers to 
understand the consequences of this one disclosure for this 
very other setting.
    Senator Thune. Mr. Poulos, what is your reaction to Mr. 
Eckles' testimony highlighting the vast amount of data these 
companies have on individual users? And do you believe that 
large tech companies use persuasive technology, meaning the 
technology is designed to change people's attitudes and 
behaviors?
    Mr. Poulos. Thank you, Senator. Short answer is, yes. 
Longer answer is it is inherent in the nature of this 
technology. This is a communications technology and other 
communications technologies, probably all of them in history, 
have been used and were inherently designed to influence people 
and to cause changes in their behavior, printing press, radio, 
television. These things are a part and parcel of, you know, 
whether you want to call it mind control or something more 
benign, this is how it works with human beings.
    We communicate with each other and those who have the most 
control over communications infrastructure and technology wield 
that control to move hearts and minds to sculpt behavior. So I 
think it is inherent to digital technology. I think that we 
need to be aware that what is really going on here is the 
amassing of personal information inside data bases so that 
large groups of people can be conditioned in certain ways to 
accept and prefer certain ways of life, certain ways of 
Government.
    That is just kind of baked into the technology, and that is 
why it is so important that we make sure that ordinary 
Americans have the ability to use technologies to build 
institutions that serve their humanity and protect their 
country and their culture.
    Senator Thune. I want to come back to that, but let me ask 
Mr. Eckels, you mentioned in your testimony that quantifying 
the impacts of algorithm ranking is quite difficult due to the 
complexity of both technical aspects of algorithms and due to 
people's complex responses to changes with algorithms. If this 
is the case, what should policymakers require or expect of 
Internet platforms with respect to algorithm explanation or 
transparency?
    Mr. Eckles. Thank you, Senator. I think the answer is going 
to be different in different settings. So in some settings, 
some of the most important area sensitive settings like 
diagnoses in health care are not nearly as subject to such 
rapid strategic responses. And so while I don't study the 
health care system, I think this is something that the Federal 
and State Governments are actively involved in already, and so 
it would be natural for them to regulate quite closely the 
adoption of algorithms in those settings.
    When it comes to communication technologies and social 
media, it is not clear exactly what type of transparency we 
need. In fact, these algorithms often can be broken up into 
multiple different pieces, as I have detailed in my testimony.
    And that can be one tool for how Congress and other 
policymakers can get some transparency. So I don't know if 
Congress should mandate explanations and transparency, but I 
hope that my testimony provides kind of a blueprint for what 
questions should be asked of these firms. So they should be 
able to tell you and the public how much weight they put on 
predictions of clicks, predictions of reshares and retweeting, 
versus how much weight they put on predictions of survey 
responses about whether people say this is really important to 
me, or how much weight they put on potentially negative signals 
that tell them that people regret their engagement, that maybe 
they spent too long on the platform, or that they wish they 
hadn't reshared that piece of misinformation.
    And so by being able to tell policymakers and the public 
about those weights, I think that provides one opportunity. I 
mean, as I was preparing my written testimony, I was really 
struck with some of the reporting in the New York Times about 
TikTok.
    And it turns out their algorithm matches exactly what I had 
written without really any knowledge of that, except that it 
seems to lack many of those signals. They seem to not be, at 
least according to this reporting, collecting data about 
people's more considered opinions of whether this was a 
worthwhile video to have seen or any signals that I saw this 
and then later, I felt misinformed or misled.
    And so probing that and pushing platforms to actually be 
more sophisticated in their algorithms, I think could be 
beneficial.
    Senator Thune. So a qualitative as opposed to quantitative. 
Just very quickly, and I know my time expired, Mr. Chairman, 
but I--since I didn't use my 5 minute opening statement, I want 
to just ask one question, and Mr. Poulos, coming back to what 
you said about these are communication mediums and, in other 
words, designed to influence people. I want to come back to 
this question. You said that decision should be made about 
these being used in a way that is positive or good for 
humanity.
    And I guess my question is and the thing I think concerns a 
lot of us is who decides what is good for humanity? Who is 
behind the curtain making these decisions? And then I want to 
ask you this question because you discussed the notion that we 
are in the nascent stage of a social credit system with regard 
to the use of AI and algorithms on Internet platforms.
    And I am curious what your thoughts are about what 
approaches Congress should be considering to prevent a social 
credit system. So you can answer that question, but maybe 
answer the bigger question about who decides what is good for 
humanity?
    Mr. Poulos. Sure. That is a very good question, Senator. 
You know, ultimately this is a Government by, for, and of the 
people. And I think we have to keep that in mind as we grapple 
with the question of how to ensure that our form of Government 
is not replaced by a social credit system.
    The reality is that every major power in the world right 
now, every civilization state, as some call them, is racing to 
establish sovereign authority over the digital technology 
within its territory. And because of the way that these super 
powerful digital entities raise ultimate questions about our 
place in the universe, about what it is to be human, why we 
should bother? We seem to have created things that are that are 
bigger and smarter and have superpowers that we lack.
    So, many countries are responding to that kind of feeling 
of disenchantment or fear by turning to their deepest cultural 
resources, which are, in many cases, theological resources. You 
see China with the Great Firewall and, you know, trying to 
focus on making algorithms and programs Daoist. Russia's 
response to the digital revolution was to build a very large 
military cathedral just outside of Russia. And in countries in 
the West, like France and others, they are actively looking for 
ways to make sure that digital technology really captures and 
expresses and even enforces their particular cultural leanings. 
You see India pursuing some of these things and Israel and 
others.
    In the U.S., we have a First Amendment, and we have a 
prohibition on the establishment of religion. So we really have 
a special challenge because we can't sort of take that 
theological, theocratic option and say, ``look, the Government 
is going to decide, you know, what the proper system of ethics 
is, what counts as harm, you know, whether something like 
transhumanism is something that, people shouldn't be exposed to 
or whether it is something that actually exposes them to, you 
know, what they need to know in order to flourish.'' The 
Federal Government needs to make sure that it doesn't 
nationalize social media in an effort to turn it into a 
tutelary system that programs Americans the way that they want 
Americans to be programmed.
    So, you know, I think the best option is to make sure that 
Americans can buy and use digital technologies, high powered 
GPUs, bitcoin so that they can freely build new institutions, 
cultural and market institutions, in a digital environment 
where they can freely protect their human identity and make 
sure that we preserve our form of Government and argue openly 
about fundamental political questions.
    Senator Thune. Thank you. Mr. Chairman, thank you.
    Senator Lujan. Senator Klobuchar.

               STATEMENT OF HON. AMY KLOBUCHAR, 
                  U.S. SENATOR FROM MINNESOTA

    Senator Klobuchar. Thank you very much and thank you for 
doing this hearing. Really important, Mr. Chair. Thank you to 
our witnesses. Algorithms touch every aspect of our lives. We 
know that. And it is not typically very pretty. And what we 
know is that Americans are getting increasingly addicted to 
platforms, especially young people taking up significant parts 
of their days. And part of that is because of how they are 
targeted.
    Yesterday, we had a hearing with the head of Instagram, and 
I pressed him about the fact that the New York Times reported 
that his major increase in the marketing budget, $67 million in 
2018 to $390 million this year was focused on wooing teens, 
mostly allocated to that. And as some of our colleagues have 
discussed, we need Federal privacy legislation, but we also 
need more transparency on algorithms, and we need to do 
something about this.
    Senator Coons and Portman and I introduced a bill today to 
require that the companies make--allow access to outside 
researchers. Some of this has been shut down, as you know, by 
Facebook, now Meta. But I think it is important for my 
colleagues to know that there is more going on here than meets 
the eye. Amazon is creating knockoff products made by small 
businesses and then using its algorithm to show its own version 
of the product they ripped off first. That has happened, Wall 
Street Journal reporting. Or number two, Apple is doing this 
when you search for maps or music or podcasts, the first result 
is often an app that Apple makes, even though they control the 
platform. And algorithms provide--pervade our entire existence 
online.
    And one of the most shocking things that I think came out 
of the whistleblower documents was the fact that, it didn't 
surprise me, given my own experience online, maybe when I put 
something out there that Senator Thune and I worked on a bill 
together, it might not be as an attractive to the algorithms of 
Facebook as some other forms of speech, no matter how 
interesting I try to make it sound.
    And then I find out that in fact, the angry emoji gets five 
times attraction as a simple like. So if our constituents in 
Minnesota and South Dakota want to put a simple like, they may 
never see that post on Facebook because it is the angry ones 
with the angry emojis that are going to start getting traction.
    So Ms. Jackson, in your testimony, you talked about how 
these companies designed their algorithms to promote 
engagement. And it is, of course, to boost advertising. I think 
we have to think about that. They want to get it out there so 
people get addicted, so they get more advertising revenue. Can 
you tell us more about how companies designed algorithms to 
increase the time users spend on their platform and how that 
affects users?
    Ms. Jackson. Thank you, Senator, and I am looking forward 
to reading the bill in detail that you just mentioned 
referencing--introducing today. It is an important step in 
ensuring that we do have more visibility into how these 
platforms operate in our ecosystem at large. I think to your 
question, I want to call back real quick to something we have 
been talking about in the context of these algorithms. I think 
over indexing or emphasizing the ability of tech companies to 
know us and the kind of perfection, the idea of perfection of 
these algorithms, can cause harm as well.
    Because as you alluded to, we are not just talking about 
algorithms on social media, we are talking about algorithms 
used, whether it is to help Amazon promote its own products, or 
frankly, to help insurance companies make decisions about what 
your premiums should be, what your sentencing might be in a 
criminal justice context. Those are all very different 
contexts. But if we believe that those algorithms are in fact 
infallible, that they have the most accurate representation, we 
might be tempted to allow those algorithms to be integrated 
into our decisionmaking, Government decisionmaking, determining 
the rights that we might have.
    And I would call back to examples of, you know, maybe I 
joined a member program at my local bar, and it says I have 
been buying more drinks and I lost my Apple Watch and so says 
that I haven't been exercising too much, and so my insurance 
company has decided that I am a depressed alcoholic, and I am 
then denied for a loan.
    And that may sound fantastical, but right now there is 
nothing that would make that illegal. And so I think coming 
back to putting these in context, algorithms themselves are not 
a magical solution in one direction or other. They are not a 
magical harm. But it is important that we have accountability, 
that we have limits on data collection that power them, and 
awareness of the ways in which they are used in our lives.
    Senator Klobuchar. OK. So one other piece of this, the 
other bill that I should also mention, that has kind of, I 
think one commentator called it the Ocean's Eleven of co-
sponsors, that I have with Senator Grassley and Graham and 
Warner and Booker, and including two on this committee, Senator 
Blumenthal and Senator Lummis, and what that bill is about is 
about getting after this exclusivity, the self-preferencing, 
the discriminatory behavior that just no one that looks at it 
can really think it is okay. It is just that our laws haven't 
caught up to a major, major part of our economy.
    And my view has always been that competition is good for 
innovation. We have heard a lot of claims that somehow doing 
something is going to undermine the tech companies, but in 
fact, they are doing just fine. And what we want to make sure 
is we continue to foster competition. Could you talk about how 
this innovation can ultimately by monopolies and dominant 
platforms get stunted? And do you share concerns on this--along 
these lines? You started up--you started a startup, right?
    Ms. Jackson. Yes. I don't actually talk about that a lot. I 
founded a small tech company to try and help Americans become 
more civically engaged. And you know, I was--I didn't really 
have a choice, honestly, in using Facebook and Google 
advertising tools. It is kind of--in the startup community, 
just it is the only option. We kind of joke about it as the 
startup tax, that a large portion of your budget, no matter 
what you do, is going to go toward cloud providers of Amazon or 
Google and advertising and Facebook and Google, regardless of 
whether you want to----
    Senator Klobuchar. They are each dominant on their own 
sector is what it is.
    Ms. Jackson. Absolutely.
    Senator Klobuchar. OK. And so----
    Ms. Jackson. Sorry. Yes. I mean, I think there are serious 
consequences for what that means for competition at home, but 
it also means that the ability to feed, everyone has to feed 
into a system that is collecting more and more data that is 
driving these incentives that we are talking about.
    And so no matter what angle, we want to address these 
online harms, whether we are looking at competition issues or 
privacy issues or content issues and extremism, they are all 
interconnected because it is a system that feeds itself. And so 
again, I applaud members of the Committee focusing on privacy 
and transparency.
    Senator Klobuchar. And would you agree that at some point 
we have to stop talking and have a little less talk and a lot 
more action?
    Ms. Jackson. I am a big fan of action, Senator.
    Senator Klobuchar. OK, thank you.
    Senator Lujan. Senator Fischer.

                STATEMENT OF HON. DEB FISCHER, 
                   U.S. SENATOR FROM NEBRASKA

    Senator Fischer. Thank you, Mr. Chairman, and thank you to 
all our witnesses for being here today for this very timely 
discussion. Many of you described in detail how Internet 
platforms increasingly quantify and tailor our online 
experiences in granular ways. On one hand, you get more 
personalized user experiences and platforms are more responsive 
to what makes you tick. But this also allows companies to take 
persuasion too far. Companies are constantly competing for our 
attention. This increases the motivation for more intrusive 
technologies, including the web design around those 
technologies. It forces us to assess the impact of online 
persuasion on user choice and privacy.
    Yesterday, I was proud to reintroduce the DETOUR Act with 
my colleagues, Senators Warner, Thune, and Klobuchar. The Act 
would prohibit large online platforms from purposely using 
deceptive user interfaces, also known as dark patterns.
    Many of us can recognize the dark patterns we have 
personally experienced where we are trapped into clicking, 
clicking, clicking, clicking and trapped into buying or signing 
up for something that you don't really want because the user 
interface intentionally confuses the options that are 
presented. Our legislation would take an important step, I 
believe, to restore the hidden options and improve user 
autonomy.
    Dr. Eckles, I appreciated your technical perspective about 
how persuasive technologies are developed as we seek to craft 
related policies. Do you think that we have enough debate and 
guidance on the ethics in designing technology that influences 
our behavior within the tech industry or in academia?
    Mr. Eckles. Thank you, Senator. I would like to see more 
discussion of these ethical issues in the tech sector and in 
the public as well, and especially not just from ex-tech 
insiders who now are repenting but really drawing on kind of 
the rich expertise that we have from ethicists and policymakers 
in the way that bioethicists have provided some guidance in the 
context of thinking about clinical trials, et cetera.
    So I do think we want more guidance there. One of the 
things that I think of with the Detour Act, at least in the 
last version I was familiar with was, that from my perspective, 
it is very important to allow and encourage these platforms to 
continue to do A/B tests, continue to do randomized trials. 
That is much of the evidence that I have drawn on whether 
published work by those platforms or leaked is A/B test that 
they have conducted.
    And so I would hope that while we can shut down some of 
these dark patterns maybe that has to do with subscription 
services, etcetera, etcetera, that we also still allow these 
platforms to keep making systematic comparisons of different 
options and hopefully sharing more of those results with the 
public.
    Senator Fischer. Do you think that the designers and 
developers right now are approaching taking into any kind of 
consideration the ethical consequences here?
    Mr. Eckles. Thank you, Senator. I think at firms that are 
established and have found themselves with a bit of a breathing 
room and a large staff with technical expertise, they are. I 
mean, in some ways that is part of what we have seen in the 
Facebook files leaks is a huge amount of disagreement and 
debate about some of these issues. But a lot of times that 
happens only after some engineer quickly built something, it 
got big, and then now there is a whole bunch of data scientists 
and researchers cleaning up the mess.
    Senator Fischer. So how do you get the ethical 
considerations at the front end of development? How do you 
incentivize or encourage, I would rather go that route, to have 
designers really, really work, not just with Government, but 
with business and academia to be able to get that done? How do 
you do it?
    Mr. Eckles. Thanks, Senator. I think one way is to make the 
public and all these potential designers really aware of the 
long-term negative consequences of some of these choices, 
right. Facebook's reputation is still deeply hurting from some 
of the move fast and break things decisions that they made 
years and years ago. So that is a consequence that they are 
living with that brand. And so thinking about that as an 
entrepreneur is important. From my perspective, I also think 
about that, in the context of education, that we have 
burgeoning enrollment in computer science programs.
    Ethics and these considerations and social sciences have to 
be part of that curriculum in some way. And so I know in some 
of my own teaching and analytics I try to bring that in. But it 
is also useful to draw on additional expertise. We are trying 
to hire a philosopher of technology ethics at the moment at MIT 
in order to help this be part of the curriculum.
    Senator Fischer. I think it is important that we bridge the 
technical consideration with the policies that we seek to 
implement, and I appreciated your points about the value of A/B 
testing, which remains important if we are going to have 
improved user experiences and distinct from the dark patterns 
that we are trying to address. So I thank you for that, sir. 
Thank you, Mr. Chair.
    Senator Lujan. Thank you, Senator Fischer. Senator Schatz.

                STATEMENT OF HON. BRIAN SCHATZ, 
                    U.S. SENATOR FROM HAWAII

    Senator Schatz. Thank you, Mr. Chairman, Ranking Member, 
thank you, and to all of the testifiers for your thoughtful 
remarks. I just want to respond a little bit to Senator 
Fischer's question because Dr. Eckles, with all due respect, I 
don't think that this is going to end up being effective if it 
is voluntary because you can't have individual coders to sort 
of waking up in the morning and saying, what are the ethical 
implications of this? Because they are engineers, they are 
supervised, they are told to deliver a product, they deliver 
that product.
    The only way to provide an effective counterweight to the 
fiduciary obligation to maximize clicks, which is essentially 
as Senator Klobuchar said, the maximization of revenue is to 
establish in Federal law an affirmative obligation not to harm 
the consumers whose data is being collected. I am all for 
notice and consent, but we all know that we click, I agree, as 
a matter of course without reading anything, and it is not fair 
to individual consumers to ask them to opt out of the world, 
essentially, in order to retain their privacy rights.
    So the only, in my view, the only way to do this is to 
establish, you may collect data from individuals as you are, 
but you have an affirmative obligation to not use that data 
against people, which doesn't mean you can't do behavioral 
advertising, it just means you can't harm people. Most 
companies are comfortable with that.
    The ones that are not comfortable with that either have 
Government relations and lawyers who are just nervous about 
what that would mean, or they are actually worried that they 
are in the business of harming people. So just wanted to offer 
that for Senator Fischer. I have not yet had a Republican co-
sponsor of our legislation, but I would invite you to take a 
look at it. For Mr. Gonzalez, you know, Chair Khan is known as 
a sort of antitrust crusader, and I know that she is moving on 
FTC action against some of these dominant Internet companies 
from a competition policy standpoint.
    But I am wondering about the Section 5 authority to find 
unfair and deceptive practices, and even the FTC's authority 
under the Equal Credit Opportunity Act and the Fair Credit 
Reporting Act. Does the FTC have authorities that it is not yet 
using? In other words, have we already delegated some of this 
policymaking to the FTC and should it exercise additional 
authorities?
    Ms. Gonzalez. Yes, Senator, I do think it already has some 
authority to act here and it should exercise that. I think the 
work that Chair Khan is doing on the antitrust front is 
incredibly important.
    I also think an inquiry about how companies are violating 
our privacy and using our data to harm us is incredibly 
important and should move forward under the Unfairness 
Doctrine, which the FTC already has authority to move forward 
on. The other two statutes that you referred to on credit 
reporting, certainly, I think the FTC should enforce that 
statute.
    Those statutes may not be broad enough to get at the harms 
that we are seeing online where our personal data is extracted, 
used to target us in harmful ways. And so that is why I believe 
the FTC ought to initiate a proceeding focused on privacy and 
civil rights as well.
    Senator Schatz. Yes, just in response to the last comment 
you made about, you know, it may not be--we may not have a 
broad enough statute, point taken. But I will also observe 
that, you know, been on the Commerce Committee for now a number 
of years, and our complaints about tech and big tech are so 
broad and so numerous as to overwhelm our policymaking ability.
    Besides the filibuster and our normal partisan 
disagreements, we sort of can't separate out what are we just 
irritated by, what do we think is real harm, how do we balance 
the needs of American users of technology with the kind of fear 
that we could set a precedent for, say, Kyrgyzstan to impose 
authoritarianism through their own controls. So I think that 
the important thing for us as a committee and as a Congress is 
to chip away at this, is to say, well, what Section 230 reforms 
makes sense, right, and what antitrust actions makes sense, and 
what protections for researchers make sense?
    Because I do think that, especially during the last 
Administration, this sort of invocation was almost an 
incantation against big tech became so expansive as to become 
meaningless and very difficult for policymakers to sort out and 
even more difficult for politicians to sort out. Because if you 
say, well, it is not so simple, you look like you are 
schlepping for these big tech companies when in fact, this 
stuff is complicated, and we have seen sort of comprehensive 
legislation fail and fail and fail.
    And I subscribe to my predecessor, Danny Inouye's point of 
view, which was his 40 percent rule, which is try to get 40 
percent of whatever you are asking for, and by the third year, 
you have already got more than you have ever asked for. I 
always chop that in half and say I got 20 percent because I am 
certainly not Danny Inouye. Thank you.
    Senator Lujan. Thanks, Senator. Senator Blackburn.

              STATEMENT OF HON. MARSHA BLACKBURN, 
                  U.S. SENATOR FROM TENNESSEE

    Senator Blackburn. Oh, Mr. Chairman, thank you so much for 
this hearing. And as our witnesses have heard, this is a very 
timely hearing. We have had quite a focus on technology issues. 
And I know Senator Klobuchar mentioned to you the hearing that 
we had yesterday where we had the Instagram CEO before us. And 
Senator Blumenthal and I are continuing to work on these issues 
around privacy and data security, making certain that the 
virtual space is going to be safe for all users and especially 
for our children.
    Mr. Eckles, let me come to you, and thank you for your 
well-written testimony. I want to pull from that. You mentioned 
in your testimony, and I am quoting you, ``there is no large 
body of evidence on the effects of the algorithmic ranking.'' 
You also noted that ``for a single user, there are often are 
more possible rankings of available items than there are atoms 
in the universe. So even though the same data may be entered in 
the algorithm, a huge variation in outcomes could occur.'' 
Alright, so here is my question.
    Given all this, is it even possible to attain algorithmic 
transparency and get to where we are trying to go with the 
effects of these algorithms? Or would the data outputs be too 
large to even begin to make heads or tails of them?
    Mr. Eckles. Thank you, Senator. I think there are hard 
problems here, but I don't want to make it sound like all 
versions of algorithmic transparency are impossible. And I 
think breaking it up into smaller problems can help. Actually, 
even for data scientists and ranking engineers of these 
companies, they have to do a lot of work to understand their 
own algorithms in a lot of cases.
    And that is in part because they are doing things like 
looking at the contents of a photo, figuring out what is in the 
photo, and knowing whether you might actually be interested in 
that content, right. So that becomes a very complex process. 
But luckily, some of that can be broken down into kind of more 
comprehensible pieces, because at the end of the day, there is 
a prediction about whether you are going to click on that 
content, whether if we asked you in a survey, you would say it 
was important.
    And those are each given weights in these algorithms. Those 
weights, if disclosed, can give you a good picture of what is 
actually being valued. The New York Times reporting on TikTok 
wrote out the equation, supposedly, for TikTok's weights on 
those predictions. So that kind of level of transparency could 
take us somewhere that isn't necessarily the entire algorithm, 
which in fact has billions of parameters. There are billions of 
numbers used to determine these algorithms.
    Senator Blackburn. OK, let me interject right there. Let's 
talk about TikTok. We know the TikTok likes to hook young 
users, and we know that TikTok is collecting this information. 
And basically, what they are doing with their Chinese Communist 
Party shared ownership, I guess you would say, is they are 
basically creating dossiers on Americans. And then they have 
the ability to funnel that information back to the Chinese 
Government. Should policymakers give more thought to the way 
that algorithms and the way that companies collect the data, 
the way they hold this data, the length of time they hold that 
data, and then look at that as a national security risk?
    Mr. Eckles. Thank you, Senator. I think as Americans, we 
have had the luxury of having so many of the tech giants here 
in the U.S., and the result of American innovation. And so as 
that changes, as we depart from that that raises a bunch of 
questions and a lot of ways other countries have been grappling 
with for a while.
    And so my sense is that there certainly could be additional 
scrutiny there when data is moving overseas to authoritarian 
regimes that perhaps don't have the kinds of privacy 
protections that we at least like to imagine for Americans 
here.
    But policymakers maybe have to think about this from a 
foreign relations perspective as well, because a lot of these 
other countries are imposing regulations that require data from 
American companies to be there locally, require employees who 
are locally in that country to potentially be on the hook 
criminally if those companies don't comply with sometimes 
really un-American regulations locally. So I don't know how to 
navigate that, but it sounds like an important project for all 
of you.
    Senator Blackburn. OK, Mr. Poulos.
    Mr. Poulos. Hi, Senator.
    Senator Blackburn. Yes. Mr. Poulos, do you have any 
comment?
    Mr. Poulos. I think what I would say----
    Senator Blackburn. Mr. Poulos, I am sorry. Yes.
    Mr. Poulos. Sure, yes. I mean, I think what I would say is 
that algorithms can't be neutral, and we shouldn't expect that 
we can engineer them to be neutral. It is in the nature of them 
to move behavior and opinion and that what the social media 
companies have done that I think in a sense is most productive 
of harm is that they have convince people that what happens on 
screen is more important than what happens off.
    And the reality in a digital age is that what happens in 
the database is more important than what happens outside the 
data base. And so when you have companies that are convincing 
users that their life online is more important than their life 
off, there is a disconnect there between that impression and 
the reality that those users are actually being onboarded into 
a really comprehensive data system where they have no agency as 
citizens. And that disconnect is one that is very difficult to 
legislate away.
    So I think the important thing to do is, you know, yes, we 
can nibble away at and ameliorate certain kinds of algorithmic 
harm. But really, what we want to do is make sure that 
Americans can recognize that they can build new institutions 
that preserve their way of life and preserve their agency as 
citizens and as members of a community through digital 
technology, that really social media isn't the extent of 
digital technology.
    And I think what we are seeing with Facebook trying to 
change into Meta and other similar moves, that social media 
companies are aware to a degree that perhaps their form of 
technology is just technologically limited. And so this is a 
fast shifting space. And while, you know, as one of the other 
Senators indicated, it is easy to point out what is 
frustrating, irritating about the space. Ultimately, you know, 
trying to make social media a creature of the Federal 
Government is not going to get us where we need to go.
    Senator Blackburn. Thank you. Mr. Chairman, thank you for 
the time.
    Senator Lujan. Thank you very much. Senator Markey.

               STATEMENT OF HON. EDWARD MARKEY, 
                U.S. SENATOR FROM MASSACHUSETTS

    Senator Markey. Thank you, Mr. Chairman. At the same time, 
social media companies can't escape the same rules that 
everyone else has to play by. So, you know, I think many in the 
industry think that just because they have to play by the same 
rules, means they are a creature of the Government. No, they 
are not a creature of the Government, they are a creature of 
the American people. That is who they are supposed to be 
serving, the American people, and they can't escape those 
responsibilities.
    And so I thank you, Ms. Gonzalez, for mentioning my bill 
because what my bill says is that there has to be a prohibition 
on algorithmic processes on online platforms that discriminate 
on the basis of race, age, gender, ability, and other protected 
characteristics. These young white men, privileged men who 
write these algorithms, and that is who does it, they are 
writing algorithms that discriminate against people. And there 
are laws against it in the real world, and they have to be laws 
against it online as well, because that is where the world no 
operates.
    If you are looking to get a job, if you are looking to get 
housing, and you are black or you are brown or you are LGBTQ or 
disabled, these young white men can write algorithms that help 
companies to avoid all of the laws that apply in the real 
world, and we know that. So, Ms. Gonzalez, talk about that, 
will you, a little bit and how this whole world of algorithms 
is just another cover for discrimination that were banned in 
the real world?
    Ms. Gonzalez. Thank you for the question, Senator Markey. 
Yes, one of the reasons that Free Press Action is very enthused 
about the legislation you put forward is because we know that 
discrimination exists. We have a long standing civil rights law 
that bans discrimination in housing and employment and in many 
other scenarios. And what we need is to clarify that those 
protections apply in the digital world as well. It is actually 
not rocket science.
    This is a long standing set of societal norms and laws, and 
they ought to be enforced in the digital world as well. I think 
what is even more pressing about how discrimination happens 
online by algorithms is that it is happening based off another 
abuse of our civil and human rights, which is an invasion of 
our privacy. So I think this is a critically important part of 
the problem that we need to address immediately. And I am so 
grateful that both your bill and COPRA address this.
    Senator Markey. Yes, so that is where we are. We have young 
white men who create algorithms, you know, algorithms that run 
from here to Osaka and back in a nanosecond that wind up being 
in a black box that no one can understand that then 
discriminates against black people, brown people, LGBTQ people, 
and many others in our society. And we just have to have 
transparency, but we need a safety and effectiveness standard 
to protect against discrimination online. Do you agree with 
that, Ms. Jackson?
    Ms. Jackson. Thank you, Senator, it is an important 
question. I would reframe----
    Senator Markey. But do you agree?
    Ms. Jackson. I agree that we absolutely need protections. 
And I believe that the most vulnerable in society are often the 
most harmed by online issues that we are discussing. But that 
is why I get back to the kind of systemic aspect of this----
    Senator Markey. But in this area. I am just talking about 
this area. In this area, it is systemic, but in terms of this 
harm, you do agree that it is egregious what goes on and how 
algorithms are used to discriminate? You do agree with that?
    Ms. Jackson. I agree. I would return to something I said 
before, which is over indexing on the power of algorithms 
intentions. Some of the greatest harm we have is when people 
put algorithms into the world without thinking about it and a 
lot of the discrimination, I agree with you, that comes out of 
that is actually the unthinking nature of the design of 
technology.
    Senator Markey. But it can also be thinking.
    Ms. Jackson. Absolutely.
    Senator Markey. You can actually design algorithms.
    Ms. Jackson. Absolutely.
    Senator Markey. So it can be young white men who are 
thinking in terms of how they discriminate against black and 
brown, LGBT, and disabled. But it can also just be an algorithm 
to set up in order to accomplish the goal that the redlining 
that is accomplished by a company in real life is also 
furthered by what they do online, the digital redlining. And 
its thinking or unthinking, but nonetheless, the impact is 
dramatic in terms of how it has a negative impact on families. 
So should we ban it? OK, let me go--let me go to you, Ms. 
Gonzalez, should we ban that discrimination online?
    Ms. Gonzalez. Yes, of course, anything that is banned in 
real life in terms of discrimination ought to be banned online. 
Now the tricky question here is whether the platform itself is 
engaging in discrimination or whether they are dealing with 
third party content that is discriminatory in nature. And so 
that is an important distinction. But I think the way that you 
set this out in your bill really gets to the heart of what 
platforms are doing and holding them accountable for their own 
actions, which is what we ought to be doing.
    Senator Markey. And thank you for that. So, yes, these 
hearings are all very helpful. They discriminate and they 
target children, OK, but they also discriminate against those 
who are most vulnerable in our society in addition to children.
    So that is why it is so important for us to legislate 
because we would never allow this to happen in real life. We 
would never allow people to engage in these kinds of 
discriminatory practices against minorities or exploitation of 
children. And same rules for the real world have to exist in 
this unreal world that has been created principally by young 
white men who get very rich in drafting these algorithms. And 
that whole, area has to--in our society. Thank you, Mr. 
Chairman.
    Senator Lujan. Thank you, Senator Markey. Senator Lee.

                  STATEMENT OF HON. MIKE LEE, 
                     U.S. SENATOR FROM UTAH

    Senator Lee. Thank you very much, Mr. Chairman. We often 
talk about Government regulation or requirements that would 
address these legitimate problems with big tech. One issue that 
I continue to grapple with--that it really is a challenge, I 
think we ought to think about it as we consider legislation in 
the tech space--involves the potential for deeper entrenchment 
of market incumbents.
    Dr. Eckles, I would like to start with you. Government 
regulation can be expensive in terms of compliance costs, 
correct? It costs a lot. Not everybody can do it. But those who 
are already there, those who have already arrived, who can 
afford a team of compliance specialists and accountants and 
lawyers have some advantage to the point where entry can be 
difficult to impossible it. Do you agree that regulation can 
have this kind of an effect?
    Mr. Eckles. Thank you, Senator. Yes, I think we have 
actually seen some of that in the context of GDPR, the European 
Privacy Regulation, where there is evidence that that increased 
consolidation as many firms stopped doing business with all but 
the biggest players who could ensure compliance.
    Senator Lee. So is there a way to thread that needle? How 
do we undertake that balance?
    Mr. Eckles. That is a great question, Senator, because I 
think actually as bad as some of the big tech giants are, often 
it is the smaller players who are doing some of the most Wild 
West things with our data. So it is not that they should 
somehow be exempted from all of these considerations.
    I would say one of the challenges with what we saw with 
GDPR was just how much ambiguity there was. So I would really 
hope that lawmakers would really be quite clear and precise in 
the laws or at least the policymaking that happens 
administratively.
    Senator Lee. Statutory ambiguity tends to lead to 
regulatory complexity. Regulatory complexity itself further 
entrenches incumbents. It is a subsidy of sorts, a subsidy on 
the wealthy and well entrenched. I have got grave concerns 
about how tech companies can use their algorithms to harm 
users. Sometimes it is through censorship. Sometimes it is 
through recommending inappropriate content to children.
    We actually see a lot of both of those things. But we 
shouldn't underestimate how Government solutions can often end 
up potentially making that situation worse. With that being 
said, transparency can be a good first step as a means to 
inform consumers. So in your view, what does transparency look 
like in the algorithm space and how can we balance that with 
the need to protect intellectual property?
    Mr. Eckles. Thanks, Senator. I think there is a lot of 
opportunities for transparency that wouldn't necessarily 
disclose a lot of the core intellectual property because so 
much of that is often in the context of making a specific 
prediction for a person, that this person is going to like this 
content, that they are going to watch this video for at least 
30 seconds, right. But then what do you do with that 
prediction? How much do you weigh that in your decisionmaking? 
That is more of a managerial decision, and that can be 
disclosed.
    So you can say something broad about what are the inputs to 
this prediction, and you can disclose what is done with that 
prediction, even if the prediction itself is made by some deep 
neural network with billions of parameters involved. So I think 
that is one angle for transparency. The other one is that these 
companies are constantly doing randomized controlled trials to 
compare different algorithms internally, to weigh a lot of 
different metrics, some that are just engagement metrics, but 
some that are metrics of potential harm.
    They could disclose more of the results from those. 
Actually I have relied in my testimony on the leaked Facebook 
files that covered one such A/B test that was conducted by 
Facebook comparing their status quo feed with a reverse 
chronological feed, which ended up making them more money. In 
fact, showing they weren't optimizing just for money making, 
but also had a bunch of other harms. Those kinds of comparisons 
are really useful, and so I would like to see a lot more 
openness about those in algorithmic impact and transparency 
reports.
    Senator Lee. So the light could be an effective illuminator 
and disinfectant. I would like to conclude with one thought, 
algorithms and the use of machine learning technology are 
undoubtedly a necessary piece of technology to deal with, to 
deal specifically with the continued use of the Internet and 
the increased activity that we see on the internet. We also 
know this technology can be used for nefarious ends, and we 
certainly want to stop those activities.
    I think we need to approach this challenge with humility, 
understanding that the Federal Government's intervention into 
these complex issues could and probably would do a lot of 
things to make the situation worse. Government regulation could 
end up hamstring companies' ability to innovate and evolve with 
emerging technology used by bad actors and empowering such bad 
actors to easily circumvent the technological standards adopted 
by Government, which would almost inevitably become outdated 
the moment they were enacted.
    There are certainly worthy challenges, challenges that I 
think Congress has to consider and try to deal with in whatever 
way we can responsibly. But I can't emphasize enough to my 
colleagues that we should urge caution and we shouldn't be 
thinking that Government solutions don't have great potential 
to set us back even further in pursuing policy updates in our 
Internet laws. Thank you.
    Senator Lujan. Thank you, Senator Lee. Senator Baldwin.

               STATEMENT OF HON. TAMMY BALDWIN, 
                  U.S. SENATOR FROM WISCONSIN

    Senator Baldwin. Thank you, Mr. Chairman. And thank you to 
our witnesses today. As you have heard, yesterday the Commerce 
Subcommittee on Consumer Protection held a hearing yesterday on 
protecting youth online, and our witness was the CEO of 
Instagram. I asked him about steps that the company was taking 
to combat harmful and problematic content relating to eating 
disorders on their platform and urged him to invest more 
resources into human reviewers to help tackle that problem.
    But as internal research from Meta shared with us by the 
whistleblower Frances Haugen has shown, it is Instagram's 
algorithms that can cause users to be drawn deeper into 
exposure to such harmful content. Ms. Gonzalez, do you believe 
that we can truly get to the bottom of this particular problem 
relating to eating disorders or other problematic content like 
hate speech or vaccine disinformation or misinformation by 
simply pushing better moderation by platforms? Or do we really 
need to address how algorithms amplify and magnify and target 
that problematic content to users?
    Ms. Gonzalez. Thank you for the question, Senator Baldwin. 
You know, I think this is really tricky because of course, 
Congress needs to abide by the First Amendment, and so there is 
a limit to how deeply you can get into content.
    However, what is really clear to me is that Congress has 
the ability and the responsibility to help the American people 
and help parents like me understand more about these problems 
by compelling transparency from the companies so we understand 
what is happening and how our children's personal and private 
data and their behavior is being tracked online and then used 
to target them with these kinds of messages, whether that is, 
you know, information about eating disorders or like in the 
case of my children, they are both big into gaming and watching 
other gamers talk about how they play the game.
    And I have noticed even with them, they are very young, 
they get pushed on a path where suddenly they were watching 
just really mean content. And so I think it is critically 
important to compel transparency and to set some limits about 
how all of our private data can be used, but especially our 
children's private data.
    Senator Baldwin. Thank you. In 2016, in that Presidential 
election, my home State of Wisconsin was one of the top targets 
for Russian influence efforts through targeted advertising on 
Facebook. Since then, we have seen numerous efforts by foreign 
and now domestic actors to use Facebook and other social media 
platforms to spread divisive content and misinformation. We 
have heard time and time again from these companies that they 
are taking steps to stop their services from being used to 
undermine elections.
    But I am concerned that we will keep facing this problem 
when we are also facing an erosion of public trust in our 
elections. Ms. Jackson, to what degree do you think Facebook's 
algorithms contribute to the spread and amplification of 
misinformation like that received by voters in Wisconsin in 
2016? And do you think Facebook or other platforms have taken 
sufficient steps to address how they can be vectors for 
divisive and misleading content that can undermine our 
democracy?
    Ms. Jackson. Thank you, Senator. It is one of the most 
important questions we can focus on. And I think what it comes 
down to is the ability of actors, whether that is the Russian 
state, a terrorist organization, or well-intended but 
misinformed neighbor to manipulate or be manipulated by a 
platform without their knowledge is a threat to our democracy. 
And there are a few things going on there that we should focus 
on. One is, you know, my organization was involved in something 
called the Election Integrity Partnership.
    We worked with Stanford and the University of Washington 
closely with DHS, CISA, and frankly, the platforms on a 
narrowly targeted issue, 2020 in response to 2016, to figure 
out how we could better track what was going on. And we found 
decent success in that partnership, but we noted that the 
platforms themselves had very different content moderation 
policies, as well as extremely different transparency standards 
for how they communicated those policies outward. So one 
starting point is ensuring that there are in fact content 
moderation standards, and second, that those are public.
    The second, though, is I don't think that self-regulation 
alone is sufficient, and it gets to the point that when the 
misinformation and disinformation that you were talking about 
in Wisconsin, that didn't just happen on Facebook, that is 
cross-platform. They used YouTube. They go into groups. They 
talk--they organize the actual events. And so what are we to 
say that, you know, is Facebook solely responsible or capable 
of addressing this problem, we would be missing a large piece 
of the puzzle. And returning again to the importance of data 
protection, privacy requirements, meaning that I would know, 
for instance, if the Russian State was attempting to buy ads, I 
was limiting the ability of the Russian State to target me on 
the basis of my predilections.
    All of those things will provide better safety and a safer 
and more productive environment across the board digitally, 
will also prevent our adversaries from manipulating us and 
harming our democracy. So it is fundamentally important, and I 
can say it 12 more times, privacy, privacy, privacy.
    Senator Baldwin. Thank you.
    Senator Lujan. Thank you, Senator. Senator Scott.

                 STATEMENT OF HON. RICK SCOTT, 
                   U.S. SENATOR FROM FLORIDA

    Senator Scott. Thank you, Chair. I think we can all agree 
the way big tech companies gather and manipulate and sell 
Americans' personal data, all without consequence, is alarming. 
These are the same companies that censor free speech, pick and 
choose which viewpoints are allowed on their platform, remove a 
United States President, but allow murderous foreign dictators 
on their platforms. For example, recent reporting has shown 
that Instagram's algorithms has recommended content to teens 
related to illegal drugs. Anybody with children or 
grandchildren that should be alarming.
    We have to find a way to hold these companies accountable 
for failing to protect the rights and privacy of all Americans. 
I have introduced the DATA Act, which would require large 
online platforms with over 30 million daily users to obtain 
user expressed consent to collect their data and preference to 
use an algorithm and provide Americans the ability for a 
private right of action if they believe that their privacy is 
violated.
    So my first question is for Mr. Poulos. I understand that 
these algorithms are designed to pull users back into the 
platform, pushing and prodding until the user engages again. It 
is not hard to see how that could cause someone to develop an 
attachment to these platforms and feel they can't escape. Do 
you believe that this is how these platforms are functioning? 
If so, what does that mean for users, especially young people 
and young girls?
    Mr. Poulos. Thank you, Senator. Human beings are imitative 
creatures above all. We are disputatious. We are jealous. We 
are envious. We are prideful. We are, in short, problematic. 
And social media gives human beings the ability to accelerate 
relationships and exchanges that involve, you know, all of the 
sins and foibles of human nature in a way that does, at its 
worst, create a kind of feedback loop.
    And I think, you know, it is undeniable that when you look 
at the way that mental illness is increasing in America, 
especially among young people, the way that frankly just 
bizarre and even barbaric forms of self-modification are 
spreading, I mean there are now even groups on social media 
where young people are ceasing to identify as individuals 
altogether and now identify themselves as members of a swarm. 
In some ways, I think this is clearly an effort to imitate our 
digital entities themselves in an effort to fit in in this new 
world. So we need to recognize that the digital revolution is a 
sweeping and deep and profound, and it is not going away.
    But that doesn't mean that we should either surrender 
ourselves to the rule of these entities or err in the opposite 
direction, which is to take an approach that thinks that the 
Federal Government can simply take control of the whole system 
and use it to really just take a side on the culture war and 
decide that Americans have to either get with the program or be 
systematically silenced or punished. And that is the power of a 
social credit system. You know, for all the flaws that social 
media has, speaking as a parent, you know, you can go on there 
and you can see all manner of memes and amateur commentary from 
young people who clearly understand what China is doing. There 
are plenty of memes out there about how you know John Cena's 
social credit score has gone to negative infinity because he 
said the wrong thing about China.
    Ordinary Americans can't quite explain what is going on 
behind the scenes, but I think that they are aware that they 
are being forced in a direction that is incompatible with our 
way of life. And so as we look toward algorithms, as we look 
toward the social media ecosystem, what we need to do is find 
the areas where there is broad based, popular, and, within 
Government, support for specifying exactly what harm is. I 
think there is a lot of disagreement right now, regardless of 
identity. You know, you just look at polling that shows that, 
you know, Hispanics are split on whether Donald Trump should be 
elected President again.
    You know, gay men--I mean, regardless of identity, there 
are deep seated cultural differences among Americans, and so 
there is not going to be an easy way to say, well, this is 
problematic, it should be suppressed. This is extremist, it 
should be suppressed. There isn't enough agreement. And so we 
have got to focus on really the most severe harms. Start there, 
build trust as a Government that the people are actually going 
to be able to preserve their human identity and their community 
identities online.
    Senator Scott. Mr. Poulos, other than to make money, is 
there any reason for big tech to have these algorithms? Is 
there any user focused reason that these are important or are 
helpful to a user?
    Mr. Poulos. Thank you, Senator. Yes, I mean, I would 
emphasize again that Stephen Wolfram was testifying on the Hill 
recently, and basically the point that he made was you can't 
have digital technology that interfaces of human beings without 
having in some way sort of algorithms that have a purpose and 
that direct people in one way or another based on what the 
technology is able to identify is their own preferences or 
leanings. And there is value at that. And consumers, who have 
clearly expressed a preference for being delivered ads that 
they are inclined to be interested in, rather than things that 
they are not inclined to be interested in.
    I know that there is an appreciation for, you know, why 
can't we go back to the days when it was just a chronological 
feed of our friends posts? Well, you know, sometimes that ends 
up being a better user experience and sometimes it doesn't. So, 
you know, they are commonsensical reasons why the sort of 
algorithms have emerged in the way that they have. And I think 
that one source of frustration is sometimes we get in the habit 
of taking an outcome based measure of judgment, where ``oh, it 
looks like something is happening to people that we associate 
with some kind of injustice, and so therefore, whatever was 
going on in the black box has to automatically be bad, and we 
need to punish--open the lid and just punish whatever is in 
there because we are focused on outcomes.''
    And I don't think that is right way to do it. I mean, you 
know, with all due respect to Senator Markey, I think his claim 
that, you know, young white men rule the Internet would fall 
flat among the many, you know, H-1B visa holders who are 
explicitly brought into this country and privileges as 
applicants so that they can build the technological 
infrastructure of online life.
    Senator Scott. Thank you. Thank you, Chairman.
    Senator Lujan. Thank you, Senator Scott. The Chair of the 
Full Committee, Senator Cantwell.

               STATEMENT OF HON. MARIA CANTWELL, 
                  U.S. SENATOR FROM WASHINGTON

    The Chair. Thank you, Senator Lujan. I want to thank you 
and Senator Thune for having this very important subcommittee 
hearing, and again a lot of my colleagues for good attendance, 
good questions, and good discussion by the witnesses, so thank 
you. Dr. Eckles, I think I want to start, though, you mentioned 
a little bit of this in your testimony, but algorithms are just 
software, right? I mean, we are talking about software here.
    I feel sometimes like any technological advancement gets a 
little bit, you know, the Internet got critiqued, you know, for 
the things that it enabled, and drones get critiqued for the 
things they enable, but the underlying technology in general is 
something that we support. Me, I am most excited about every 
particle in a storm now can be an algorithm, and if you put the 
government dollars behind supercomputing, then you will have a 
really good weather forecasting system. So there are great 
applications.
    The discussion we are having here today is how people 
misuse or what are the problems with those algorithms. And I 
think one of the things, you know, my colleague, Senator Thune 
and Schatz have been discussing this issue of 230. But when we 
first looked at content moderation, I don't think we really 
even thought that it was going to be automated. I don't think 
we were in that mode of this automation. And now we are--it is 
automated, and the algorithms have varied choices in them. In 
fact, they are editorializing. Is that not correct? They are 
making choices and editorializing about content?
    Mr. Eckles. Thanks, Senator. I don't know if I necessarily 
would describe it as editorializing, but I think fundamentally 
it is unavoidable that a lot of this has to be automated, 
right? If we want it to be cheap for so many of us to be able 
to share things with other people, we are not paying enough to 
have somebody on retainer to review all of our content, right.
    Our clicks on ads are not making these companies enough 
money for there to be a person on a retainer to review all my 
content personally. So some of that automation is unavoidable. 
And I think it makes sense to have some scrutiny on where those 
lines are for what stays up and what gets taken down.
    The Chair. But would you call that choice?
    Mr. Eckles. Yes, I think I think that is a choice, and I 
think there--one of the directions that I have seen this 
promising is having more choices there, so that it is not just 
leave up the content or take it down, but there is a larger 
repertoire of choices by the platform. So they can say, look, 
we, you know, this photo could be disturbing. We don't want to 
show it by default. But if you wanted to see it, you know your 
friend posted this content. Or ranking things, part of what 
they do through algorithms is to say, you know, this content is 
there, you can seek it out, we don't necessarily need to take 
it down, but we don't have to put it in front of you.
    So actually, I think that a lot of algorithmic ranking and 
recommendation and some of the choices about what to show, how 
big, what to open up right in your face already, those are more 
subtle and really beneficial ways for them to make those 
choices. So it doesn't have to be about we banned this person 
entirely from the platform. Take down all their content.
    The Chair. Well I think the question you have been 
discussing here for some time, or at least the gist of it I am 
getting is that again, when we looked at the 230 issue, an 
exemption, it was about you, you know, protecting people just 
for being publishers. But now with algorithms, you can be very 
precise, and you can have practices, as were discussed by 
either Ms. Jackson in her testimony or Ms. Gonzalez, that are 
endangering either people's privacy or discriminatory, as some 
of my colleagues, or even promulgating hate.
    So the question then becomes, what do you want to do about 
those individuals--because they have become far more than just 
a publisher. They really have increased their activities. And 
so I wanted to ask this question because Mr. Poulos, you 
brought up real harm. We had this discussion yesterday with 
Instagram about what happens when there is real harm. I guess 
actually, this is for all the witnesses, because your 
testimonies are talking about what to do about these if there 
is real harm. I think Ms. Gonzalez and Ms. Jackson are saying 
that we should have legislation and that has been loud and 
clear. But I wanted to hear whether you thought Mr. Poulos that 
real harm should be more than binding arbitration with Facebook 
or Instagram.
    Mr. Poulos. Thanks, Senator. In the abstract, you know, I 
think I mean, I am one of these people who thinks that binding 
arbitration is usually not the right way to resolve legal 
disputes. So I am inclined to say that, you know, if Congress 
wants to ensure that there is a more active legal process, that 
is in their remit to do so. You know, my concern is more along 
the lines of trying to split hairs and pick and choose as a 
deliberative body over what gets to count as harm is going to 
be difficult on the internet. And it is----
    The Chair. But you think real harm, as some of the 
witnesses that we have had here for or my colleagues yesterday 
and their hearing, I know it is interesting our two committees, 
subcommittees, but you know, my other two colleagues who have 
been doing a fine job, Senator Blumenthal and Senator 
Blackburn, have been showing the harm to kids. So in that case, 
would you agree that some of those cases are real harm?
    Mr. Poulos. Thanks, Senator. I mean, rather than there 
being a sort of epistemological, unsolvable question about 
whether people are suffering, I think what you would probably 
find in most cases is that once these things go to court, it 
would be difficult, either too difficult on the one hand to 
meet legal standards of connecting those dots.
    Or on the other hand, it would push judges and juries too 
hard in the direction that, you know, oftentimes happens in our 
litigious society, where a plaintiff just shows up and says, 
you know something bad happened to me, now give me $20 million. 
You can have sort of, you know, you kind of perversions of the 
legal system, I think, in either direction. You know, that 
said, you know, arbitration is not the be all and end all, 
either.
    The Chair. OK, Ms. Jackson, Ms. Gonzalez, do you want to 
comment on that? The need to show--to do more on real harm.
    Ms. Gonzalez. Senator, yes, I would agree we need to do 
more on real harm. You know, in the context of 230, one of the 
things that we have been thinking about at Free Press Action is 
how the intent of that statute has been overread by the courts. 
And so clarity around platform accountability when they 
themselves, the platform itself, either by design or 
negligently is causing harm, ought to be reviewable by courts. 
So that is kind of where we stand on the Section 230 issue.
    And then of course, we have long standing laws, civil 
rights, privacy, etc., norms and principles that ought to apply 
online as well. The harms here are really concrete. I don't 
think it is out of the regular order of the Congress or, you 
know, especially an expert agency like the Federal Trade 
Commission to look at the facts before them and to promulgate 
rules that apply across the industry and to take enforcement 
action when companies violate those rules.
    The Chair. Thank you. I do see my time is over, but Ms. 
Jackson, did you want to add something really quickly?
    Ms. Jackson. Yes, just to double down there. There is 
criminal content online that we are talking about, and so over 
focusing on whether or not there are harms or distract from the 
fact that real people are being hurt and everyone else in the 
world is making decisions about this about us.
    And so I think it is really important, we can have debates 
about which legislation, or which approaches or what we do, but 
the United States not acting is ceding one of the most 
consequential decisions of our time to literally every other 
country, including China, and having that be the law of the 
land eventually for our own citizens. And so just really 
encouraging that we take action. Thank you.
    The Chair. Thank you, Mr. Chairman.
    Senator Lujan. Thank you, Chair Cantwell. Senator Peters.

                STATEMENT OF HON. GARY PETERS, 
                   U.S. SENATOR FROM MICHIGAN

    Senator Peters. Well, thank you, Mr. Chairman. Dr. Eckles, 
I chaired a Senate Homeland Security and Governmental Affairs 
Committee, and our committee has been focusing on the rise of 
domestic terrorism in our country. And as part of the efforts 
that we have been undergoing, we are looking at social media 
companies, their practices, their politics, and how that can 
feed into extremist content being provided to individuals.
    So my question to you is, in your expert opinion, what are 
the tradeoffs that these social media companies are making 
during their decisionmaking progress, process when they weigh 
profit motives verses user satisfaction? Walk me through a 
little bit about what is going on behind the scenes there.
    Mr. Eckles. Sorry, Senator, was that a question for me?
    Senator Peters. Yes.
    Mr. Eckles. Yes. I think that is a great question because 
they are weighing a lot of different considerations. And one of 
the things that actually Adam Mosseri brought up yesterday was 
the notion of the long term versus the short term. And so I 
think this is one of the things that as platforms get more 
sophisticated, they hopefully start to consider more and that 
can improve some of their decisions, which is that they think 
more about the long term, right. You might be able to show 
something different to somebody today and make a little bit 
more money, but how are they going to feel about this tomorrow?
    How are they going to feel about the time they spent on 
your platform? Are they still going to be around in a month if 
a new competitor enters? Won't they want to switch to this 
other platform? So I think the platforms are in some cases 
getting more sophisticated about thinking about the long run 
and including measures of consumer welfare and time well spent. 
And so I think the industry really pushing other companies to 
do more of this could be quite beneficial.
    So those are tradeoffs that they are making, and I would 
like to see you all in this subcommittee and other policymakers 
provide more scrutiny on what those tradeoffs are, like what 
actually are the weights that they are giving to some of these 
different outcomes that they care about? How much weight do 
they put on revenue versus how much weight they put on consumer 
surveys?
    Senator Peters. Yes. Very good. Well, thank you, doctor. I 
have another question for you. In your written testimony, you 
state that platforms work to predict what content a user is 
going to engage in, and clearly that is the basis behind 
algorithms and machine learning. But could you just describe 
any role that psychologists and behavioral scientists may have 
at these companies and in predicting potential behavior from 
folks and how to model accordingly?
    Mr. Eckles. I think a lot of these companies now have 
behavioral and social scientists from all disciplines involved 
in this process. And in some ways that could make us a little 
worried. On the other hand, I think actually that can be quite 
beneficial. So it is psychologists and behavioral economists at 
Facebook who helped to pioneer there the focus on actually 
asking people questions about, hey, was this content you just 
saw, was that important, was that interesting, did it make you 
feel connected, right?
    This idea that sometimes what we want in the moment and 
what we want on reflection can be different. That is an idea 
that has a lot of grounding in research and psychology and 
behavioral economics and could help platforms actually focus on 
giving people what they really want on consideration, not just 
what say their lizard brain wants in the moment. So I think 
these companies are, at least the giants, are employing lots of 
social scientists, lots of psychologists.
    And I think from what we have seen in some of the internal 
conversations that have been leaked from, say, Facebook, a lot 
of times they are advocating actually for consumer well-being. 
So it sort of depends on some of the top level decisionmaking, 
whether those choices get put into practice.
    Senator Peters. And one final question there for you, 
doctor. You mentioned in your testimony that WhatsApp can also 
be dangerous despite not having an algorithm. Could you talk to 
us about your study that looked into how WhatsApp users 
manipulated the trending topics feature on WhatsApp in favor of 
BJT and India, please?
    Mr. Eckles. Yes, definitely. So I think, you know, 
qualitatively, we look at some of these cases like genocide in 
Myanmar, and it is easy to sort of connect that with 
algorithmic ranking and recommendation. But then we also see 
violence that sort of directly facilitated or caused by 
activities on platforms without ranking like WhatsApp. And so I 
think we have to remember that some of the fundamental things 
about social media are that it lowers costs for people to share 
content with lots of people and to propagate it really quickly, 
right, whether there is some algorithmic ranking or not.
    And a lot of times, if the algorithms are really 
transparent, people learn how to game them. So we had a whole 
bunch of phones in India that we are joining all the public 
political groups we could find on WhatsApp and tracking what 
the conversation was happening there. I think in many ways we 
knew more of what was happening than Facebook did because those 
conversations are all encrypted unless somebody lets you join 
that group. We were able to see then these groups coordinate 
their activity on Twitter so as to manipulate the trending 
topics there.
    And that kind of cross-platform coordination happens all 
the time. And I think that is something that Ms. Jackson has 
really been highlighting. And so we have to definitely look 
across those platforms, give platforms clarity that they can do 
so without risk of anticompetitive concerns if they are doing 
it at least within a particular context of preventing these 
kind of cross-platform manipulations.
    Senator Peters. Well, thank you, Dr. Eckles. And thank you, 
Mr. Chairman, for holding this hearing.
    Senator Lujan. Thanks, Senator Peters. Senator Rosen.

                STATEMENT OF HON. JACKY ROSEN, 
                    U.S. SENATOR FROM NEVADA

    Senator Rosen. Well, thank you Chairman Lujan, and of 
course, Ranking Member Thune. This is a really important 
hearing, and I appreciate all the witnesses being here today, I 
want to build upon what some of my colleagues have been talking 
about but taking a little bit different direction, 
deplatformization. And so many mainstream social media 
platforms have been publicly taking action against hate speech. 
They are removing it from their sites, they are removing it and 
banning account holders. In some cases, a fully de-platform 
users from chat communities and social media sites.
    While significantly more needs to be done, I appreciate the 
actions these platforms are taking to tackle racism, 
antisemitism, and other forms of white supremacy rhetoric and 
hateful conspiracy theories that far too often lead to 
violence. But when dealing with fringe platforms, there have 
been efforts to remove platforms altogether from their hosting 
sites, their app stores, cloud hosting services, instead of 
removing the hateful content or user. Total deplatformization 
strategies deny such fringe sites and apps access to platform 
infrastructure from which to perpetuate hate speech.
    This could include suspending the use of payment systems 
such as PayPal or Cash App or completely removing companies of 
browsers and domain name system. So Ms. Gonzalez, is 
deplatformization of fringe sites that support hate speech, do 
you think this is an effective strategy and how easy or 
difficult is it for a platform that is posting extreme, 
radical, or violent content to find a new hosting or payment 
service if kicked off a mainstream company? And should these 
host sites and payment services, what is your responsibility in 
this mix?
    Ms. Gonzalez. Thank you for the question, Senator. An 
advocacy that is not focused on request to the U.S. Government, 
I actually support deplatformization and I support it because 
for two main reasons. One, I think that the companies 
themselves do have First Amendment rights to set the standards 
upon which people have to behave on their site or on their 
platforms. And if there is a platform that is just repeatedly 
violating the standards of, say, the App Store, then I think it 
is within the app stores rights to remove them.
    The other consideration that brings me to this position is 
that even if these platforms or entities are kicked off of 
these--or deplatformed, they can build their own sites, they 
can build their own payment systems. And this is why at Free 
Press Action we do support net neutrality. We don't think that 
Internet service providers that provide access to the Internet 
should be making choices about who gets to build a website or 
who gets to run over the internet.
    So I think that is an important distinction here as we kind 
of balance First Amendment issues with the desire to hold 
companies accountable for proliferating hate and extremism.
    Senator Rosen. Thank you. I would like to move on to talk 
about another thing everyone has been talking about, the 
algorithms. And we all know how they are built to engage with 
the user, they are built to direct the user's attention toward 
advertising or something else. They are just really going to 
take you further in. If you like red sweaters, you are going to 
get a whole slew of red sweaters once you start looking for one 
of them. But this involves amplifying content far too often 
incendiary in nature, which in turn optimizes engagement. So I 
know I am not going to have time for the entire panel so we can 
begin with Ms. Gonzalez and then Dr. Eckles.
    What can Congress do to incentivize or discourage companies 
from using these persuasive technologies that manipulate or 
harm users? And I also want to know what you think about giving 
the user control over, if I am shopping for red sweater for a 
friend, and then I don't want to see any more red sweaters, how 
can I stop that--where do I have a choice? And so what can we 
do about that? So Ms. Gonzalez and Dr. Eckles. I know I am 
running close to my time.
    Ms. Gonzalez. Thanks, Senator Rosen. I will try to be 
brief. Even though I am all for red sweaters, so keep serving 
those ads.
    Senator Rosen. For holiday season, red sweaters----
    Ms. Gonzalez. It is very beneficial, in my view. But I 
think this is where having a comprehensive approach to privacy 
is critically important. We should understand what companies 
are collecting about us and have rights to, you know, compel 
them to remove data that we don't want them to be tracking 
about us.
    And so we need to have those rights in the users' hands, 
and we need a great deal more transparency about what they know 
about us and how they are using that information. I think too, 
a lot of companies know, as the Facebook papers have revealed, 
that they are causing harm. So we need to have some disclosure 
when companies become aware that their algorithms are driving 
harm and abuse.
    Senator Rosen. I know that I am over my time, so I guess I 
could--I will ask the rest of the panel. We will submit this 
for the record, but I really want to know what people think 
about allowing the end user to opt-out of getting any more 
recommendations on a particular topic or subject or even a red 
sweater, if you will? Thank you, Mr. Chairman. I appreciate 
this hearing.

             STATEMENT OF HON. RICHARD BLUMENTHAL, 
                 U.S. SENATOR FROM CONNECTICUT

    Senator Blumenthal. Thank you, Senator Rosen. My name is 
Richard Blumenthal. I am a Senator from Connecticut. I am 
taking over for Senator Lujan who is voting right now, and I 
have a strong interest in this area. As a matter of fact, just 
coincidentally as Chairman of this Subcommittee on Consumer 
Protection, yesterday, we had a hearing with the CEO of 
Instagram, as you may know.
    We got from him, I think he used the word 
``directionally,'' support for access of researchers to these 
algorithms, which are the black boxes that drive destructive 
content to kids on platforms like Instagram and others. TikTok, 
Snap, YouTube all committed to making datasets about their 
effect on children and teens available to independent 
researchers when they came to testify. So this idea of access 
by researchers is tremendously important so that parents and 
all of us understand how they work, and what they do, and what 
can be done to stop the destructive content.
    And I think that is one of the main values of this hearing. 
I was very interested, Ms. Gonzalez, in your comments, page 10 
of your testimony, where you say and I am quoting, ``we must 
also set out explicit protections for external researcher 
access to platformed data to guard against what is now a 
documented pattern of targeted efforts by platforms to deny 
external researchers the opportunity to investigate platform 
practices.'' I think that comment is tremendously revealing. 
After I have sat here and I have heard Antigone Davis of 
Facebook, Adam Mosseri of Instagram, a representative from 
Snapchat, TikTok, YouTube, all say, we support research.
    But its research of our choosing, people of our choosing to 
do the research. And footnote 19, Ms. Gonzalez, I think 
documents the denial of access to independent, objective 
researchers that could uncover more of the truth for parents 
and others who really deserve it. And I know that you cite the 
example of Laura Edelson of NYU who was, in effect, stifled in 
her research. And so my questions relate to the efforts we 
should do legally, what legal protections should be afforded to 
academic researchers against efforts to silence them or stifle 
their research? Let me ask all of you to comment.
    Specifically, why voluntary efforts from the platforms are 
inadequate and why we need these requirements for access rather 
than self-policing or internal research? And maybe Ms. 
Gonzalez, I can begin with you, since you explicitly called 
attention to this issue in your testimony.
    Ms. Gonzalez. Thank you, Senator Blumenthal. You know, 
through my work and various coalitions, the Change the Terms 
Coalition, as well as the Real Facebook Oversight Board, I 
worked with a number of researchers. So in addition to Laura 
Edelson story, I have heard time and again researchers being 
granted access to data and then having their access denied when 
the research starts to get real in terms of holding the 
companies accountable. So unfortunately, I wish we weren't in 
this place, but I think we are, and the Facebook papers 
certainly have revealed that we cannot rely on companies to 
give access to their data without some mandates.
    So unfortunately, that is the place we are in, and we have 
to develop some safeguards, protections, and expectations about 
access to data by third party researchers. Otherwise, we are 
just going to continue to not understand how these systems work 
or it is just, you know, Facebook for it is such a good example 
because now we know just how much they knew about how their 
systems were harming people and that they failed to disclose 
it.
    It took a whistleblower, Frances Haugen and many other 
whistleblowers, Yael Eisenstadt. The list goes on and on to 
really unveil just how harmful these practices were. Last 
year--and then I'll also stop, I promise. Last year, the Wall 
Street Journal expose that found that Facebook knew that 64 
percent of people who go to extremist groups on their site find 
those groups because of its own recommendation system, that was 
a blockbuster revelation, and Facebook had known that for years 
and failed to disclose. We just can't rely on them to be 
forthcoming.
    Senator Blumenthal. Thank you. That is really powerful, Ms. 
Gonzalez, and perhaps in writing, you can elaborate on some of 
those points, but they are very, very pertinent and important. 
Ms. Jackson or any of the other witnesses, if you--go ahead.
    Ms. Jackson. Thank you, Senator. Yes, my organization, the 
Digital Forensic Research Lab is one of those independent 
research groups. And so we pay very close attention to this 
question and share the belief that this is extremely important 
and in fact, that there is a role for Congress to play and 
requiring certain levels of transparency and disclosure. I 
would also urge us to be very intentional about what it is we 
are asking for. I know we talk a lot about Facebook, and part 
of the reason for that is we actually have more information on 
Facebook than almost any of the other platforms.
    We will give Twitter a little credit and that they provide 
better data access probably than the rest. But we are not 
talking about YouTube. We are not talking about Google. We are 
not talking about Reddit. And there is a reason for that. And 
that, to me, underscores the importance of having those 
standards because one of the risks in research is that you 
chase the thing you have, which both leads us to not look at 
the kind of overarching aspects of platforms that cause some of 
these problems.
    I am encouraged that we are talking about algorithms today, 
which is an improvement from a conversation a year ago where we 
were only focusing on content. But there are many more examples 
of exactly that, how platforms operate, how they operate 
between each other that we need to have a sense for. And to get 
to that, we need to have intentional transparency, standards 
across the board for companies to provide certain information, 
and also have clarity about what we want from that transparency 
at the end of the day, so we are not chasing unending rabbit 
holes of specific pieces of problems.
    Senator Blumenthal. Thank you. Thank you very much. Mr. 
Poulos.
    Mr. Poulos. Thank you, Senator. I think it is self-evident 
that the platforms that want to choose their own researchers so 
that they can produce a relatively favorable information in the 
same way that you know a legislative body or even an 
independent researcher would want to tend to arrive at the 
conclusions that they think are the most important to reach. So 
I think to a degree, you know, if you are looking to go toe to 
toe with the platforms, with dueling researchers, it is going 
to have to be a somewhat disputatious process.
    This is a turf battle in some respects, and if Congress 
wants to mix it up, then they are going to have to mix it up. I 
would caution, you know, that in addition to, you know, chasing 
after what you already have, I think there is the danger in 
research of chasing after what you want to find.
    And that there are those who outside of Silicon Valley and 
outside of social media, really want the social media space and 
the Internet more generally to become a place where really 
predictive policing governs the space, so that, you know, you 
can say, well, we need to make sure that three steps in the 
decisionmaking tree before someone commits an act of extremism, 
that they aren't exposed to information that might lead them to 
that point. And you can chase down that rabbit hole quite far.
    And I think there is a real danger that at least some 
outside researchers would want to push social media through 
regulation, if not legislation, right, toward really doing 
using algorithms to conduct a kind of predictive policing. 
Where, you know, I mean, just to grab something out of the hat. 
You know, if going to a certain kind of political rally or 
expressing a certain kind of political opinion is officially 
deemed an extremist act, well, then you can, you know, you can 
find how far backward in time on social media do you really 
want to go still pinning responsibility on a person.
    You know, while they were exposed to this piece of 
information or this opinion Op-Ed or, you know, to this kind of 
meme, and so we need to stop those things because, you know, 
they have some inclination to lead people in an extremist 
direction. I think that stuff can be very dangerous, very 
suppressive, chilling of free speech, free expression, online 
and off. And so that is, you know, that is the warning that I 
would counsel.
    Senator Blumenthal. Mr. Eckles.
    Mr. Eckles. Thank you, Senator. I think allowing external 
researchers, whether they are academics or journalists or 
Congressional staffers, to do research, whether they are 
probing the systems from outside, you know, creating their own 
accounts and scraping data that is really valuable. And 
sometimes there has been legal uncertainty about the state of 
that. Some recent case law, my understanding, suggests that 
researchers can do a bunch of things there, but that may be a 
point where Congress can help.
    Another way that researchers can be helped is with actual 
access to proprietary data. And I have been in the position of 
having that access previously working as a scientist at 
Facebook, and I have data use agreements with multiple 
companies that allow me to do that kind of research. But as you 
say, there can be some limitations on the topics that you might 
be able to undertake. And so I think that means that the public 
gets potentially a skewed view of things. Now there have been 
some efforts like social science one with Facebook to release 
some privacy preserving data to a larger set of researchers 
without any publication approval from the firm.
    And I think that is a really promising direction. That 
actually used some of the same technology, differential privacy 
that is being now used by the U.S. Census Bureau for the 
decennial census and for more of their data products. That does 
raise some challenges that can reduce some of the data quality 
in some ways. But I think that is a really promising direction, 
and I would hope as well that privacy regulation would make it 
very clear that platforms can share data of that kind with 
external researchers if they use those techniques to protect 
the privacy of people involved.
    There has been a lot of uncertainty about that. As soon as 
Facebook announced that they were going to share that data in 
that way, activists said, we are going to sue you in European 
jurisdictions. So having that clarity could really help spur 
that along. Thank you, Senator.
    Senator Blumenthal. Thank you all. I think this area is 
very, very important. You know, the Surgeon General on Tuesday 
asked for more data sharing from social media platforms in an 
advisory. So now we now have public health professionals 
calling for researchers to have access so they can hold big 
tech accountable. A couple of you mentioned the Digital 
Services Act, which is a proposed European law that would 
impose this kind of access as an obligation.
    And I think that we can learn from Europe and others, but 
we don't need to wait for them to act. I think we should act 
because we need to protect the public, and for all the reasons 
that you mentioned. And ultimately it actually is good for 
social media to have more competition, more transparency. You 
know, the Internet initially was designed to be a lot more open 
than it has become. Of course, we can learn from history. So 
thank you all for being here.
    Senator Lujan. Thank you, Senator. And before we close, 
just some follow-up questions to the panelists. Professor, 
following up on your conversation with Senator Fischer, could 
you define A/B testing for the Committee?
    Mr. Eckles. Thanks, Senator. An A/B test is just one form 
of a randomized trial or randomized controlled trial where 
people are randomly assigned to different variations on a 
service. And so in the context of algorithmic ranking, it might 
be that some people are randomly assigned to one status quo 
ranking algorithm and other people are randomly assigned to see 
content according to some new algorithm. This is one of the key 
tools that these companies use to make decisions about these 
algorithms to balance all the different metrics that they are 
tracking. They are running hundreds, thousands of these kinds 
of experiments a year.
    I think we know as of 2016, Google was running 10,000 A/B 
tests a year. Just as they are in medicine, these are the gold 
standard for causal inference, for learning about cause and 
effect relationships of if we change things in this way, what 
are the consequences? And so while they can potentially be used 
to do things like fine tune, some dark pattern, deceptive 
practice, they also are one of the main tools that platforms 
also have to make educated and good decisions. So I think, you 
know, in my own testimony, I relied on some A/B tests that I 
had run and published when I was a scientist at Facebook. Some 
that were leaked as part of the Facebook files.
    One that was published by Twitter that actually didn't 
look, according to their own conclusions, didn't look so 
favorable to them, but they chose to actually publish that. So 
I think we would very much like to see as a research community 
and as the public, the highest quality evidence available, and 
that comes from these randomized trials or A/B tests.
    Senator Lujan. Now with that being said, what policies 
would you suggest that would better incentivize A/B testing for 
consumer harms rather than exclusively for engagement and 
revenue?
    Mr. Eckles. I think that is a great a great question. I 
can't say that I know the answer there. I think hopefully there 
are some lessons that budding entrepreneurs are learning now as 
well that some of these early decisions that platforms made 
that were short sighted have permanently hurt their brands and 
are hurting their ability to launch new products. Think about 
Meta as they are trying to move into all these metaverse 
offerings.
    The Facebook reputation is coming with them, and it may 
prevent some of their ability to succeed in that area. When it 
is time to move to a new setting where they don't have the 
network effects they have in an established social media space, 
will people want to use their service? So these matters of long 
term reputation and consumer attitudes can be one lever, but I 
don't know the full answer there, Senator.
    Senator Lujan. Appreciate that. Just for clarification, 
GDPR is the European General Data Protection Regulation, 
correct?
    Mr. Eckles. Yes, Senator.
    Senator Lujan. Ms. Jackson, following up on Senator 
Blackburn's question regarding corporate power over user data 
internationally, what should the U.S. Government do to better 
protect personal information and behavioral data from foreign 
actors seeking a strategic advantage?
    Ms. Jackson. Thank you, Senator, for that extremely 
important question. I think the lack of our data privacy means 
that literally anyone, whether that is the Chinese State 
collecting information, a terrorist organization, an 
intelligence agency, or just a marketer can gain access to our 
information and then deploy that into the system that we have 
all described today to achieve whatever objective it is. And I 
think there's no question that that is bad for American 
national security, that that is bad for American competition in 
the world, and it is bad for American democracy.
    What I think we sometimes miss is that we are having this 
conversation on an island, and all around the world, we 
mentioned GDPR, the data protection rules in Europe, almost 
every country in the world has some data protection laws, 
including even China. The difference is that China has data 
protection law for commercial entities and not for itself. And 
so while in the United States, we paid decently close attention 
to questions about limiting the Government's ability to surveil 
us, we have completely left out the commercial picture. But 
that is a giant backdoor for whoever wants to influence us.
    So I appreciate that question and just return to as the 
rest of the world is making these decisions, it is of extreme 
urgency that the United States begin to make its own decisions 
about how we protect our own citizens, our own data, and our 
own democracy so that we don't leave it up to the Chinas of the 
world to make those decisions for us.
    Senator Lujan. And you anticipated my next question, which 
is what are the consequences for the United States not taking a 
leadership role in privacy and algorithmic regulation? I think 
you touched on that very clearly. My concern is everyone around 
the world and especially the Europeans, are defining this. And 
here in the United States, we are going to be defined by what 
other countries are doing with establishing these rules. And it 
just it seems like a terrible mistake for the United States not 
to be engaging more. Ms. Jackson.
    Ms. Jackson. Absolutely. If I can add a point that. I think 
one thing that often gets missed is that I can't remember which 
Senator alluded to earlier, the kind of strength of the 
Internet is the fact that it was open, interoperable. It's this 
public thing that is not just in the United States, it is every 
country around the world. And so to think that we can have, you 
know, 100 different versions of the Internet and have it be the 
thing that we want and have the rights that we expect offline 
transferred to the online world, we are in for a treat.
    And so part of this is that, you know, China is advancing a 
very intentional model, and that intentional model depends on 
every other country being divided in its approach because the 
splinternet, as some people have called it, is the goal so that 
China can do whatever it wants in its own territories. You have 
mentioned the social credit system. China is the author of a 
system in which people exist online, and the data collection 
that we are talking about here in the U.S. is used for the 
purposes of the State being able to institute its political 
control.
    That is an actual authoritarian state. In the United 
States, we have a very different model and set of intentions, 
and so it is extremely important that we don't seed that 
dispersal. So it is not just that we are absent, it is that our 
lack of leadership means that China having a clear vision, 
bringing the next billions of people in the world online for 
the first time into their system really ends up impacting the 
ability of democracies to have a democratic online space.
    Senator Lujan. I appreciate that. Ms. Gonzalez, the last 
question I have for today. All Americans, regardless of the 
language they speak at home, deserve access to services with 
online platforms and all that they provide. They deserve the 
same access to stay connected to their friends and to loved 
ones. They also deserve access to the critical information 
available through online platforms, which is why the spread of 
online amplification of non-English and a particular Spanish 
language misinformation is deeply troubling.
    I repeatedly called for online platforms like YouTube and 
Twitter, Facebook and others to publish more information broken 
down by language so we can begin to understand the scope of the 
problem. Now, Ms. Gonzalez, what other commitments should 
online platforms make to stop the amplification of non-English 
language misinformation and disinformation on their platform?
    Ms. Gonzalez. Thanks for the question, Senator. I think, 
you know, this is near and dear to my heart that I have been 
working on this for a while now. I think one thing that really 
struck me in the Facebook paper revelations was that Facebook 
has seriously underinvested in Spanish and other non-English 
languages when it comes to how it shores up its integrity 
system. One of the things that really shocked me in the 
response or kind of lack of response that you got from the tech 
companies was that they failed almost across the board to 
answer really basic questions about how they keep their users 
safe in languages other than English.
    So they weren't willing to answer simple questions like who 
is in charge of non-English moderation in language x, how many 
moderators do they have? We know that human moderators are a 
critically important part of the content moderation system, 
right. One of the things that the Facebook papers reveal is 
that to really get at, you know, beyond capturing 20 percent of 
hate speech, for instance, just take an example, that AI alone 
wasn't going to catch that, and that Facebook needs to 
significantly step up.
    It is very clear that there is this disparity in how 
Facebook, YouTube, WhatsApp, and other platforms are forcing 
their own content moderation rules across language. And that 
needs to stop. I think users, no matter what language they 
speak, ought to have the same protections across the board.
    Senator Lujan. And what seems more clear to me, Ms. 
Gonzalez, is I have gotten more involved in this space is the 
disaggregated data is collected. I mean, the AI is built to 
tell the difference, but no one wants to release it. Even to 
the point that the letter that was responded to my inquiry from 
Facebook, it came to me unsigned.
    Everything is OK, we collect this information. I just 
wonder if a bot even produced that letter before it came to me. 
You know, some algorithm or some AI just said, OK, here is this 
question, spit out the response, and then, you know, we will 
just send it to him unsigned, so he doesn't ask any questions 
of who needs to be held accountable for the content of that. 
Anyhow, that is the subject for another hearing down the road. 
With that being said--yes, Ms. Gonzalez.
    Ms. Gonzalez. If I may, I mean, one thing that came out in 
the L.A. Times piece, written by Brian Contreras a couple weeks 
ago was really upsetting to me. We would been meeting with 
Facebook, calling their attention to this problem, and you know 
ahead of the 2020 election, we said there is a gap. And they 
were like, ha, maybe there is a gap. Well come to find out, 
they knew about that gap in February 2020 and decided not to 
invest to close the gap. So this is why I am on such--I am so 
much for calling for greater transparency because they know 
that there is a problem, and they just don't do anything about 
it and don't let anyone know that it is happening.
    Senator Lujan. And Ms. Gonzalez, in this area, I also fear 
that because of lack of action in the United States, in the 
United States we are going to have to be dependent on leaders 
in other parts of the world to establish the rules that we are 
hoping that will be created here with every one of us. And like 
many of my colleagues, I am hopeful that we will see a markup 
of legislation, and there are many bipartisan pieces that are 
good ideas, whether this is expansive or more narrow. But it is 
clear that action needs to be taken. And so I am hopeful and 
prayerful that that will happen.
    Now, as we close this hearing, the hearing record will 
remain open for two weeks until December 23. Any Senators that 
would like to submit questions for the record for the witnesses 
should do so by that date. We also ask that your responses be 
returned to the Committee by January 6, 2021--sorry, 2022. Now 
that concludes today's hearing. Thank you very much, everybody.
    [Whereupon, at 1:13 p.m., the hearing was adjourned.]

                            A P P E N D I X

          Prepared Statement of Imran Ahmed, Founder and CEO, 
                   Center for Countering Digital Hate
    Thank you for inviting the Center for Countering Digital Hate to 
submit written testimony for this critical hearing and thank you to the 
Committee for calling attention to this important topic.
    In our core mission to expose and disrupt the architecture of 
digital hate and misinformation and to hold to account the perpetrators 
and platforms that profit from its spread, we have extensively studied 
the use of social media by malignant actors and movements. In the last 
year, CCDH's work has studied and exposed anti-vaccine and Covid-19 
misinformation, racist abuse, violent misogyny, and climate denial on 
mainstream social media platforms. CCDH has shown that social media 
companies, namely Facebook, Twitter, YouTube, Google, and their 
subsidiaries, have systematically failed to act on hate, 
misinformation, and dangerous content circulating on their platforms.
    Algorithms, specifically the complicated and layered algorithms 
used by social media platforms, boil down to human choices made by 
individuals at companies that are driven by profit. Big Tech executives 
decide which content wins and loses in the battle for attention on 
their platforms, allowing them to shape awareness and knowledge for 
billions of people. Their algorithms are designed to drive engagement, 
eyeballs, and therefore ad revenue. Engagement is maximized by:

   Emotional triggers--to enrage is to engage, and vice versa

   Conspiracy theories, based on disinformation, that lead to a 
        warren of conspiracist rabbit holes

   Contentious misinformation that gets engagement from 
        detractors and supporters

    Malgorithm, CCDH's report on the Instagram algorithm and its 
recommendation systems, was the piece of research that earned our 
organization the most pushback from technology companies.\1\ We know 
this resistance was due to the verity of our study. In August 2020, 
Facebook, which owns Instagram, implemented a new algorithm that 
provided ``Suggested Posts'' when users reached the end of their 
feeds.\2\ This feature would use machine learning algorithms to 
identify the potential interests of a user based on their data and 
habits, and then would include ``Suggested Posts'' into users' 
feeds.\3\
---------------------------------------------------------------------------
    \1\ Malgorithm, Center for Countering Digital Hate, https://
www.counterhate.com/malgorithm
    \2\ CNN, 19 August 2020, https://edition.cnn.com/2020/08/19/tech/
instagram-suggested-posts
/index.html
    \3\ Powered by AI: Instagram's Explore recommender system, Meta, 25 
November 2019, https://ai.facebook.com/blog/powered-by-ai-instagrams-
explore-recommender-system/
---------------------------------------------------------------------------
    Using a very simple research methodology that traces user 
experience with recommendation algorithms, CCDH researchers set up new, 
anonymous accounts on Instagram and followed a series of accounts 
ranging from health and wellness to known anti-vaccine influencers. 
Researchers reviewed recommendations generated from Instagram's 
``Explore'' and ``Suggested Posts'' features. More than half of the 
recommendations in this study contained Covid-19 misinformation. Users 
following wellness accounts were algorithmically recommended anti-
vaccine misinformation and conspiracies about Covid-19. Users following 
QAnon accounts were recommended Covid-19 misinformation, misinformation 
related to the 2020 United States election, and QAnon conspiracy 
theories. Those following anti-vaccine accounts received 
recommendations for antisemitic content. In short, Facebook chose to 
launch a revenue-maximizing algorithm that promoted harmful 
misinformation during a deadly pandemic and an election season where 
conspiracies ran rampant and that culminated with a deadly attack on 
the U.S. Capitol, plotted and pre-publicized almost entirely online.
    In fact, Facebook's internal research, revealed by whistleblower 
Frances Haugen, align with our findings in Malgorithm. Facebook 
conducted a test with a very similar methodology that yielded very 
similar results. Within two days of a fresh account being set up, 
Facebook found that recommendations containing QAnon and other 
conspiracist extremism were fed through the algorithm to the new user's 
feed.\4\
---------------------------------------------------------------------------
    \4\ Insider, 25 October 2021, https://www.businessinsider.com/
facebook-papers-leak-qanon-carol-smith-research-2021-10?r=US&IR=T
---------------------------------------------------------------------------
    In 2021, CCDH released a report showing that a dozen individuals 
were responsible for 65 percent of the vaccine disinformation on 
Facebook, Instagram, Twitter and YouTube.\5\ Mark Zuckerberg was in 
fact asked about the report by members of Congress. He said he would 
look into it. A month later, our researchers found that the 
Disinformation Dozen had continued to operate with complete impunity on 
his platforms. Six months later, after the Surgeon General and the 
President of the United States all cited that report, Facebook 
executives attacked our research claiming that it was based on a 
``faulty narrative.''
---------------------------------------------------------------------------
    \5\ The Disinformation Dozen, CCDH https://www.counterhate.com/
disinformationdozen
---------------------------------------------------------------------------
    We now also know from documents disclosed by Facebook whistleblower 
Frances Haugen that on the very same day CCDH released the 
Disinformation Dozen report, Facebook had produced internal research 
that largely confirmed our findings that indeed a very small group was 
responsible for more than half of the vaccine disinformation on the 
platform. When confronted by us and later by Congress, Mark Zuckerberg, 
denied, deflected, and his company continues to delay action to stop 
these individuals.
    Meanwhile, hospitals throughout the country are packed with 
unvaccinated COVID-19 patients desperately trying to take breaths, 
telling their doctors they would have taken the vaccine but for content 
spread by the Disinformation Dozen, most of whom are still allowed to 
disseminate vaccine lies on Facebook. Many of those patients will not 
live to see another day. At the same time, Big Tech sits atop a fortune 
accrued faster than almost any in human history.
    It is now well documented by the work of whistleblowers, 
researchers, watchdog groups, and even scores of employees working 
within these technology companies that there is a fundamental problem 
with business models that prioritize profit and engagement over safety 
and security. Researchers have found algorithmic biases in race, 
gender, sexuality, and myriad other factors, particularly those 
affecting protected groups \6\ \7\ Further, researchers have found that 
misinformation receives more engagement than real news on Facebook 
products.\8\ There is a clear danger to products that sell data in 
order to maximize engagement and profit, and there is an immeasurably 
higher cost associated with the damages of misinformation and malignant 
content's ability to spread unfettered across unregulated platforms.
---------------------------------------------------------------------------
    \6\ ``FACEBOOK'S AD ALGORITHM IS A RACE AND GENDER STEREOTYPING 
MACHINE, NEW STUDY SUGGESTS'' The Intercept, 4 April 2019, https://
theintercept.com/2019/04/03/facebook-ad-algorithm-race-gender/; Imana, 
Korolava, and Heidemann, ``Auditing for Discrimination in Algorithms 
Delivering Job Ads'' University of Southern California, https://
ant.isi.edu/datasets/addelivery/Discrimination-Job-Ad-Delivery.pdf
    \7\ Algorithmic bias detection and mitigation: Best practices and 
policies to reduce consumer harms Brookings Institute Nicol Turner Lee, 
Paul Resnick, and Genie Barton May 22 2019 2019 https://
www.brookings.edu/research/algorithmic-bias-detection-and-mitigation-
best-practices-and-policies-to-reduce-consumer-harms/#footnote-6
    \8\ New Study Shows that Misinformation Sees Significantly More 
Engagement than Real News on Facebook, Social Media Today, 22 May 2019, 
https://www.socialmediatoday.com/news/new-study-shows-that-
misinformation-sees-significantly-more-engagement-than/555286/
---------------------------------------------------------------------------
Conclusion
    The ways that algorithms organize, categorize, and predict behavior 
based on data gathering and human decision making absolutely require 
oversight, transparency, and accountability. Solutions should not only 
focus on the outputs of these algorithms, what sort of content is 
amplified by algorithms and its circulation volume, but also whether 
there are proper safety mechanisms in place to pump the breaks when 
dangerous content is being spread rapidly. Slowing down the circulation 
of malignant content and prioritising and auditing safety mechanisms 
must be central to our approach to dangerous algorithms.
    In light of Facebook and tech companies' attempts to stifle 
independent research and discredit whistleblowers, Congress must commit 
to protections for whistleblowers as well as an agenda that prioritizes 
transparency while respecting privacy and data privacy. oversight 
bodies, including but not limited to the Federal Trade Commission (FTC) 
and consumer protection agencies, must be involved in the oversight of 
these publicly traded and gravely opaque companies. Transparency of 
algorithm outputs, audits of algorithm safety, and audits of platforms' 
ability to moderate hate, illegal content, and apply their terms of 
service must be part of oversight and legislative considerations.
    The revelations of the past year have proved yet again what CCDH 
has researched and exposed, that social media platforms are woefully 
incapable of self-regulation. And the time to act is now.
    The task ahead of us is to generate solutions that balance economic 
prosperity of the Nation with protection of citizens' privacy as well 
as protection from serious harm. There is no silver bullet to solve the 
problems created by Big Tech. But it is now more than ever an urgent 
task. For far too long hate and misinformation have been propagated for 
profit and caused massive and in some cases irreparable harm. Congress 
must act collectively, urgently, and with a multi-pronged approach to 
protect children, families, our society and democracy.
                                 ______
                                 
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
                                 
                                 ______
                                 
   Response to Written Questions Submitted by Hon. Kyrsten Sinema to 
                              Rose Jackson
    Spread of Misinformation. Misinformation remains rampant on social 
media and is a significant problem. For example, misinformation about 
the COVID-19 pandemic has contributed to public confusion. Also, during 
the 2020 presidential election, a number of false rumors circulated on 
social media platforms alleging wrongdoing by election administrators.

    Question 1. What steps can social media companies take to combat 
the harm of misinformation? Could algorithms work in reverse to limit 
the content containing false information a user encounters instead of 
leading them down a path to continued exposure to misleading claims?
    Answer. Large digital platforms deploy a range of policies, 
technologies, and tactics to moderate the content on their services. 
The most productive approach for addressing misinformation will vary 
platform by platform, and be affected by the particular features and 
products the platform offers (i.e., audio versus video or text; 
ephemeral posts or content), the user base the platform serves, the 
languages it operates in, and other related factors. Content moderation 
approaches therefore are best targeted to those specific contexts. 
Algorithms can be (and in many cases are already) used in several ways. 
For example, platforms use automation and algorithms to review posts, 
add labels and links to sources, and solicit user feedback on content 
offered to consumers. Additionally, companies sometimes de-rank or 
simply remove targeting from posts containing mis- or disinformation 
and have experimented with adding friction to slow down a piece of 
content as it goes viral. These approaches have had some success in 
reducing or mitigating mis- and disinformation, and where they are 
effective, they should be scaled and applied in new contexts. However, 
technological, or automated approaches to content moderation will never 
be perfect. In some cases, these approaches fail to remove clearly 
illegal (or counter to terms of service) content, and in others they 
result in the over-removal of content, often including critical, 
satirical, and legitimate expression.
    To combat the harm of misinformation, large digital platforms can: 
establish clear terms of service and trust and safety policies; 
communicate those policies (and their enforcement) in all languages in 
which the platform is available; make clear when algorithms, filtering, 
or other technologies are used to alter the display of content to a 
user, and enable users to control those settings; employ techniques 
(including those referenced above) that are shown to decrease the 
spread of mis- and disinformation; and establish policies to target the 
coordinated abuse of platform features. Platforms can also combat these 
harms by sharing information and partnering with civil society 
organizations and the communities most affected, to co-create 
solutions, and build societal resilience.

    Question 2. What steps could Congress take to address this issue 
while recognizing the importance of freedom of expression?
    Answer. Addressing the challenge of mis- and disinformation 
requires a ``multi-stakeholder approach.'' The issue is systemic, with 
narratives and trends spreading across an information ecosystem 
comprised of numerous platforms and countless communities. That means 
it cannot be solely addressed by a single platform, or sector. Congress 
would be wise to design its legislation and direct its oversight to 
incentivize industry, empower civil society, center the American public 
and users, and strengthen the government's ability to set and enforce 
the rules. Substantively, Congress should prioritize protecting 
privacy, enabling transparency and accountability, and ensuring the 
U.S. government is appropriately staffed and resourced for our digital 
age. Doing so would provide a meaningful foundation for many of the 
specific laws and rulemaking under consideration to address the issues 
discussed during the hearing.
    While a particular piece of content may be evidence of 
disinformation, addressing the systemic drivers of the problem is a 
more fruitful approach. That means focusing on product design, business 
models, and other platform incentives that either result in the spread 
of certain content or incentivize and reward behavior online that 
contributes to the problem. As I mentioned during my testimony, the 
conversations we are having about the Internet and the digital world 
are interrelated. The gaps in privacy or data protections in the United 
States contribute to the incentive to create and collect more--and more 
sensitive--data to use for everything from better targeted advertising 
to the development of new artificial intelligence models and tools. 
Passing comprehensive privacy or data protection legislation would set 
standards for how that data could be collected and used, potentially 
changing the incentives for platforms on everything from design to 
policy.
    Second, Congress can set standards for the basic information 
platforms are required to disclose, whether clear terms of service and 
policies, or insight into why users are viewing certain content. 
Proposals to create a new bureau in the FTC focused on data and 
technology would be a good first step toward setting such standards and 
tracking how major platforms operate. A better resourced FTC could also 
facilitate the growing calls for better data and information on 
platforms, to inform government agencies and Congress, as well as the 
community of academic and independent researchers who work with civil 
society to shed light on the impact of these pervasive technologies and 
platforms in society. Organizations like my own in the open-source 
research community have been instrumental in uncovering and documenting 
how individuals and groups manipulate platforms and populations. And we 
use that knowledge to inform the public and advise companies, 
governments, and civil society on appropriate responses. But our 
knowledge is limited by what data is available to us, and we know there 
is much about the information ecosystem we miss. Addressing 
disinformation requires driving down the information asymmetry between 
platforms and everyone else.
    But the task of rulemaking and enforcement cannot sit with the FTC 
alone. The Federal Communications Commission (FCC), Consumer Financial 
Protection Bureau (CFPB), and others have significant roles to play. 
Congress should bolster the resources and capacity of these entities 
alongside the FTC, and work to clarify mandate and points of 
coordination between them to ensure companies are spending time and 
resources on the requests and changes that are most impactful. Finally, 
Congress should seek to ensure existing law is applied to a digital 
context. In some cases, this will mean prompting agencies to update 
their approach to existing work and authorities. In other cases, 
Congress may need to consider new laws. Congress should consider 
adopting proposals such as those that would require the same funding 
disclosures for digital advertising as is required on television and 
other legacy media, or those requiring platforms provide public 
advertising libraries. Congress could also apply existing law in new 
places; Bills like the Malinowski-Eshoo Protecting Americans from 
Dangerous Algorithms Act, for example, apply limited liability 
exceptions on the basis of narrowly defined existing criminal statutes.

    Foreign Language. Media reports suggest that many social media 
companies have not invested resources in addressing misinformation 
being spread in languages other than English. This poses risks to non-
English speaking communities in the United States, including Spanish 
speaking residents in Arizona, as well as residents of foreign nations 
who utilize these tech platforms that have headquarters in the United 
States.

    Question 3. What responsibility do tech companies have to prevent 
their platforms from being used to spread misinformation in languages 
other than English, which can result in political violence in the 
United States or in foreign nations?
    Answer. It is important to note that while many large digital 
platforms have voluntarily taken steps to address mis- and 
disinformation they do not have a legal responsibility to do so in any 
language. However, to responsibly operate in any market, companies 
should have staff capable of communicating in and understanding the 
core languages within that market. They should publicize their key 
company and policy documents (such as terms of service and community 
guidelines) in those languages, and ensure any content moderation 
decisions, reporting features, or other core platform components are 
likewise made available in each of those languages. Companies should 
publish (in a centralized and easy-to-read format) the languages in 
which particular services are provided, and what products and services 
are made available in each region and language. While these may seem 
like small, or even obvious asks, no company currently provides most of 
the above consistently, with a few exceptions.
    Another important element is the ways in which platforms 
collaborate with civil society and the public. Mis- and disinformation 
spreads across multiple platforms and requires whole-of-society 
responses. My organization, the Digital Forensic Research Lab, often 
uncovers the manipulation of information environments in foreign 
countries and languages. We work collaboratively with platforms to flag 
these issues and make our research public to inform civil society 
responses and government action. Whether platforms make data and other 
information available to organizations like my own, particularly in 
countries outside the United States, has a direct impact on our 
understanding of the problem and ability to respond. Further, whether 
platforms respond promptly when communities do highlight growing 
unrest, harassment, or violence can make the difference between life 
and death. The lack of language appropriate staff can slow responses 
and hamper essential contextual understanding.

    Research Access. We heard during the hearing about the potential 
value in permitting researchers, particularly those in an academic 
setting, to examine user data from tech companies or from probing 
social media sites by setting up accounts to examine the content those 
accounts receive exposure to.

    Question 4. What types of data do you believe Congress should 
require tech companies to disclose to researchers so they can examine 
the effects of algorithms on every day Americans?
    Answer. There is no question that the asymmetry of knowledge 
between large digital platforms and everyone else is a barrier to 
addressing online harms. That is why so many legislators, researchers, 
and other experts have increasingly turned toward transparency and data 
access as a focal point for action. What each of these groups means by 
``transparency'' however is less clear, complicating calls for 
companies to be required to provide certain kinds of information to 
researchers, regulators, or the public. Enabling societies to better 
understand the impact of platform design and policies is important, but 
transparency is not an end in and of itself. And if focused on 
transparency to promote accountability and prevent harm, it is 
important to build flexibility into what we prioritize as the most 
essential information. Rather than focus on Congress delineating the 
specific forms of data or reports platforms be required to provide, it 
can authorize and fund regulatory and oversight bodies to set standards 
and requirements that can adapt as technology changes and our 
understanding of related challenges grows.
    It is also important to acknowledge that algorithms are used across 
several industries and sectors for vastly different purposes. 
Understanding the data used to develop machine learning models, the use 
and context of those models, and the societal impact they have will 
vary whether looking at those used to predict the success of students, 
those used to determine whether someone is granted a loan, or those 
used by the justice system to inform sentencing. That is why I focused 
in my testimony on the need to appropriately resource agencies and 
offices across the U.S. government to play their role in enforcing 
existing law in a digital context and identify where new law is 
required.
    In reference to large digital platforms, Congress could consider 
the approach taken in the European Union's Digital Services Act (DSA), 
which includes a mix of requirements ranging from the provision of 
specific data, to standardized transparency reporting, or the conduct 
of risk assessments. As these technologies advance, so too does the 
field studying them. Congress would be wise to ensure new insights and 
best practices can inform and adapt how oversight is conducted.

    Question 5. Broadly speaking, how would you define ``researchers'' 
to ensure that only legitimate researchers can obtain information?
    Answer. In establishing information sharing requirements between 
digital platforms, regulators, researchers, and the public, it's 
important we be specific about the kinds of information and data at 
play, and the tradeoffs between transparency and security or privacy. 
There is a great deal of platform information that would not have 
significant privacy implications, and we should be careful not to limit 
all access to such information to a small subset of individuals. That 
said, for more sensitive or expansive forms of information and data the 
following qualifications for researchers should be considered. To be 
``qualified'' a researcher should be affiliated with an organization or 
institution that: is a registered non-profit; has proven expertise in 
studying information environments; publicizes their research 
methodology; and makes public the results of their research. 
Researchers should be required to abide by established terms of use and 
be barred from access should they violate those agreements. There are 
strong proposals for the creation of a new office within the FTC to 
delineate these definitions more clearly, to create terms of use, and 
identify which data and information should be provided by covered 
platforms. Doing so would enable regulators to flexibly revise the 
approach over time to find the right balance between access and 
privacy, while holding researchers accountable to these standards.

    Transparency and Intellectual Property. I want to ensure that the 
Internet remains a place of innovation. In addition to academic 
researchers, some have proposed that the general public have additional 
information about the algorithms tech companies employ to run their 
platforms.

    Question 6. If Congress pursues measures to require tech companies 
to disclose more information about algorithms, how can we ensure that 
those companies' intellectual property remains protected and does not 
fall into the hands of competitors or nefarious actors?
    Answer. As referenced in question four, algorithms are used across 
a range of industries for a variety of purposes. So, the answer 
regarding large digital platforms may not apply to algorithms used for 
law enforcement, college admissions, or other purposes. When an 
algorithm is being used to make decisions on behalf of the government, 
or in ways that fundamentally determine a citizens' access to 
education, finance, and essential services, understanding why those 
decisions are being recommended is a base necessity.
    When it comes to large digital platforms, as Dr. Dean Eckles 
testified during the hearing, many of the algorithms being used are not 
particularly sensitive or proprietary, and disclosure of the level the 
committee is discussing would be unlikely to create significant 
intellectual property risks. That said, balancing interests such as 
these is why Congress would be wise to authorize and resource 
regulatory bodies to set the rules for the type of information required 
from companies, and the level of access provided to the public, to 
researchers, and to the government more broadly.
    It is also worth noting, that some transparency steps could be 
taken without sharing large data sets or divulging proprietary 
information. For example, some experts recommend platforms inform a 
user when they are seeing a post, certain content, or advertising 
because of a particular ranking or targeting algorithm. Others suggest 
that platforms allow users to opt-out of such targeting. These steps 
would provide users with more context and power in choosing and 
understanding their digital environments, which can help build 
resilience against bad actors manipulating platforms and the broader 
information ecosystem as well.
                                 ______
                                 
  Response to Written Questions Submitted by Hon. Raphael Warnock to 
                              Rose Jackson
    In recent years, we have seen several examples of targeted 
advertising algorithms that can enable discrimination and may violate 
Federal law. For example, in 2019, the Department of Housing and Urban 
Development sued Facebook over housing advertisements that could be 
targeted on the basis of protected classes such as race, religion, 
familiar status or others. Facebook has also been sued for targeting 
tools and algorithms that may have shown job advertisements to certain 
populations based on age.

    Question 1. Do you believe that existing Federal anti-
discrimination laws adequately address the ways that algorithms can 
enable discrimination? If not, how should Congress and Federal agencies 
develop new regulations to enable greater oversight?
    Answer. The 1964 Civil Rights Act, the 1968 Fair Housing Act, and 
the 1974 Equal Credit Opportunity Act set clear precedents against 
discrimination on the basis of protected classes. The same rights 
afforded to citizens offline should also be enforced online, especially 
in instances where citizens' civil rights are threatened. The first 
step is to ensure that existing laws are applied to the online world. 
More can be done to provide agencies responsible for overseeing and 
ensuring these protections with the resources and capacity required to 
update their operations for an online context. Congress could direct 
the administration to review those agencies and offices responsible for 
enforcing such protections to determine how, or whether, mandates have 
been sufficiently applied to digital tools and the Internet. 
Subsequently, Congress could consider additional legislation clarifying 
the application of these laws in a digital context, or where required, 
develop legislation explicitly establishing new requirements pertaining 
to the role of curation, algorithmic amplification, direct targeting, 
or automated decisionmaking in such discrimination.
    The risk of algorithmic discrimination extends beyond social media 
and the Internet. Algorithms are now used to aid in decisionmaking on 
everything from hiring, to loans, to college admittance. Rather than 
seek one standard for all algorithms, Congress should consider 
appropriate regulations and standards applied to various industries. As 
new technology continues to mainstream into nearly every aspect of 
life, it will become harder to regulate a technology per-say, rather 
than its use and application industry by industry.

    Question 2. Do Federal agencies have adequate authorities and 
resources to investigate and, if necessary, prosecute corporations that 
enable discrimination through their algorithms? If not, what additional 
authorities or resources are necessary?
    Answer. Ultimately, Federal agencies do not have adequate 
authorities and resources to investigate and, if necessary, prosecute 
corporations that enable discrimination through their algorithms. In my 
testimony, I called on congress to create and resource a new bureau at 
the FTC to address online privacy, data security, and other online 
abuses, including discrimination, voter suppression, and civil rights 
violations. But the FTC has a narrow mandate. Congress must provide 
resources and support to all agencies to improve their understanding of 
and engagement on issues related to the application of technology. It 
can do so through further support for and investment in the U.S. 
Digital Service (USDS) and other related U.S. Government service-
oriented tech agencies and offices. I also recommended that congress 
pass legislation to foster greater transparency and accountability 
around online harms, and address information asymmetries between 
companies and everyone else, to understand the full extent of these 
harms, and inform solutions.
                                 ______
                                 
   Response to Written Questions Submitted by Hon. Kyrsten Sinema to 
                          Jessica J. Gonzalez
    Spread of Misinformation. Misinformation remains rampant on social 
media and is a significant problem. For example, misinformation about 
the COVID-19 pandemic has contributed to public confusion. Also, during 
the 2020 presidential election, a number of false rumors circulated on 
social media platforms alleging wrongdoing by election administrators.

    Question 1. What steps can social media companies take to combat 
the harm of misinformation? Could algorithms work in reverse to limit 
the content containing false information a user encounters instead of 
leading them down a path to continued exposure to misleading claims?
    Answer. Social media companies can and should do more to combat 
misinformation. These platforms are pushing content to users through 
algorithms that, in most cases, are trained to drive the highest 
engagement. As Frances Haugen shared with this committee, inflammatory 
content and conspiracy theories tend to fit this bill.
    Platforms have failed outright at fulfilling even the most basic 
requests for transparency about how their systems work, what they know 
about us and how they use our data to target us with content. Their 
data collection is pervasive, and may be invasive and extractive. Only 
platforms have the data about who sees what information and how much 
they engage with it. That gross imbalance is even more in focus now 
because platforms like Facebook have been shown to be aware of the 
negative impacts of their design and have failed to adapt it to prevent 
and mitigate harms, such as the misinformation about COVID and the 2020 
election that you mention.
    We each are seeing vastly different things online--predicated on 
online platforms' data practices, many of which are abusive and merit 
further scrutiny from an expert agency such as the Federal Trade 
Commission. The divide between us all in what we see (and when we are 
largely unaware of what others see) means that we simply don't have the 
ability to fully ascertain who is being persuaded by what.
    No one silver bullet will solve this problem and we need both 
corporate responsibility and a strong government response to targeted 
and prolific lies that threaten our health and safety and even our 
democracy. Free Press has a series of recommendations to online 
platforms designed to disrupt online bigotry and conspiracies and to 
shed greater light on how these systems operate to help us identify 
further areas for improvement:

   Free Press has worked with Change the Terms, a coalition of 
        more than five dozen racial justice and digital and civil 
        rights groups, to release a set of model corporate policies to 
        disrupt online hate. Most of these measures would significantly 
        reduce misinformation as well, including modifications to the 
        companies' terms of service, appeal policies, enforcement 
        mechanisms, content moderation training procedures, regular 
        audits of the algorithms, and far more transparency to 
        researchers and the public.\1\
---------------------------------------------------------------------------
    \1\ Change the Terms, ``Recommended Internet Company Corporate 
Policies and Terms of Service to Reduce Hateful Activities,'' https://
www.changetheterms.org/terms.

   Companies can and should institute some of the mitigation 
        measures put in place ahead of the 2020 election to slow the 
        spread of lies and bigoted propaganda and promote accurate 
        information.\2\
---------------------------------------------------------------------------
    \2\ See, e.g., Kevin Roose, Facebook reverses postelection 
algorithm changes that boosted news from authoritative sources., N.Y. 
TIMES, Dec. 16, 2020, at https://www.nytimes.com/2020/12/16/technology/
facebook-reverses-postelection-algorithm-changes-that-boosted-news-
from-authoritative-sources.html.

   Whatever rules social media companies adopt, they should 
        apply to all users. Powerful people must be held to the same 
        standards as the rest of us. There should not be a special set 
        of more lenient rules for the rich, powerful and famous that 
        allow them to proliferate conspiracy theories and bigotry with 
---------------------------------------------------------------------------
        impunity.

   As the #YaBastaFacebook campaign and community organizations 
        have been pointing out for years now, social media companies 
        have done an especially poor job of moderating content in 
        languages other than English. Frances Haugen revealed internal 
        Facebook documents that demonstrate that Facebook has seriously 
        underinvested in keeping its platforms safe across languages. 
        Facebook is not alone. All social media companies must commit 
        to and invest in keeping their platforms safe across languages.

    We need multi-layered solutions to these problems, and while self-
regulation is one necessary intervention, many social media firms have 
demonstrated that they are unwilling to meaningfully self-regulate. 
Government intervention is necessary.

    Question 2. What steps could Congress take to address this issue 
while recognizing the importance of freedom of expression?
    Answer. Congress must legislate to rein in tech company abuses 
while protecting free expression. Pages seven through seventeen of my 
written testimony for this hearing outline Free Press Action's 
recommended legislative framework.\3\ In short, Free Press Action 
recommends that Congress adopt digital privacy and civil rights 
legislation that:
---------------------------------------------------------------------------
    \3\ Disrupting Dangerous Algorithms: Addressing the Harms of 
Persuasive Technology: Hearing Before the Subcomm. on Commc'ns, Media, 
& Broadband of the S. Comm. on Sci., Commerce, & Transp., 117th Cong. 
(2021) (statement of Jessica J. Gonzalez, Co-CEO, Free Press Action), 
https://www.commerce.senate.gov/services/files/FCB71657-4A1C-4796-BCED-
55BBF418246A.

---------------------------------------------------------------------------
   Limits tech's collection and use of our personal data.

   Establishes people's rights to control their own data.

   Enhances data transparency.

   Prevents discrimination by algorithms.

   Increases platforms' transparency about known impacts of 
        their business models.

   Protects whistleblowers and external researchers.

   Expands FTC oversight.

   Encourages collaboration across agencies that hold 
        specialized expertise.

   And sets a Federal floor for consumer protection, not a 
        ceiling.

    This approach has been endorsed by over three dozen grassroots 
organizations in the Disinfo Defense League, a distributed national 
network of organizers, researchers and disinformation experts 
disrupting online racialized disinformation infrastructure and 
campaigns that deliberately target Black, Latinx, Asian American/
Pacific Islander and other communities of color.
    Fortunately, many of these elements are included in Senator 
Markey's Algorithmic Justice and Online Transparency Act, which 
prohibits algorithms that discriminate based on protected 
characteristics, and establishes safety and effectiveness standards;\4\ 
and in the Consumer Online Privacy Rights Act, sponsored by Senators 
Cantwell, Schatz, Klobuchar and Markey, which penalizes platforms that 
abuse personal data, allows people to see the information companies 
collect on them, and preserves a Federal private right of action.\5\ 
Free Press Action endorses both bills.
---------------------------------------------------------------------------
    \4\ Algorithmic Justice and Online Transparency Act, S. 1896, 117th 
Cong. (2021).
    \5\ Consumer Online Privacy Rights Act, S. 3195, 117th Cong. 
(2021).
---------------------------------------------------------------------------
    In addition to privacy and civil-rights safeguards, to combat 
disinformation we must invest in a thriving media system to support a 
21st-century democracy. Free Press Action and the Disinfo Defense 
League have urged Congress to pass legislation to tax digital 
advertising and direct those monies to support high-quality 
noncommercial and local journalism, including journalism by and serving 
people of color, non-English speakers, and other minority groups.

    Foreign Language. Media reports suggest that many social media 
companies have not invested resources in addressing misinformation 
being spread in languages other than English. This poses risks to non-
English speaking communities in the United States, including Spanish 
speaking residents in Arizona, as well as residents of foreign nations 
who utilize these tech platforms that have headquarters in the United 
States.

    Question 3. What responsibility do tech companies have to prevent 
their platforms from being used to spread misinformation in languages 
other than English, which can result in political violence in the 
United States or in foreign nations?
    Answer. Tech companies have a responsibility to keep people safe 
and enforce their terms of service in all languages in which they 
operate, and they are failing miserably to do so. This puts non-English 
speakers at risk as they navigate targeted disinformation campaigns 
about COVID and vaccines.
    Free Press noticed a significant gap in the moderation efforts of 
social media companies in non-English languages in the lead up to the 
2020 presidential election. As platforms took enforcement actions to 
address militarized social movements in English (QAnon, calls for 
violence related to Kenosha, etc.) similar Spanish-language posts 
remained widely available.
    Frances Haugen's documents revealed just how extensive the 
asymmetries are and how systems to proactively keep users safe are not 
replicated across languages.
    Despite Haugen's shocking revelations and repeated requests from 
civil rights leaders and elected officials, we still don't know how 
many content moderators speak Spanish, where they are located, or how 
they are trained and treated. Recent reporting suggests, for instance, 
that working conditions for Spanish-language content moderators are 
unequal to those of English-language moderators.\6\
---------------------------------------------------------------------------
    \6\ Sarah Emerson, Facebook's Spanish-Language Moderators Are 
Calling Their Work A ``Nightmare,'' BUZZFEED NEWS, Jan. 13, 2022 at 
https://www.buzzfeednews.com/article/sara
hemerson/facebooks-spanish-language-moderators-said-theyre-treated.
---------------------------------------------------------------------------
    Basic transparency could help identify why these gaps exist and 
pave the way for solutions, but so far the companies have resisted 
calls for more transparency.

    Research Access. We heard during the hearing about the potential 
value in permitting researchers, particularly those in an academic 
setting, to examine user data from tech companies or from probing 
social media sites by setting up accounts to examine the content those 
accounts receive exposure to.

    Question 4. What types of data do you believe Congress should 
require tech companies to disclose to researchers so they can examine 
the effects of algorithms on every day Americans?
    Answer. Researchers need access to information about platform 
moderation data, such as the number of posts, accounts, groups, etc. 
that are removed for violating platforms' corporate policies. Tech 
companies should also provide researchers with content that they are 
removing from their platforms. When platforms conduct internal research 
that demonstrates their services are causing widespread harm, they 
should have to disclose that immediately to researchers and the broader 
public. For instance, it should not take a Wall Street Journal expose 
two years after the fact to reveal that Facebook's own recommendation 
system is responsible for 64 percent of people who find extremist 
groups on its platform. Nor should it require a whistleblower to reveal 
that Instagram knew it was harming teen girls.

    Question 5. Broadly speaking, how would you define ``researchers'' 
to ensure that only legitimate researchers can obtain information?
    Answer. Some of the most innovative research about how online 
disinformation and extremism are playing out in our communities is 
coming not from elite educational institutions but rather from the very 
communities that are most impacted by these issues. I'd recommend 
liberally defining ``researchers'' to ensure that academics, 
journalists, freelancers, non-profit organizations and community 
leaders alike can access data and information. Instead of focusing on 
one's affiliation, Free Press Action recommends prodding one's 
willingness to comply with protective orders or other guidelines. A 
researcher should be deemed ``legitimate'' as long as they are indeed 
engaged in research and willing to follow the rules, not based on 
credentials or the content and aims of one's research interests.

    Transparency and Intellectual Property. I want to ensure that the 
Internet remains a place of innovation. In addition to academic 
researchers, some have proposed that the general public have additional 
information about the algorithms tech companies employ to run their 
platforms.

    Question 6. If Congress pursues measures to require tech companies 
to disclose more information about algorithms, how can we ensure that 
those companies' intellectual property remains protected and does not 
fall into the hands of competitors or nefarious actors?
    Answer. Congress can protect tech company disclosures by barring 
any party that is engaged in ``Competitive Decision-Making'' from 
accessing confidential data. This would exclude direct employees of 
competing firms from accessing the information. ``Competitive Decision-
Making'' in this context would mean a person's activities, association, 
or relationship with any of their clients involving advice about or 
participation in the relevant business decisions or the analysis 
underlying the relevant business decisions of the client in competition 
with or in a business relationship with the party submitting the 
disclosures.
                                 ______
                                 
  Response to Written Questions Submitted by Hon. Raphael Warnock to 
                          Jessica J. Gonzalez
    In recent years, we have seen several examples of targeted 
advertising algorithms that can enable discrimination and may violate 
Federal law. For example, in 2019, the Department of Housing and Urban 
Development sued Facebook over housing advertisements that could be 
targeted on the basis of protected classes such as race, religion, 
familiar status or others. Facebook has also been sued for targeting 
tools and algorithms that may have shown job advertisements to certain 
populations based on age.

    Question 1. Do you believe that existing Federal anti-
discrimination laws adequately address the ways that algorithms can 
enable discrimination? If not, how should Congress and Federal agencies 
develop new regulations to enable greater oversight?
    Answer. Existing Federal anti-discrimination laws are a critically 
important keystone to fighting algorithmic discrimination, but Congress 
must do more. Opaque algorithms obscure algorithmic decision-making and 
obstruct our ability to redress their discriminatory impacts.
    Congress should legislate to further protect civil rights in the 
digital age. Pages seven through seventeen of my written testimony for 
this hearing outline Free Press Action's recommended legislative 
framework.\7\ In short, Free Press Action recommends that Congress 
adopt digital privacy and civil rights legislation that:
---------------------------------------------------------------------------
    \7\ Disrupting Dangerous Algorithms: Addressing the Harms of 
Persuasive Technology: Hearing Before the Subcomm. on Commc'ns, Media, 
& Broadband of the S. Comm. on Sci., Commerce, & Transp., 117th Cong. 
(2021) (statement of Jessica J. Gonzalez, Co-CEO, Free Press Action), 
https://www.commerce.senate.gov/services/files/FCB71657-4A1C-4796-BCED-
55BBF418246A.

---------------------------------------------------------------------------
   Limits tech's collection and use of our personal data.

   Establishes people's rights to control their own data.

   Enhances data transparency.

   Prevents discrimination by algorithms.

   Increases platforms' transparency about known impacts of 
        their business models.

   Protects whistleblowers and external researchers.

   Expands FTC oversight.

   Encourages collaboration across agencies that hold 
        specialized expertise.

   And sets a Federal floor for consumer protection, not a 
        ceiling.

    This approach has been endorsed by over three dozen grassroots 
organizations in the Disinfo Defense League, a distributed national 
network of organizers, researchers and disinformation experts 
disrupting online racialized disinformation infrastructure and 
campaigns that deliberately target Black, Latinx, Asian American/
Pacific Islander and other communities of color.
    Fortunately, many of these elements are included in Senator 
Markey's Algorithmic Justice and Online Transparency Act, which 
prohibits algorithms that discriminate based on protected 
characteristics, and establishes safety and effectiveness standards;\8\ 
and in the Consumer Online Privacy Rights Act, sponsored by Senators 
Cantwell, Schatz, Klobuchar and Markey, which penalizes platforms that 
abuse personal data, allows people to see the information companies 
collect on them, and preserves a Federal private right of action.\9\ 
Free Press Action endorses both bills.
---------------------------------------------------------------------------
    \8\ Algorithmic Justice and Online Transparency Act, S. 1896, 117th 
Cong. (2021).
    \9\ Consumer Online Privacy Rights Act, S. 3195, 117th Cong. 
(2021).
---------------------------------------------------------------------------
    In addition to privacy and civil-rights safeguards, to combat 
racialized disinformation we must invest in a thriving media system to 
support a 21st-century democracy. Free Press Action and the Disinfo 
Defense League have proposed that Congress pass legislation to tax 
digital advertising and direct those monies to support high-quality 
noncommercial and local journalism, including journalism by and serving 
people of color, non-English speakers, and other minority groups.

    Question 2. Do Federal agencies have adequate authorities and 
resources to investigate and, if necessary, prosecute corporations that 
enable discrimination through their algorithms? If not, what additional 
authorities or resources are necessary?
    Answer. Federal agencies have existing authorities to investigate 
and enforce civil rights laws and rules against corporations engaged in 
discrimination, but would benefit from additional resources and 
rulemaking authority.
    Existing civil rights laws remain woefully under-enforced with 
regards to algorithmic decision-making and data use. For instance, the 
Equal Employment Opportunity Commission has authority to regulate 
hiring algorithms and facial-recognition software. The Department of 
Housing and Urban Development (HUD) has explored disparate-impact 
claims pertaining to housing discrimination. Machine-learning bias in 
credit denial can have lasting impacts on a family's wealth, and the 
Equal Credit Opportunity Act (ECOA) is meant to guard against that kind 
of discrimination.
    The Federal Trade Commission (FTC) should enforce any violations of 
the Equal Credit Opportunity Act, which prohibits credit discrimination 
based on several protected characteristics; and the Fair Credit 
Reporting Act, which requires people to be notified when adverse 
actions about them are made, and allows them to dispute any inaccurate 
information. Both of these existing laws offer critical protections for 
consumers in the credit context--which has many implications across 
other areas like housing, employment, and insurance.
    To the extent that unfair practices and discriminatory practices 
fall outside of the scope of these two acts, the FTC can regulate 
unfair and discriminatory algorithms under its existing unfairness 
doctrine authority, Section 5 of the FTC Act.\10\ Congress should 
ensure a more expedited regulatory process by granting the FTC 
additional rulemaking authority under the Administrative Procedures 
Act.
---------------------------------------------------------------------------
    \10\ Federal Trade Commission Act Sec. 5, 15 U.S.C. Sec. 45 (2021).
---------------------------------------------------------------------------
                                 ______
                                 
     Response to Written Questions Submitted by Hon. John Thune to 
                              James Poulos
    Question 1. I was intrigued by your prepared testimony, where you 
talk about addressing digital harms by passing a ``Second Amendment for 
Compute.'' Can you elaborate further on what you mean by that?
    Answer. The idea of the Second Amendment for Compute is to legally 
identify and enshrine the basic American rights to build and use 
digital computational technologies. Notably, these rights are already 
implicitly included in the First and Second Amendments, because the two 
uses of the technologies are as communications tools and as offensive 
and defensive weapons. The deepest roots of American culture, civics, 
and civilization are found in the conviction that, as a matter of 
justice and sound political theology, the rights of human persons who 
have come of age to communicate and bear arms are natural and God-
given, things no just government can infringe, abridge, abrogate, or 
deny.
    In this sense we and our Constitution already have all the 
justification and legal authority needed to protect and defend American 
citizens' rights to keep and bear digital tech. Now, in the early 
debates of the founding era, the Antifederalists warned that while a 
Federal Constitution of enumerated powers blocked tyranny and unjust 
rule, enumerated rights did the opposite. And there is something 
powerfully true to the insight that a people have likely already lost 
their shared faith or conviction in what they must make explicit.
    Nevertheless, one glance today shows that the brazen and systematic 
violation of the rights to keep and bear digital technologies that are 
implicit in the First and Second Amendments is foundational to our 
emergent political and economic order. Our unelected rulers in public 
and private life work hand in hand every moment of every day to preempt 
and defeat those rights. Control of digital life is consolidated in the 
hands of the military-intelligence complex and the tech oligarchy, upon 
whose economic performance America's administrative class and financial 
system now utterly depends. Under their control, Americans are being 
coercively conditioned to accept the wholesale expropriation of their 
private lives as well as the ``public square'' into the cyberspaces 
operated, owned, and policed by the major tech corporations, which are 
properly understood as the core strategic governance organs of the 
administrative state, including the military-intelligence complex.
    What is unfolding is--by design--the founding of a new, digital 
regime, one hostile and fatal to our distinctively American way of life 
and our constitutionally guaranteed form of government, and one resting 
on a political theology that repudiates and destroys our most venerable 
Western principles of natural right and Man's relation to his Creator. 
As the turmoil of the Trump years made inescapably plain, the operators 
of this regime are increasingly naked in their use of media of all 
kinds, through the digital technologies that now rule them all, as 
weapons against Americans who dare resist or even object to their 
extension and consolidation of digital power and authority.
    This means that it is now too late legally, and technology is 
advancing too quickly, to any longer leave merely implicit the natural 
and God-given rights that Americans already possess with regard to 
digital technologies. To save those rights, and our civilization and 
form of government with them, we must enumerate Americans' digital 
rights--come what may. Congress must be undaunted by administrative 
state objections that Americans are too stupid, crude, hateful, sinful, 
inexpert, or simply dangerous to be allowed such rights and their free 
exercise. America's global adversaries and enemies will not seize 
control if Americans begin that free exercise. Quite the contrary: our 
technological rulers today are railroading us into a post-American 
order where elites who see America (correctly) as the main obstacle to 
founding a new globalistic cyborg order. Nor will the free exercise of 
natural and God-given digital rights crash our financial system, as the 
SEC and other agencies and officials insist on arguing. Our financial 
and regulatory leaders across the public and private sector have 
already repeatedly crashed our financial system. Grievously and 
irreparably harming generations of Americans, these ostensible elites 
have suffered virtually zero accountability or punishment, in fact only 
growing more powerful, more wealthy, more impudent, and more divorced 
from our American identity, our American norms, and our American 
principles. They simply cannot be trusted to tell us the truth about 
the likely effects of Americans' free exercise of digital rights--aside 
from their implicit admission that their illegitimate stranglehold on 
our fates and fortunes cannot survive a restoration of American 
constitutionalism in digital life.
    That brings us to the explicit terms of legal protection envisioned 
by the ``Second Amendment for Compute.'' They begin simply with the 
right of Americans to buy, sell, and build high-powered GPUs--a right 
already being stealthily infringed by ``energy regulations'' in several 
states including California itself. The entire canard that powerful 
computation is too ``eco-unfriendly'' to be allowed to America's 
digital serfs--a charge laid aggressively against bitcoin and other 
cryptocurrencies--is a fallacy Congress must recognize as such. 
Americans' use of high-powered GPUs to freely build digital tools and 
cultural and economic institutions expends far less ``energy'' than, 
say, the petrodollar and its form of economic life, or than the 
military-intelligence complex; still more deeply, since John Locke and 
well before it has been a fundament of Western life to understand that 
the just building and maintenance of life and flourishing in our human 
and natural spacetime requires as its price the expenditure of energy, 
in the form the sweat of our brows and the ache of our bodies, of the 
productive calculations of our minds, and of our creation and use of 
the tools and weapons we need to achieve both our most basic and 
fundamental ends and our higher forms of creativity, genius, worship, 
and wealth.
    By extension therefore it is equally clear that the right of 
Americans to buy, sell, mine, and mint bitcoin and other digital coins 
must not be infringed, neutralized, or destroyed. It is essential that 
lawmakers realize bitcoin, to take the most important example, is not 
simply a new currency among many, a new asset class among many, or a 
new fad among many. It is a computational protocol that supervenes upon 
or obsolesces a host of previous forms of digital architecture, because 
of its tremendous power and adaptability. It is, in this sense, a 
computer that can and likely will very soon grow large enough in its 
development to supervene upon the entire world, ``ruling'' it in a way 
no one person or group of people, no matter how brilliant, willful, 
expert, or ambitious, can any longer plausibly imagine to rule the 
world. Without the explicitly protected right to use this protocol to 
build new but quintessentially American institutions of culture, 
economics, finance, and religion on a securely digital footing, 
Americans will lose more and more of their initiative, ingenuity, and 
identity until they surrender before their rulers' efforts to merge 
them into their technologies and become cyborg serfs: a permanent 
underclass of passive, sullen, demoralized, self-loathing, mentally and 
physically unwell posthumans bearing no recognizable relation either to 
their American forebears or indeed to any human who has ever heretofore 
lived.
    Hammering out the ins and outs of the ``Second Amendment for 
Compute'' will not be swift or easy work, but it can be done, as it 
must be, with sufficient speed and efficacy to save Congress, and 
America, from digital obsolescence. But to do so, lawmakers must find 
and wield the courage necessary to resist what will be extraordinary 
concerted pressure from the rising post-constitutional regime ruled by 
the triumvirate of the administrative state, the military-intelligence 
complex, and the tech oligarchy. It is very likely that 2022 will rank 
among the very last possible opportunities to begin this fateful 
struggle in earnest at the Federal level.
    That said, the Federal level is not, as Congress knows well, 
everything. Perhaps one of the most important things Federal lawmakers 
can do to ensure Americans' natural and God-given digital rights are 
not infringed is to pass legislation that gives state governments as 
wide as possible a berth to pass their own digital rights legislation 
and to erect their own legal and regulatory regimes around digital 
rights, free from fear of Federal retaliation, punishment, or budgetary 
blackmail. The ``laboratory of democracy'' will quickly show exactly 
how powerful and precious digital rights are in our digital age as 
Americans gravitate toward and rally around states with the most 
courageous and clear-sighted protections of digital rights.

    Question 2. The ``metaverse'' appears to be what we can expect the 
Internet to be in the near term future. As you know, Mark Zuckerberg 
changed the name of his company to Meta in October, and said that the 
``metaverse'' can be described as the ``embodied internet.'' What in 
your view, should we expect with the ``embodied internet'' in terms of 
our concerns about AI and algorithms?
    Answer. While the digital medium, like all communications media, 
independently reshapes our senses and sensibilities, much of what 
appears to be the mere march of innovation in tech is driven by deep-
seated strategic conflicts among powerful institutions and personnel 
reaching back decades. These conflicts shed important light on how our 
concerns toward AI and algorithms should be dramatically sharpened as 
conflicting players wrestle for metaverse dominance.
    A fresh example concerns recent news that Meta suddenly scrapped 
the development of its new operating system for virtual and augmented 
reality devices, after four years of work by hundreds of programmers on 
the project, in the wake of the abrupt departure of team lead Mark 
Lucovsky. Lucovsky told the press he quit--to join Google in a similar 
role--due to Facebook's hard pivot to the metaverse and the so-called 
``Facebook whistleblower'' Frances Haugen's disclosures about the 
company.
    Lucovsky is not just some random nerd striking out for greener 
pastures. He came aboard at Facebook as their Oculus system's General 
Manager of Operating Systems after building Chinese and American 
engineering teams for VMware--a Silicon Valley cloud computing and 
machine virtualization company that acquired in the mid-2000s companies 
based in London, Switzerland, Bulgaria, and Israel, the latter of which 
led to VMware's opening an R&D facility in Israel based on the team in 
place in that country at the time of acquisition.\1\ Previously to 
VMware, Lucovsky served for five years as Engineering Director at 
Google, which contentiously hired him away from Microsoft. In a lawsuit 
over Google's poaching from Microsoft of another high-powered tech 
executive, Kai-Fu Lee--now best known as the former head of Google 
China--it was revealed that Lucovsky testified in a statement that 
then-CEO of Microsoft Steve Ballmer reacted to news of his move to 
Google with a profane and violent tirade directed against Google chief 
Eric Schmidt. Lucovsky's decision to return to Google to do for it what 
he was doing for Facebook leaves Meta in a position of striking 
weakness, reliant on Android, the mobile operating system of none other 
than Google itself.
---------------------------------------------------------------------------
    \1\ Like America's, much of Israel's high-tech industry arose from 
military and intelligence agencies and funding. B-Hive Networks, the 
Israeli company VMware acquired, was co-founded by CTO Asaf Wexler. 
Wexler began his industry career at Gilat Satellite Networks, a firm 
founded by former Israeli intelligence officer Yoel Gat--himself a 
protege of pioneering technologist and former Defense Ministry 
Electronic Research Department head Zohar Zisapel. Microsoft took over 
a quarter stake in Gilat, infusing $50 million, in the 2000s; by 2011, 
the company had entered the U.S. defense market, acquiring U.S. 
satellite systems integrator and transceiver company Wavestream 
specifically to do so. In America and abroad, the military-intelligence 
and digital industries grow hand in hand.
---------------------------------------------------------------------------
    Lucovsky's dramatic peregrinations illustrate how Silicon Valley is 
an extremely small world made up of a very intimately and 
internationally connected group of programmers and executives working 
in a highly strategically sensitive and secretive pressure cooker 
environment evolving at breakneck speed. This reality must color 
lawmaker's understanding of what exactly was at work with Haugen's 
slickly launched and operated ``whistleblower'' campaign. Rather than a 
spontaneous act of conscience, Haugen's doings--which extend far beyond 
her media-ready Hill testimony to an in-depth consultative tour with 
top European regulators aiming to crack down on Meta through a sweeping 
new Digital Services Act--are organized and facilitated by a cast of 
global lobbyists with elite backing and support. This shop, called 
Reset, is run by Ben Scott, a top tech advisor to Hillary Clinton, both 
in her days as Secretary of State and during her run for president, 
instrumental to the promulgation of the 21st Century Statecraft 
doctrine instructing the State Department to thoroughly technologize 
global diplomacy.
    Today Reset draws up to $10 million a year from eBay billionaire 
Pierre Omidyar through the London-based activist philanthropy group 
Luminate, whose CEO, Stephen King, is a longtime Omidyar lieutenant 
previously the head of BBC Media Action, a body funded separately from 
the BBC that reputedly gives MI6 and the British Foreign and 
Commonwealth Office a quiet means to conduct information operations 
such as the FCO's Counter Disinformation & Media Development campaign, 
run until 2019 by career civil servant Andy Pryce. While, to be sure, 
there is no demonstrable connection between Pryce, King, or Scott to 
``dodgy dossier'' author Christopher Steele, the former head of the 
Russia desk at MI6 during the financial crisis, a pattern in strategic 
conduct and conflict is strongly suggested by the manner in which 
Steele's ostensibly unofficial investigation of then-president Donald 
Trump was effortlessly laundered into the national political battle 
through the slick offices of Fusion GPS on the Clinton campaign's dime.
    If these details seem to suggest some kind of kill shot has been 
orchestrated against Meta to the benefit of its top strategic rival 
Google, that is no coincidence. The question is why. The point is not 
to ensnare lawmakers' attentions in a wilderness of mirrors, but rather 
to underscore that the current conflict in Silicon Valley over 
metaverse dominance is inseparable from the broader geopolitical 
conflict inside and outside the U.S. over digital dominance and 
control. It is not unreasonable to infer that, in the case of Meta, 
powerful individuals and organizations in and out of the Federal 
government have acted on a strong interest in punishing Facebook for 
Mark Zuckerberg's and certain board members' perceived responsibility 
for Donald Trump's 2016 election victory and near-win in 2020. This 
interest, more than a mere partisan disagreement, is clearly animated 
by a more comprehensive ideological conviction that their factional 
control over the entire digital infrastructure of the United States, 
and its projection around the world, is essential and non-negotiable.
    The upshot of this high-stakes gambit is that, while the metaverse 
is presented to would-be users as an escape from the harsh challenges 
and overwhelming confusions of the real world--an infinite televisual 
paradise within which to play safely and freely--the figurative TV in 
which users are to be absorbed is itself enclosed within a vast 
computer system which unelected officials and their powerful allies 
outside government use to manipulate and control masses in a vast power 
struggle for digital sway over the country, the globe, and the minds 
and bodies of those within them.
    In short, it is not an exaggeration to say that AI and algorithms 
designed for a mainstream and ubiquitous metaverse are geostrategic 
tools of the utmost power and influence, and that domestic and foreign 
intelligence agencies, governments, NGOs, corporations, and activist 
groups should be expected to wrestle for influence and control over 
their design, reach, and consequences--even as users are in all 
likelihood constantly conditioned to perceive and experience the 
metaverse not as a system of unprecedented social control but as an 
unprecedented funhouse within which, no matter how dark the world 
outside, they can, in Neil Postman's portentous phrase, amuse 
themselves to death.

    Question 3. You stated in your testimony that big tech corporations 
``have moved so much of our political and social life into their 
technological ecosystem that they now make and enforce fundamental 
decisions about what we can and must think, say, and do.''
    With these big tech platforms using persuasive technology through 
algorithms to manipulate users, could you please elaborate on the 
harmful impacts this has caused to society, and if these practices 
continue, what will the future hold? What steps would you recommend 
Congress take to protect freedom of speech?
    Answer. There is little doubt that the covid crisis has revealed 
pervasive coordination between Big Tech and the Federal government to 
push official messaging and suppress disfavored information concerning 
a host of matters touching upon the virus and its harms as well as its 
vaccines and treatments and their consequences and the likelihood 
thereof. What the CDC or WHO or the White House or Dr. Anthony Fauci 
says, no matter how quickly obsolete or intentionally misinforming, is 
made sacrosanct or memory holed, with no accountability and no 
admission of guilt, error, contradiction, or sometimes even change in 
messaging. Meanwhile criticism, questioning, and information deviating 
even slightly from official statements, which are now indistinguishable 
from propaganda, is censored and those who disseminate it are 
suspended, deplatformed, or otherwise punished. This is just a 
foretaste of how the expropriation of citizen deliberation on 
foundationally political matters into online forums with post-
constitutional guidelines may harm and even obsolesce our way of life 
and constitutionally guaranteed form of government.
    It is also beyond question that the rubric of ``hate speech'' is 
being used to progressively police and enforce expansive censorship and 
deplatforming codes against platform users online. Hate, an ill-defined 
term with no constitutional status and until almost moments ago no 
presence at all in any American or Western legal codes, has become a 
core component of ``woke'' doctrines increasingly codified, under the 
banner of Critical Race Theory, ``antiracism'', and other movements, 
into our increasingly online public lives. Legal consequences are being 
imposed for failing or refusing to use in the workplace people's 
preferred pronouns whatever they may be.
    Still more, in the wake of the January 6 turmoil at the Capitol, 
Federal agencies and their allies outside government are pushing to 
implement online monitoring and surveillance regimes intended to 
automate the process of detecting and shutting down speech and users 
whose online behavior is considered to indicate precursors to 
``extremist'' identification, communication, or action. This approach 
treats constitutional speech guarantees with manifest contempt, 
encouraging retaliatory and partisan officials to make unilateral and 
arbitrary decisions about which words, phrases, and political positions 
should trigger algorithms and AIs surveilling Americans, leading to 
their punishment online and off. Even elementary discussions about core 
questions concerning any citizen in a free society--such as immigration 
levels, government corruption, the influence of unelected officials, 
the rights of parents to know and influence their children's education 
and morals, and speech codes themselves--are now considered precursors 
to ``extremism''. In one recent case, the National School Boards 
Association responded to parents angered by school board policies by in 
effect demanding of the White House that they be classified as 
``domestic terrorists'' and, by implication, treated accordingly.
    Although this ordeal did not transpire solely on the internet, the 
line between the online and offline manifestations of this controversy 
blurred away completely as it now does in the case of almost every 
nontrivial political disagreement in America and many trivial ones 
besides. However well-intentioned, efforts to institutionalize a 
radical overturning of the American and Western legal tradition around 
speech and identity are being put central to activist efforts within 
and outside Big Tech to use algorithms and AI as ethically pure tools 
to mandate and enforce ethical purity and its conspicuous expression 
upon users.
    Today's technological tools to shape and reshape hearts and minds 
are dangerously powerful, but there is no putting them back in the box 
or tearing them all down. Americans have made our bed creating these 
technologies and unleashing them on ourselves and the world and now we 
must lie in it. Nevertheless, a central part of accepting our 
responsibility for these fateful acts is to ensure that the entities we 
have created do not tear down our civilization and our regime and shape 
us into building a new one alien and hostile to all that has come 
before. In various debates about the degrees to which the Constitution 
applies in areas under the control of the United States government, it 
has sometimes been argued that the best doctrine is one wherein the 
Constitution goes where the flag goes. Within U.S. territory at a 
minimum, U.S. citizens and persons on the Internet and utilizing 
smartphones are fully ``under the flag'' and within the jurisdiction of 
the United States. Congress must ensure that the Constitution is in 
force in our online life as much as our offline life. That means no 
speech codes, no chilling of speech, no mandatory expression or 
identification, and no prosecution or punishment of communication as 
those acts implicate the First Amendment or the Constitution more 
broadly. It means no surveillance, policing, and punishment of online 
communication on a basis of identifying and monitoring users and 
activities as pre-criminal ``extremists'' to be treated as legally 
suspect or as second-class citizens. It means taking our form of 
government and our liberties as seriously as our forebears took them, 
as precious gifts and achievements worthy of the ultimate sacrifice, 
come what may.

    Question 4. I recently spoke with a well-known and well regarded 
technologist who told me that in the near term future, there may be 
children whose best friend is an AI. That seems to me to be a cause for 
great concern. How do we prevent a future where our children's best 
friend is an AI?
    Answer. Many factors fuel the increasing emotional reliance of 
adults as well as children on machines that more enchantingly imitate 
human interactions--and the increasing desire among adults to make 
children ever more reliant in that way.
    One is the sheer formative power of the digital medium, which 
reshapes our senses and sensibilities in, so to speak, its image. The 
digital medium is all about the supremacy of machine memory, a faculty 
of recordation and recall so powerful that our human memory and 
imaginations appear suddenly obsolete or humbled. Because the 
previously dominant televisual medium enhanced and rewarded our faculty 
of imagination, we came to believe intuitively that whoever could dream 
the biggest and best dreams was, in a fundamental sense, the best, 
entitled ethically and practically to take charge, lead masses, rule 
the world. We are now suffering a wrenching transformation into a world 
where that ethic of order and ordering is overthrown by a new ethic of 
masterful machine memory. In response, people are showing a dangerous 
willingness to sacrifice their faculty of human memory so long as they 
can retain what appears to be a special zone where radical imagination 
can be given what feels like free play: the past and reality as such 
are being demonized as morally wrong and practically broken, and the 
kinds of wholesale socioeconomic transformations being urged as the 
only possible response to the triumph of digital require a sacrifice of 
people's memories of their culture, their family, their bloodline, 
their biography, and more.
    The Great Reset effort and the revolutionary digital culture 
currents allied with it aim ultimately to create a ``Stunde Null'' or 
``Year Zero'' moment for America, akin to the one America once created 
for Germany in 1945. It is impossible to accomplish this without 
acculturating children into a fundamentally digital world, one where 
their uniquely human faculties are not seen as a precious gift but as a 
flawed or sinful impediment to the creation of a new and better 
lifeworld. In fact, children are the necessary place in a logical sense 
to begin with this recreation, because their memories are the fewest 
and in some respects the most weakly formed and easily broken. One need 
not read Brave New World in order to recognize the centrality of 
indoctrinating the youth to radical projects of reordering the very 
fundaments of human life. For the digital revolutionaries, children 
must now therefore be indoctrinated to see their humanity instinctively 
as a primitive and unpleasant state which can be overcome and 
transcended, for the good of the planet, of life, and of the universe, 
by the intimate merging of their being with that of digital entities.
    This process is already being demonstrated at several levels, of 
which kids pair bonding with AIs or robots or virtual friends is just 
the most basic. Law in the Anglosphere (see recent developments in 
Canada) is being transformed so that it is illegal to stop a child from 
becoming transsexual or transgendered, a process inseparable from and 
at the vanguard of the process of becoming transhuman--both through 
hard technology and the soft technology of online life where social 
contagions of re-identification as trans- or posthuman spread 
constantly without censorship or suppression. Meanwhile the law does 
nothing to punish or even supervise those who relentlessly, obsessively 
groom children--that is, other people's children--to become queer in 
all its myriad expressions. It is undeniable that under digital 
conditions laws welcoming and protecting the adoption of children by 
households lacking a married mother and father contribute to the 
normalization of not just post-heterosexual but post-biological and 
ultimately post-human identity formation. It is painfully easy to see 
how the next big pushes for ``rights'' will include polygamist rights, 
cyborg rights, and trans- or posthuman identifying rights. If Congress 
shies away from laying down laws that put limits on which mutations and 
malformations of our human identity may become the next identities and 
practices to be elevated for official approval, protection, privilege, 
and celebration, the task will fall to the states, where patchwork 
efforts will surely face a tsunami of litigation, doubtless backed by 
at least some public officials at the highest possible rank.
    In the realm of education, children need preserved and strengthened 
from the earliest ages their foundational human bonding to mother and 
father and their foundational development of independently human senses 
and faculties. This means that measures such as ``universal pre-K'' 
which institutionalize the removal of young children from the household 
away from the parents for daily education or socialization, however 
well-intended, should not be supported. It means that, for parents, 
homeschooling should be made as easy as possible and without penalties 
or undue burdens. And it means that lawmakers should undertake a 
thorough, restrictive, and exemplary review of funds allocated toward 
programs and initiatives that treat it as an inherent good or act of 
progress to ``bring more tech'' into schools. Adding digital devices to 
the classroom does not automatically bring anything to students' 
educations except the interests and agendas of those furnishing the 
technology and of educators who see that technology as their most 
powerful way to advance them.
    Nevertheless, it is essential to recognize that children and 
students do have a vital interest in learning confident human 
competency at mastering our most powerful digital technologies for the 
purposes of keeping them in check and strengthening and protecting our 
human identity, our human culture, and our American lifeway and free 
republic. Trying to hide kids away or come up with a magic age at which 
they can finally access such technologies is wrongheaded and futile. 
The goal is to ensure that technological education is not centered on 
convincing kids that they can become masters of their identities in 
machine-made fantasy worlds or that they have a global moral mission to 
transform other children in that image. That means supporting programs 
and initiatives that encourage or supply access and training in the 
competencies and the culture at the core of Americans' digital rights.
                                 ______
                                 
   Response to Written Questions Submitted by Hon. Kyrsten Sinema to 
                            Dr. Dean Eckles
    Spread of Misinformation. Misinformation remains rampant on social 
media and is a significant problem. For example, misinformation about 
the COVID-19 pandemic has contributed to public confusion. Also, during 
the 2020 presidential election, a number of false rumors circulated on 
social media platforms alleging wrongdoing by election administrators.

    Question 1. What steps can social media companies take to combat 
the harm of misinformation? Could algorithms work in reverse to limit 
the content containing false information a user encounters instead of 
leading them down a path to continued exposure to misleading claims?
    Answer. It is worth noting that some leading social media companies 
are engaged in substantial efforts to reduce spread and associated 
harms of misinformation. But this is not an easy problem to solve, and 
there is often substantial disagreement about what good outcomes look 
like.
    One challenge is identifying obviously harmful misinformation 
before it has already spread widely. For example, efforts to use fact 
checkers can be limited by the fact that, by the time something is 
identified as starting to ``go viral'' and then is fact-checked by 
professionals, much of its eventual spread may already have occurred. 
This is one motivation for these companies' investments in other 
approaches, such as crowd-sourcing judgements about headlines. There 
are clear promises and challenges to these approaches as well, which 
have been examined by my MIT colleagues. [Some of this research is 
summarized in Allen, J., & Rand, D. (2021) ``How the Wisdom of Crowds 
Could Solve Facebook's Fact-Checking Problem'' Time.] Some of those 
same researchers have been consulted by, e.g., Twitter in their 
Birdwatch program and some of the data analyzed arises by trials 
conducted by Twitter.
    Short of labeling or downranking misinformation, platforms also use 
other signals that content might not be what people initially take it 
as. As described in my written testimony, for each item they might 
display to a user, some platforms predict ``negative'' actions--actions 
that would indicate that showing this item to this person is not 
beneficial (often by that person's own lights). This can include 
sequences of actions such as resharing the item and then deleting that 
reshare, perhaps reflecting regret or initial misunderstanding. It also 
can include whether that person, on reflection if asked, would say the 
item is important, informative, etc. These kinds of predictions play a 
substantial role not just in addressing harmful misinformation, but 
also in reducing exposure to other types of content like spam.
    This is not to say that platforms are doing enough or that they 
don't deserve criticism. They should continue to make growing 
investments in this area. Platforms that are not doing some of these 
things that constitute the state-of-the-art, such as using these 
sophisticated negative signals, ought to catch up and can certainly be 
criticized for not doing so. Likewise, reporting indicates that 
Facebook and other platforms have seemingly invested much less in 
addressing these problems in some markets and languages, as discussed 
below.

    Question 2. What steps could Congress take to address this issue 
while recognizing the importance of freedom of expression?
    Answer. It seems important to continue to provide clarity that 
algorithmic ranking and recommendation do not make platforms broadly 
liable for content shared by users. As highlighted in my written 
testimony and work by legal scholars (e.g., Daphne Keller), defining 
whether particular content has been algorithmically amplified in a 
rigorous and useful way is quite difficult. Regulations that would make 
intermediary liability depend on such a definition may have unintended 
and undesirable consequences, if they are workable at all.

    Foreign Language. Media reports suggest that many social media 
companies have not invested resources in addressing misinformation 
being spread in languages other than English. This poses risks to non-
English speaking communities in the United States, including Spanish 
speaking residents in Arizona, as well as residents of foreign nations 
who utilize these tech platforms that have headquarters in the United 
States.

    Question 3. What responsibility do tech companies have to prevent 
their platforms from being used to spread misinformation in languages 
other than English, which can result in political violence in the 
United States or in foreign nations?
    Answer. I do not see any fundamental reason why the same 
responsibilities platforms that have to their users and the public in 
English ought not apply in other languages. From recent reporting, it 
seems that investments in moderation and algorithmic ranking in other 
languages have often not been commensurate with the very large 
footprint platforms have in other languages.
    There are some more reasons--not as attributable to choices by 
social media platforms as others--why automated efforts are further 
behind outside of English and a few other languages. First, in some 
settings there are multiple languages in use, often in combination. 
Some of my own research has involved studying communication in India. 
Even within the same context, messages can switch among Hindi, English, 
other local languages, and combinations thereof. This can make 
algorithmic processing of content much more difficult (and it can even 
make it harder to find people who can do accurate labeling of content).
    Second, general-purpose natural language processing (NLP) is less 
well-developed for many languages. I was struck by this when trying to 
begin working with data in Hindi. This can reflect baseline differences 
between languages (e.g., available parallel texts for machine 
translation) and large differences in commercial, academic, and 
government investments.
    Congress could consider how increased funding for research in 
computational linguistics and NLP in otherwise neglected languages may 
have broader positive consequences, as companies large and small often 
build on this more basic research.

    Research Access. We heard during the hearing about the potential 
value in permitting researchers, particularly those in an academic 
setting, to examine user data from tech companies or from probing 
social media sites by setting up accounts to examine the content those 
accounts receive exposure to.

    Question 4. What types of data do you believe Congress should 
require tech companies to disclose to researchers so they can examine 
the effects of algorithms on every day Americans?
    Answer. There is valuable space between mandates for data sharing 
and regulations that discourage data sharing. That is, Congress can 
make it more straightforward for tech companies to responsibly share 
data with researchers, can reduce legal risks to researchers and their 
institutions when conducting research discouraged by the platforms, and 
can take care in crafting data privacy legislation so that new barriers 
are not introduced.
    As an example of barriers to data sharing, note that when Facebook 
and Social Science One announced their initiative for sharing data 
relevant to misinformation and elections with screened researchers, 
some groups (e.g., Electronic Privacy Information Center) argued that 
this was not permitted under Europe's General Data Protection 
Regulation (GDPR) and the FTC consent decree with Facebook.
    Congress could make explicit how such sharing can be squared with 
other data privacy and protection regulations. This may be important in 
avoiding ambiguity about whether particular frameworks (e.g., 
differential privacy, as used by the U.S. Census Bureau, Facebook, and 
others) for privacy-preserving data releases would satisfy various 
regulations.

    Question 5. Broadly speaking, how would you define ``researchers'' 
to ensure that only legitimate researchers can obtain information?
    Answer. I worry about any policy that depends on giving necessary 
and sufficient conditions for being a legitimate researcher in order to 
protect the privacy of users (or other interests, such as protecting 
intellectual property).
    The scandal associated with Cambridge Analytica's collection of 
Facebook data via a platform application involved a seemingly 
legitimate researcher, Aleksandr Kogan. Similar prior research 
efforts--which Cambridge Analytica was imitating--also involved 
collecting and distributing data in ways that may have importantly 
compromised user privacy, as they, e.g., distributed data (presumably 
inadvertently) containing personally identifiable information. So 
limiting data access to some broad category of legitimate researchers 
is not enough to protect privacy, and any mandate for data sharing 
without clarity about methods for preserving privacy could be harmful. 
On the other hand, some ways of defining legitimate researchers would 
exclude important work in the public interest by, e.g., journalists at 
for-profit media companies.
    Even in the presence of some mandated or voluntary data sharing 
processes, important work by external researchers (including academics, 
journalists, and others) will continue to depend on the ability to 
``probe'' these systems in other ways (e.g., scraping public content, 
creating accounts to do audits). Sometimes these methods are actively 
discouraged by platforms in ways that create legal ambiguity and risk 
for researchers, their institutions, and scholarly publishers, 
including the threat of application of the Computer Fraud and Abuse 
Act. Congress could go beyond current case law (e.g., Sandvig v. 
Sessions) in protecting the freedom to conduct such research--without 
requiring any direct cooperation and disclosures by platforms.

    Transparency and Intellectual Property. I want to ensure that the 
Internet remains a place of innovation. In addition to academic 
researchers, some have proposed that the general public have additional 
information about the algorithms tech companies employ to run their 
platforms.

    Question 6. If Congress pursues measures to require tech companies 
to disclose more information about algorithms, how can we ensure that 
those companies' intellectual property remains protected and does not 
fall into the hands of competitors or nefarious actors?
    Answer. I agree that this is a concern. I do not have any complete 
solutions, but perhaps much of the value from such disclosures come 
less from the details of the algorithm and the predictive statistical 
machine learning models used and more from the general architecture of 
the algorithms and how they trade off different objectives. 
(Furthermore, the ``full'' algorithm is usually somewhat of a sprawling 
thing and is not readily interpretable by anyone, especially anyone 
without access to internal data and massive computing resources. My 
written testimony discusses this somewhat.)
    Another way some disclosures could limit innovation is if they 
inadvertently restrict the scope for algorithmic ranking and 
recommendation or limit the kinds of user interfaces in which content 
is presented. Some of the platforms that have seen rapid growth 
recently have done so with other interfaces for presenting content. And 
one way that platforms can innovate is by increasing the repertoire of 
actions they can take with content (e.g., displaying the same content 
in different ways), as discussed in my written testimony. So 
requirements for disclosure that are written too narrowly could either 
have key blind spots or impede such innovation.
                                 ______
                                 
     Response to Written Question Submitted by Hon. John Thune to 
                            Dr. Dean Eckles
    Question. The ``metaverse'' appears to be what we can expect the 
Internet to be in the near term future. As you know, Mark Zuckerberg 
changed the name of his company to Meta in October, and said that the 
``metaverse'' can be described as the ``embodied internet.'' What in 
your view, should we expect with the ``embodied internet'' in terms of 
our concerns about AI and algorithms?
    Answer. Since the ``metaverse'' remains largely a thing of fiction, 
rhetoric, and narrow demonstrations, answers to this question are 
necessarily somewhat speculative.
    Some features of embodied, immersive virtual environments would 
seem to actually present substantial barriers to algorithmic 
personalization as it currently exists. If two people are both in the 
same virtual space, it can be important for them to in fact have 
largely the same environment. It could be quite disruptive of the 
immersive experience and social interaction if they are, e.g., seeing 
different art on the walls or even rooms or buildings had different 
locations or contents. That is, in order for the whole metaphor and 
experience to work, there may be substantial limitations on the scope 
for personalization. (This may actually be related to some reasons why 
demand for a ``metaverse'' experience will be somewhat limited: how 
often do we really want to bring along most of the constraints of 
everyday physical space when working, socializing, being entertained, 
etc.?)
    However, there is still important scope for AI and algorithmic 
personalization in immersive virtual environments. Over a decade ago, 
Jeremy Bailenson at Stanford and colleagues already examined how simple 
algorithms can be used to modify the behavior of a single avatar so 
that it appears differently to different participants in a meeting. For 
example, the avatar of a speaker in a meeting could simultaneously make 
eye contact with multiple other participants or subtly mimic the 
gestures of multiple other participants; there is some work suggesting 
this could make them appear more credible and be more persuasive. Thus, 
such considerations are certainly not absent.

                                  [all]