[Senate Hearing 117-514]
[From the U.S. Government Publishing Office]
S. Hrg. 117-514
SOCIAL MEDIA PLATFORMS
AND THE AMPLIFICATION OF DOMESTIC
EXTREMISM AND OTHER HARMFUL CONTENT
=======================================================================
HEARING
BEFORE THE
COMMITTEE ON
HOMELAND SECURITY AND GOVERNMENTAL AFFAIRS
UNITED STATES SENATE
ONE HUNDRED SEVENTEENTH CONGRESS
FIRST SESSION
__________
OCTOBER 28, 2021
__________
Available via the World Wide Web: http://www.govinfo.gov
Printed for the use of the
Committee on Homeland Security and Governmental Affairs
[GRAPHIC NOT AVAIABLE IN TIFF FORMAT]
U.S. GOVERNMENT PUBLISHING OFFICE
47-981 PDF WASHINGTON : 2022
-----------------------------------------------------------------------------------
COMMITTEE ON HOMELAND SECURITY AND GOVERNMENTAL AFFAIRS
GARY C. PETERS, Michigan, Chairman
THOMAS R. CARPER, Delaware ROB PORTMAN, Ohio
MAGGIE HASSAN, New Hampshire RON JOHNSON, Wisconsin
KYRSTEN SINEMA, Arizona RAND PAUL, Kentucky
JACKY ROSEN, Nevada JAMES LANKFORD, Oklahoma
ALEX PADILLA, California MITT ROMNEY, Utah
JON OSSOFF, Georgia RICK SCOTT, Florida
JOSH HAWLEY, Missouri
David M. Weinberg, Staff Director
Zachary I. Schram, Chief Counsel
Christoper J. Milkins, Director of Homeland Security
Moran Banai, Senior Professional Staff Member
Kelsey N. Smith, Research Assistant
Pamela Thiessen, Minority Staff Director
Andrew Dockham, Minority Chief Counsel and Deputy Staff Director
Kirsten D. Madison, Minority Director of Homeland Security
Maggie Frankel, Minority Senior Professional Staff Member
Sam J. Mulopulos, Minority Professional Staff Member
Laura W. Kilbride, Chief Clerk
Thomas J. Spino, Hearing Clerk
C O N T E N T S
------
Opening statements:
Page
Senator Peters............................................... 1
Senator Portman.............................................. 2
Senator Hassan............................................... 19
Senator Johnson.............................................. 21
Senator Ossoff............................................... 24
Senator Rosen................................................ 26
Senator Lankford............................................. 28
Senator Romney............................................... 31
Senator Padilla.............................................. 34
Prepared statements:
Senator Peters............................................... 39
WITNESSES
Thursday, October 28, 2021
Hon. Karen Kornbluh, Director, Digital Innovation and Democracy
Initiative, and Senior Fellow, The German Marshall Fund of the
United States.................................................. 4
David L. Sifry, Vice President, Center for Technology and
Society, Anti-Defamation League................................ 7
Cathy O'Neil, Ph.D., Chief Executive Officer, O'Neil Risk
Consulting and Algorithmic Auditing............................ 8
Nathaniel Persily, Ph.D., Co-Director, Stanford Cyber Policy
center, and James B. McClatchy Professor of Law, Stanford Law
Center......................................................... 10
Mary Anne Franks, D.Phil., Professor of Law and Michael R. Klein
Distinguished Scholar Chair, University of Miami............... 13
Alphabetical List of Witnesses
Franks, Mary Anne D. Phil.:
Testimony.................................................... 13
Prepared statement........................................... 104
Kornbluh, Hon. Karen:
Testimony.................................................... 4
Prepared statement........................................... 41
O'Neil, Cathy Ph.D.:
Testimony.................................................... 8
Prepared statement........................................... 75
Persily, Nathaniel Ph.D.:
Testimony.................................................... 10
Prepared statement........................................... 80
Sifry, David L.:
Testimony.................................................... 7
Prepared statement........................................... 46
APPENDIX
Southern Poverty Law Center statement submitted for the Record... 115
Responses to post-hearing questions for the Record:
Ms. Kornbluh................................................. 123
Mr. Sifry.................................................... 125
Ms. O'Neil................................................... 130
Mr. Persily.................................................. 131
Ms. Franks................................................... 151
SOCIAL MEDIA PLATFORMS
AND THE AMPLIFICATION OF DOMESTIC EXTREMISM AND OTHER HARMFUL CONTENT
----------
THURSDAY, OCTOBER 28, 2021
U.S. Senate,
Committee on Homeland Security
and Governmental Affairs,
Washington, DC.
The Committee met, pursuant to notice, at 10:15 a.m., via
Webex and in room 342, Dirksen Senate Office Building, Hon.
Gary C. Peters, Chairman of the Committee, presiding.
Present: Senators Peters, Hassan, Sinema, Rosen, Padilla,
Ossoff, Portman, Johnson, Lankford, Romney, Scott, and Hawley.
OPENING STATEMENT OF CHAIRMAN PETERS\1\
Chairman Peters. The Committee will come to order. I would
like to thank our witnesses for joining us here today.
---------------------------------------------------------------------------
\1\ The prepared statement of Senator Peters appear in the Appendix
on page 39.
---------------------------------------------------------------------------
Our Committee has held several hearings this year,
examining the rise of domestic terrorism, and today's hearing
is going to focus on the role of social media platforms and the
role that they play in the amplification of domestic extremist
content and how that content can translate into, unfortunately,
real-world violence.
Yesterday marked three years since a white supremacist
gunman opened fire in a Pittsburgh synagogue, killing 11
innocent people in the deadliest attack on the Jewish community
in the United States. The attacker used the fringe social media
platform Gab prior to the attack to connect with like-minded
extremists and spread his own hateful and anti-Semitic views
online. While bad, violent, and hateful ideology has long
terrorized communities for many Americans, the shocking attack
was the first glimpse at how quickly increased exposure to
extremist content and users with similar beliefs can radicalize
domestic terrorists and drive them to act on their violent
intentions.
Less than a year after the Tree of Life attack, we saw a
white nationalist open fire in an El Paso shopping center. This
attacker was one of many who viewed video of the Christchurch
mosque massacres that widely circulated on social media just a
few months earlier, a video he reportedly cited as inspiration
for his deadly attack in a 2,300-word racist manifesto that he
also posted online.
On January 6, 2021, we saw a stark example of how
individuals went beyond seeing and sharing extreme content
across numerous social media platforms. They were spurred to
action by what they repeatedly saw online, and ultimately, a
mob violently attacked Capitol Police and breached the Capitol
Building.
In attack after attack, there are signs that social media
platforms played a role in exposing people to increasingly
extreme content and even amplifying dangerous content to even
more users. Yet, there are still many unanswered questions
about what role social media platforms play in amplifying
extremist content. We need a better understanding of the
algorithms that drive what users see on social media platforms,
how companies target ads, and how these companies balance
content moderation with generating revenue.
For the majority of social media users who want to connect
with distant family and friends or stay up to date on their
favorite topics, there is very little transparency about why
they see the content, recommendations or ads that populate
their feeds. While social media companies have promoted how
they are providing more curated content for their users, we
have seen how users can be shown increasingly polarizing
content. In worst case scenarios, users are reportedly
recommending more and more extreme content, nudging them down a
dark and dangerous ``rabbit hole.''
Recent reporting and congressional testimony and
revelations in the Facebook Papers have shed some light on
business models that appear to have prioritized profits over
safety and decisions that appear to disregard the platforms'
effect on our homeland security. It is simply not enough for
companies to pledge that they will get tougher on harmful
content. Those pledges have gone largely unfulfilled for
several years now. Americans deserve answers on how the
platforms themselves are designed to funnel specific content to
certain users and how that might distort users' views and shape
their behavior, both online and offline.
As part of my efforts to investigate rising domestic
terrorism, including the January 6th attack, I have requested
information from major social media companies about their
practices and policies to address extremist content so that we
can better understand how they are working to tackle this
serious threat. While we are continuing to work with companies
to get answers and examine relevant data, I am looking forward
to hearing from our experts today about how these platforms
balance safety and business decisions and, specifically, how
these decisions relate to rising domestic extremism.
Ranking Member Portman, you are welcome to start with your
opening remarks.
OPENING STATEMENT OF SENATOR PORTMAN
Senator Portman. Thank you, Mr. Chairman. I appreciate your
holding this hearing. It is a very important topic. I look
forward to hearing from our experts today and then, I think in
a future hearing, hearing from some of the companies
themselves.
The role that social media plays in directing content that
can lead to online and offline harm has taken on new
significance in the past several weeks as we have learned more
about this from whistleblowers and others, and news has emerged
about malfeasance by some of the largest internet platforms.
This exploitation of social media, of course, is not new. In
2016, as Chair of the Permanent Subcommittee on Investigations
(PSI) of this Committee, I held a hearing which examined the
Islamic State of Iraq and Syria (ISISs), use of online
platforms in furtherance of their violent goals. We learned
from testimony that social media accelerates, in this case,
ISIS's radicalization and recruitment efforts and can also
speed up their mobilization to violence.
Today, foreign terrorist actors continue to try to
weaponize social media to inspire radicalization and attacks
against Americans and American interests. This use of social
media for nefarious purposes is not limited to terrorists, of
course. Drug traffickers, foreign adversaries, and a host of
other threat actors exploit online platforms, particularly
social media. China and Russia use these platforms to conduct
influence campaigns targeting Americans, including being
involved in our elections. Drug cartels and gangs use these
platforms to traffic narcotics. Traffickers use these platforms
to exploit children and other vulnerable people, and domestic
violent extremists (DVE) across the ideological spectrum use
social media to spread propaganda and recruit members to their
cause. So it is a broad problem.
Social media platforms acknowledge that the threats exist,
and they talk about what they are doing to prevent bad actors
from exploiting their sites, including artificial intelligence
(AI) to help moderate content, networks of cross-industry
partnerships, and on-staff experts--some of you may have been
in that position--all to prevent or remove dangerous content.
However, there is still a persistent threat of harmful content
despite what the platforms say they are doing.
These social media companies are businesses, so it is not
surprising that Congress and others have trouble getting more
information on their algorithms and how they operate, how they
are designed to amplify content. That is proprietary
information; I understand that. But Congress has heard from
whistleblowers, like Frances Haugen recently, that Facebook has
not addressed troubling aspects of its algorithms, which have
promoted of variety of concerning and alarming posts.
This raises important questions about whether or not it is
time to revisit the immunity provided by Section 230. In 2017,
during my time as PSI Chair, I introduced legislation which
would remove Section 230 immunity from platforms that knowingly
facilitated sex trafficking. So we have dealt with this issue.
That legislation, called the Stop Enabling Sex Traffickers Act
(SESTA), was actually signed into law back in 2018. We have
figured out how to deal successfully with Section 230 at least
in this one narrow area but very important area.
I take advocates, researchers, and even platforms at their
word when they call for regulation. A regulation can take many
forms which puts a premium on having sound information and
analysis as we consider legislation to solve these problems. In
other words, we need to know more. We need to be able to look
under the hood and figure out what the issues are to be able to
regulate properly.
These experts are in front of us here today to help
evaluate the extent of the problem and also discuss what we
should be doing about it. So far, we have found out a lot about
social media business models from third-party researchers and
from whistleblowers. The findings are largely based on
anecdotal evidence. If they want to ensure that Congress
pursues evidence-based policy solutions, I think it is
incumbent upon the platforms to provide quality data.
I am already working with Senator Coons, who is Chair of
the Senate Judiciary Subcommittee on Privacy, on legislation
that would require the largest tech platforms to share data
with legitimate researchers and scholars so that we can all
work together on solutions to these problems that all of us
have identified. Dr. Persily has been an important partner in
this work. I look forward to his testimony today.
Importantly, as we look at these issues, we must take care
that our efforts hold these platforms accountable, but it is
done in a manner that balances First Amendment protections,
which I understand Professor Franks is going to discuss in her
testimony.
Mr. Chairman, again thanks for having this hearing, and I
look forward to hearing from our witnesses, and I thank you for
giving us a chance to have real experts in front of us.
Chairman Peters. Thank you, Ranking Member Portman.
It is the practice of this Committee to swear in witnesses,
so if each of you will please stand and raise your right hand,
including those who are joining us by video.
Do you swear the testimony you will give before this
Committee will be the truth, the whole truth, and nothing but
the truth, so help you, God?
Ms. Kornbluh. Yes.
Mr. Sifry. Yes.
Ms. O'Neil. Yes.
Mr. Persily. Yes.
Ms. Franks. Yes.
Chairman Peters. You may be seated.
Our first witness today is the Honorable Karen Kornbluh,
former Ambassador to the Organization for Economic Cooperation
and Development (OECD) and who currently serves as a senior
fellow at the German Marshall Fund of the United States, where
she leads its Digital Innovation and Democracy Initiative to
ensure technology supports democracies across the globe. Prior
to her role with the German Marshall Fund (GMF), Ms. Kornbluh
served in previous administrations as Chief of Staff (CoS) at
the U.S. Treasury Department and of the Office of Legislative
and Intergovernmental Affairs at the Federal Communication
Commission (FCC), where she negotiated early internet policies.
Welcome to the Committee. You are now recognized for your
5-minute opening remarks.
TESTIMONY OF THE HONORABLE KAREN KORNBLUH,\1\ DIRECTOR, DIGITAL
INNOVATION AND DEMOCRACY INITIATIVE, AND SENIOR FELLOW, THE
GERMAN MARSHALL FUND OF THE UNITED STATES
Ms. Kornbluh. Thank you, Chairman Peters, Ranking Member
Portman, and Committee Members for the opportunity to testify
on this critical issue.
---------------------------------------------------------------------------
\1\ The prepared statement of Ms. Kornbluh appears in the Appendix
on page 41.
---------------------------------------------------------------------------
To underscore the points that you both made, the National
Strategy for Countering Domestic Terror States clearly that the
widespread availability of domestic terrorist recruitment
material online is a national security threat. I would like to
stress three points today. First, the design of the platform
and its algorithms can promote radicalization. Second, this
cannot be addressed by after-the-fact, whack-a-mole content
moderation. Third, we need to change the platforms' incentives
so that they fix these dangerous design elements.
As part of a test, Facebook researchers created an account
for a fictional Carol Smith, a 41-year-old conservative mother
from North Carolina. Within days, Carol was recommended pages
related to QAnon, and within only three weeks the platform
showed her an account associated with the militia group, Three
Percenters. She did not ask to be shown this content, she had
no idea why she got it, and she had no idea who was paying for
it.
Facebook groups can be manipulative as well. Internal
research found that a full 70 percent of Facebook political
groups in the United States were rife with hate, bullying,
harassment, and misinformation. Facebook's own algorithms
recommend extremist groups to users. For instance, Facebook
directs users who like certain militia pages toward other
militia groups.
The platforms also provide tools that help extremists to
organize. So-called ``super inviters'' can create invitation
links to groups that can be shared on or off Facebook. The
platform provides inviters recommendations of specific friends
to invite, allowing them to recruit from other conspiracy and
militia groups. As an example, Stop the Steal inviters sent
these kinds of automated invitations to members of other
groups, resulting in high membership overlap with Proud Boy and
militia groups and fueling Stop the Steal group's meteoric
growth rates.
This is a national security vulnerability. It was recently
revealed that 140 million Americans were targeted by troll
farms operating out of Eastern Europe.
Similar algorithm radicalization is evidence on other
platforms. TikTok's algorithm, for instance, also promotes
content from QAnon, the Patriot Party, Oath Keepers, and Three
Percenters.
YouTube has 290 extremist channels. When researchers showed
an interest in militant movements, YouTube suggested videos to
them with titles like ``Five Steps to Organizing a Successful
Militia.'' The platform also recommended videos about weapons,
ammunition, and tactical gear.
Extremists find it all too easy to work across platforms.
Diehards can organize on less moderated platforms, like 4Chan
or Telegram, and then may retail the fringe content on more
mainstreams with just a few clicks.
Second, the whack-a-mole approach to catching content after
it has gone viral cannot work. Facebook employees themselves
admitted this problem. They said that the mechanics of the
platform were behind the hateful speech and misinformation, but
most of their ideas for changing these in order to limit
algorithm radicalizations were rejected. The content moderation
system cannot win against huge volumes of algorithmic
recommendation, but that system is further undermined by
exempting many users with large footprints. No wonder that at
Facebook the researchers said they catch only 0.6 percent of
content that depicts violence or could incite serious violence.
Third, it is critical to change the platforms' incentives.
While Congress works on more comprehensive legislation
regarding privacy and antitrust, a digital code of conduct
could help tackle algorithmic radicalization while protecting
free expression. Congress or the Federal Trade Commission (FTC)
could demand platforms commit to common-sense design changes
and transparency, and the FCC would enforce the companies'
commitments.
Platforms should implement the kinds of design changes that
research has already shown would enable more consistent
application of their own terms of service. For example, a
circuit breaker, like those used by Wall Street to prevent
market crises, could prevent the viral spread of sensitive
content in topics areas with high harm potential while human
reviewers have time to determine whether or not it violates
platform policies.
Second, platforms should commit to transparent third-party
audits, the equivalent of a black box flight data recorder,
like the National Transportation Safety Board (NTSB) gets when
a plane goes down or the data available to the Food and Drug
Administration (FDA) or the Environmental Protection Agency
(EPA), which should not need a whistleblower to access data.
Third, the FTC should enforce commitments to this code
under its Section 5 consumer protection authority. Of course,
Section 230 reform, as contemplated in a number of current
bills, would also allow users to sue in cases of terrorism or
serious harm. Or, they could require a code as a condition of
limited liability, but this would require legislation.
Mr. Chairman, the whistleblowers documents are a look in
the rearview mirror, but Web 3.0 is being built today. It is
essential that we act now to set sensible rules of the road. I
thank you for holding this hearing.
Chairman Peters. Thank you, Ms. Kornbluh.
Our next witness is Dr. David Sifry. He is the Vice
President of the Center for Technology and Society (CTS) at the
Anti Defamation League (ADL). Mr. Sifry leads a team of
innovative technologists, researchers, and policy experts
developing proactive solutions and producing cutting-edge
research to protect vulnerable populations. Additionally, Mr.
Sifry is an advisor and a mentor for companies and was selected
as a technology pioneer at the World Economic Forum (WEF). He
joined the ADL after a career as a technology entrepreneur and
as an executive at companies including Lyft and Reddit. Mr.
Sifry is also an advisor and mentor for companies and was
selected as a technology pioneer at the World Economic Forum.
Welcome, Mr. Sifry. You are recognized for your opening
comments.
TESTIMONY OF DAVID L. SIFRY,\1\ VICE PRESIDENT, CENTER FOR
TECHNOLOGY AND SOCIETY, ANTI-DEFAMATION LEAGUE
Mr. Sifry. Mr. Chairman, Ranking Member Portman, Members of
the Committee, good morning. It is an honor to appear before
you today to address the ways social media platforms amplify
hate and foment domestic terrorism.
---------------------------------------------------------------------------
\1\ The prepared statement of Mr. Sifry appears in the Appendix on
page 46.
---------------------------------------------------------------------------
For over a century, ADL has been a leading voice in
fighting hate in all forms, and we have been tracking online
hate since the days of dialup. In 2017, ADL launched our Center
for Technology and Society to respond to the threat of online
hate. My team advocates for targets of online hate and
harassment. We deeply engage with, and call out, tech platforms
to hold them accountable for their actions and their deliberate
inaction.
Before joining ADL, I spent my career as an entrepreneur
and executive in the tech sector. A trained computer scientist,
I founded six technology companies and served as an Executive
at Lyft and Reddit. I have been on the inside, and I know
firsthand how big tech companies work and how business
incentives drive product, policy, and strategy.
These platforms maximize profits by providing hyper
targeted ads to an audience that spends large parts of their
life online. Core product mechanics like virality and
recommendations are built around keeping you, your friends, and
your family engaged. The problem is that misinformation, hate
filled and polarizing content is highly engaging. So algorithms
promote that content.
As ADL's own research has long suggested and Facebook leaks
confirm, these platforms exploit people's proclivity to
interact more with incendiary content, and tech companies do so
with full knowledge of the harms that result. Ultimately, these
companies neglect our safety and security because it is good
for the bottom line.
With no accountability, no regulation, and no incentives
beyond growth and increasing ad revenue, extremists find a
haven to reach, recruit, and radicalize. Platform algorithms
take advantage of these behaviors, especially to our attraction
to controversial and extremist narratives. As a result, some
users get trapped in a rabbit hole of toxic content, pushing
them toward extremism. This has deadly consequences. ADL
reports show that extremists on mainstream platforms push
people into fringe communities that further normalize hate and
violence. Extremist ecosystems inspire individuals to commit
acts of domestic terrorism as we saw in Charlottesville, Poway,
and El Paso.
Senators, three years ago yesterday, in what was the most
lethal anti-Semitic attack in American history, 11 congregants
were massacred at the Tree of Life Synagogue in Pittsburgh.
Before he attacked, the terrorist posted his anti-Semitic
manifesto, which then spread online and was expressly cited as
inspiration by the Poway and El Paso shooters. How many lives
will be lost before big tech puts people over profit?
The leaked Facebook documents revealed that company
researchers flagged Facebook's key role in spreading conspiracy
theories, inciting extremist violence, and contributing to the
events of January 6th. Company executives were fully aware of
the problem and chose not to act. Self-regulation is clearly
not working. These billion and trillion-dollar companies have
the resources to improve systems, hire additional staff, and
provide real transparency. Yet, they claim it is too
burdensome. Without regulation and reform, they will continue
to focus on generating record profits at the expense of our
safety and the security of our republic.
The leaked Facebook documents, January 6th, and rising
domestic terrorism all confirm what ADL has been stating for
years; social media is a super spreader of the virus of online
extremism. It is time for a whole-of-government, whole of
industry, whole-of-society approach to fighting online hate.
ADL built the PROTECT Plan to address the rise in domestic
extremism and our REPAIR Plan to push back hate to the fringes
of the digital world.
Congress must establish an independent resource center to
track online extremists and make appropriate referrals,
regulate platforms including through targeted Section 230
reform, ensure academic researchers access to data, and require
regular and meaningful transparency records and independent
third-party audits.
It is well past time to hold social media platforms
accountable. Thank you for your leadership in working to bring
an end to this cycle of hate.
Chairman Peters. Thank you, Mr. Sifry.
Our next witness is Dr. Cathy O'Neil, the Chief Executive
Officer (CEO) of O'Neil Risk Consulting and Algorithmic
Auditing (ORCAA), an algorithmic auditing company that helps
companies and organizations manage and audit algorithmic risk.
As an independent data science consultant, Dr. O'Neil works for
clients to audit the use of particular algorithms in context,
identifying issues of fairness, bias, and discrimination, and
recommending steps for remediation. Dr. O'Neil earned a Ph.D.
in math from Harvard, was a post-doc at MIT Math Department and
a professor at Barnard College. She has authored the books,
Doing Data Science and Weapons of Math Destruction: How Big
Data Increases Inequality and Threatens Democracy.
Welcome, Dr. O'Neil. You may proceed with your opening
comments.
TESTIMONY OF CATHY O'NEIL, PH.D.,\1\ CHIEF EXECUTIVE OFFICER,
O'NEIL RISK CONSULTING AND ALGORITHMIC AUDITING
Ms. O'Neil. Thank you so much for having me today. I am in
the lucky position of just being sort of a background expert
here. I am going to try to explain three things: first of all,
what is an algorithm; second of all, what is a recommendation
algorithm; and third, what is a filtering algorithm because
those are the two types of algorithms that you see the most on
social media.
---------------------------------------------------------------------------
\1\ The prepared statement of Ms. O'Neil appears in the Appendix on
page 75.
---------------------------------------------------------------------------
But I am just going to start with what is an algorithm. I
am going to do the opposite of what most people who talk about
AI and big data will do. I am not going to try to make it
complicated; I am going to try to explain it in simple terms
because it is quite simple. It is predicting the future,
predicting what will be successful in the future based on
patterns in the past, what was successful in the past.
It does not even have to be formal. It could be something
you do in your head. Like for example, I cook dinner for my
children every day. I look for patterns in the past. That is
historical data. Well, historical data is just my memories. I
know what my kid eats. He only eats carrots but not broccoli.
That kid will eat raw carrots but not cooked carrots. I have a
lot of information, and I can predict what will be a successful
dinner.
But here is the thing that is really important. Besides the
historical data I am talking about is the definition of
success. I have to be very precise when I make an algorithm,
and I have to say exactly what I mean by success. In this case
of making dinner for my kids, I am going to define success as
my kids eat vegetables.
The reason it is so important how you define success is
because you actually optimize to success. So I am going to make
meals that are likely to be successful. Time after time, I am
going to learn from my past mistakes or successes, and I am
going to make meals that will be successful in the future.
Now I just want to make the point that a different choice
of success changes everything. If my son were in charge--he is
a fan of Nutella and not so much of vegetables--then we would
have very different meals. We would be optimizing to Nutella
rather than optimizing to vegetables. So just imagine what kind
of meal that looks like. It is completely different.
I want to make the point that algorithms depend a lot on
patterns in the past, but they depend even more on what you
define as success.
I will make a last point about algorithms just in general,
which is like whoever is in power, whoever owns the code,
typically is the one that defines success. I would say, to the
points I have already heard, success for social media platforms
is about money. They are always going to optimize to money
which is, of course, ads, ad clicks.
Now we are going to talk about recommendation algorithms,
which is how social media decides what content to show you or
what groups to offer you, membership for.
I want you to think about this. I want you to think about
your behavior on these platforms, it is a sort of scoring
system, like you have scores in multiple dimensions.
Actually, let's start not with social media platforms, but
let's start with like Netflix. Let's say you watch Diehard, the
movie, twice a week, every week. Then you are going to be
scored in multiple ways by the platform, and for example, you
would be scored along the lines of: Do you like male characters
in your movies or female characters in your movie? Do you like
violent movies or nonviolent movies? Do you like suspenseful
movies or nonsuspenseful movies?
If you watch Diehard twice a week, your sort of male
character, violent, suspenseful movie scores will go up every
time you do it. Every time you watch a movie that is like a
chick flick or a romantic comedy, those scores will go down a
little bit and romantic comedy scores will go up. So you should
think about your profile from the perspective of a
recommendation algorithm as just a series of scores along these
various dimensions that profile you, like what is your taste.
Now for the definition of success of those algorithms, the
point is that they want you to stay on the platform as long as
possible. For social media algorithms, the definition of
success is, again, staying on the platform as long as possible.
They will profile you and score you in all sorts of ways to
figure out how to keep you on the platform.
I want to make the point that this is completely neutral so
far. They would likely score me as very interested in crafting,
and yarn in particular, and they would offer me yarn-type
things, and every time I click on them my yarn profile score
goes up. They would peg me more and more over time as somebody
quite interested in yarn. In that sense, I would become an
extremist with respect to yarn.
Every single person is sort of nudged and profiled with
respect to their interests. I would even add that it is not
just what they are interested in initially, but they can become
more interested in certain things over time because of the
content that is offered them. Similarly, if I watched Diehard
enough on Netflix, I would be offered more and more movies like
Diehard, and that would actually inform my profile and my
tastes in the future.
I would spend time a little bit on filtering algorithms,
but just suffice to say that whereas recommendation algorithms
work quite well for the social media platforms and they make
them very profitable because they do succeed in keeping people
on the platforms, the opposite is true for filtering
algorithms. The idea of getting rid of harmful content, they do
not work well at all. They are very facile, sort of keyword
search-based algorithms, and they are quite unsuccessful
compared to recommendation algorithms.
I will stop there for now. Thank you.
Chairman Peters. Thank you, Dr. O'Neil.
Our next witness is Dr. Nathaniel Persily, the Co-Director
of Stanford Cyber Policy Center and the James B. McClatchy
Professor of Law at Stanford Law School. Dr. Persily's
scholarship and legal practice focuses on American election
law. He is a co-author of the book, Law of Democracy. His
current work has been honored as a Guggenheim Fellow, Andrew
Carnegie Fellow, and a Fellow at the Center for Advanced Study
in the Behavioral Sciences, examining the impact of changing
technology on political communications, campaigns, and election
administration.
Dr. Persily, you may proceed with your opening comments.
TESTIMONY OF NATHANIEL PERSILY, PH.D.,\1\ CO-DIRECTOR, STANFORD
CYBER POLICY CENTER, AND JAMES B. MCCLATCHY PROFESSOR OF LAW,
STANFORD LAW SCHOOL
Mr. Persily. Thank you very much, Chairman Peters and
Ranking Member Portman and Members of the Committee.
---------------------------------------------------------------------------
\1\ The prepared statement of Mr. Persily appears in the Appendix
on page 80.
---------------------------------------------------------------------------
I am going to spend my time today talking about what we
know, what we need to know, and then what to do about it.
Before I do that, let me sort of admit to what we all know,
which is why we are here, and that is that a courageous woman
revealed thousands of pages of documents that previously no one
had seen outside of the company. The revelations themselves,
the content, were quite striking, but the fact that she did it
was really momentous. It sort of brought front and center the
fact that the internal researchers at these companies know an
enormous amount about us and we know very little about then.
That is an equilibrium which is not sustainable. Right? We
cannot live in a world where all of human experience is taking
place on these platforms and the only people who are able to
analyze the data are the people who are tied to the profit
maximizing missions of the firms.
As Senator Portman mentioned earlier, I have been working
both with his staff and others trying to figure out a way to
open up these companies to outside research because they have
lost their right to secrecy. All right? We are at a critical
moment where we need to understand what is happening on these
platforms.
Let me talk about three areas of critical importance when
it comes to research on online harms. The first is how much
harmful content, appears on these platforms. The second is the
role of the algorithm. The third has to do with advertising.
First, how much harmful content is there on these
platforms? The answer is a lot. Right? If you listen or read
the transparency reports, there are millions of examples,
whether it is hate speech or disinformation or inciting content
and the like.
But those numbers are really quite meaningless, so is the
claim that, for example, Facebook takes down four billion
accounts a year. I mean, that is interesting because it seems
like a lot, but none of us really know what the denominator is.
We do not know how much of the offending content is being
shared and viewed by users of the platform and what the
experience is both of the average user and certain subsets of
users.
For most people it is quite sort of important to understand
their experience online is not filled with hate speech; it is
not filled with disinformation and the like. But that is also
the wrong question to ask.
The question--and where I think research has progressed in
the last few years--is to suggest that there is a sizable
minority on these platforms who are experiencing and producing
an enormous amount of this terrible content, whether we are
talking about incitement, hate speech, disinformation and the
like. Particularly because these are folks that it is going to
be very hard to survey, it is going to be very hard to get them
in sort of outside strategies for research, we need the
internal data to figure out exactly what is going on.
Second, on the role of the algorithm. This is an area where
the platforms and outside observers seem to have diametrically
opposed views. Whether it is the Haugen revelations or others
that--from Facebook who have made this point or even
conventional wisdom, sort of the argument is made that they are
maximizing for engagement, and salacious disinforming and
hateful content generates the most clicks, and therefore, that
is what is favored in the algorithm.
If you ask the companies, they say, no, that there are all
kinds of measures that the firms are taking in order to take
down this offensive content, that they take down, as I said,
millions and millions of examples of hate speech and the like.
I will say one other point that they make is that on
encrypted platforms such as WhatsApp and the like that you see
the same types of offending content, and if you go outside the
United States, where those kinds of platforms are much more
ubiquitous and they do not have algorithms, that you see, as
much, if not more, of the hateful content.
Finally, let me talk about advertising. These firms are
advertising monopolies, and we need to treat them as such. That
is what makes them distinctive. We do not really know a whole
lot about how advertising is affecting sort of this information
ecosystem problem that is the subject of this hearing. We know,
of course, about the meddling in the elections, incendiary
content, whether it is from Iran, whether it is China, whether
it is Russia and the like. When outside researchers try to
study advertising, as a group of New York University (NYU)
researchers tried to do, they were kicked off the platform
because they were trying to scrape and to try to find out
exactly how people were targeted and the like.
So now what to do about it? There are many areas of reform
that I think are on the table. Karen Kornbluh mentioned some
important ones. I want to focus on this question of researcher
transparency. We cannot live in a world, as I said before,
where the only people who understand what is happening on the
platforms are the internal researchers to the firms. Whether it
is immunizing outside researchers who want to scrape the
platform as these NYU researchers do or to develop a secure
pathway, maybe administered by the FTC, in order to vet
researchers so that they can analyze privacy-protected data,
that has to be our future. Right? We cannot live in this world
where the platforms hide behind their right to secrecy. Only if
we can get access to this data can we then regulate
intelligently.
Thank you.
Chairman Peters. Thank you.
Our final witness this morning is Dr. Mary Anne Franks, who
is a professor of law and the Michael R. Klein Distinguished
Scholar Chair at the University of Miami. Dr. Franks is also
the President and Legislative and Tech Policy Director of the
Cyber Civil Rights Initiative, a nonprofit organization
dedicated to combating online abuse and discrimination. Her
work is at the intersection of civil rights and technology. Dr.
Franks authored the award-winning book, The Cult of the
Constitution: Our Deadly Devotion to Guns and Free Speech and
has been awarded a grant from the Knight Foundation to support
research for a book titled Fearless Speech.
Welcome, Dr. Franks. You may proceed with your opening
comments.
TESTIMONY OF MARY ANNE FRANKS, D.PHIL.,\1\ PROFESSOR OF LAW AND
MICHAEL R. KLEIN DISTINGUISHED SCHOLAR CHAIR, UNIVERSITY OF
MIAMI
Ms. Franks. Thank you. On October 14, 2021, Facebook
announced a new artificial intelligence project called Ego4D.
The name derives from the project's focus on ego-centric, or
first-person, perception. Among Facebook's plans for this data
include equipping augmented reality glasses with the capacity
to transcribe and recall recordings of what people say and do
around the user.
---------------------------------------------------------------------------
\1\ The prepared statement of Ms. Franks appear in the Appendix on
page 104.
---------------------------------------------------------------------------
Asked whether Facebook had implemented measures to address
potential privacy and other abuses of these capabilities, a
spokesperson replied that the company ``expected that privacy
safeguards would be introduced further down the line.'' As
underscored by multiple internal documents recently released by
whistleblower Frances Haugen, this approach is characteristic
of Facebook, aggressively pushing new untested and potentially
dangerous products on the public and worrying about the
consequences later, if at all.
Documents shared with the FCC note the asymmetrical burden
on employees to demonstrate the legitimacy and user value of
harm mitigation tactics before implementation, a burden not
required of new features or changes aimed at increasing
engagement or profits. While it may no longer be an official
motto, ``move fast and break things'' still seems to be
Facebook's animating philosophy.
It is notable that Facebook chose to announce such a highly
controversial new project just as the company faces a storm of
criticism and scrutiny over documented evidence that it
knowingly allowed violent extremism, dangerous misinformation,
and harassment to flourish on its platforms. One might have
expected Facebook would be more circumspect about drastically
increasing the capacity of individuals to record people around
them without consent in light of the revelation, for example,
that it allowed nude images of an alleged rape victim to be
viewed 56 million times simply because the man she accused of
raping her was a famous soccer star.
Is it arrogance? Is it callousness? Or, is it merely
confidence? Confidence that no matter what is revealed about
Facebook's role in the disintegration of our shared reality or
the dissolution of our democracy--not its acceleration of a
conspiracy theories from QAnon to Stop the Steal, its
amplification of deadly disinformation about Coronavirus
Disease 2019 (COVID-19), its endangerment of teenage mental
health, its preferential treatment of powerful elites, or its
promotion of violently racist and sexist propaganda--that it
will face no real consequences?
After all, that seems to be the lesson that not only
Facebook but other dominant tech companies have learned from
previous scandals. Media attention will be intense for a while.
They may be called before Congress to answer some uncomfortable
questions. They may face some fines. But the companies will
reassure the public that their purpose was never to cause harm.
They will promise to do better in the future.
It should be clear by now that debates over tech companies'
intentions are a distraction and an obstacle to real reform.
Moral and legal responsibility is not limited only to those who
intend to cause harm. We hold entities accountable also when
they know their actions will cause harm or when they are
reckless about foreseeable harms and even sometimes when they
are negligent.
Facebook and other tech companies have known for years that
a business model focused on what is euphemistically called
``engagement'' is ripe for exploitation and abuse. These
companies have, at a minimum, consciously disregarded
substantial and unjustified risks to Americans' privacy,
equality, and safety.
These risks are not politically neutral. Contrary to oft
repeated claims that social media is biased against
conservatives, the algorithms of major social media sites
disproportionately amplify right-wing content. Facebook allows
right-wing news sites to skirt the company's fact-checking
rules and changed its algorithm in 2017 to reduce the
visibility of left-leaning news sites. The day after the 2020
election, 10 percent of all views of political content on
Facebook in the United States were of posts that falsely
claimed that the vote was fraudulent.
As one Facebook employee wrote, if a company takes a hands
off stance for these problems, whether for technical or
philosophical reasons, then the net result is that Facebook
will be actively, if not necessarily consciously, promoting
these types of activities.
According to recently released internal research, Twitter's
algorithms also disproportionately amplify right-wing content.
Research on YouTube's algorithms show that they create a far
more robust filter bubble for right-wing content than left-wing
content and that Fox is, by far, its most recommended
information channel, influence that illustrates how the
ecosystem of extremism and disinformation is driven by forces
beyond social media.
Lopsided political amplification is all the more troubling
given the disproportionate rate of right-wing offline violence.
Since 2015, right-wing extremists have been involved in 267
plots or attacks and 91 fatalities.
To be clear, the object of concern here is not conservative
content as such but content that encourages dehumanization,
targets individuals for violence and harassment, traffics in
dangerous disinformation, and promotes baseless conspiracy
theories that undermine our democratic institutions. Social
media, along with, in some cases, mainstream media, elected
officials, and others with influential platforms amplify these
anti-democratic forces.
Structural reform, including reform to Section 230, that
limits its protections to speech protected by the First
Amendment and denies immunities to intermediaries who exhibit
deliberate indifference to unlawful content, is necessary to
ensure that no industry and no individual is above the law when
it comes to the reckless endangerment of democracy.
Thank you.
Chairman Peters. Thank you, Dr. Franks.
Recent reports based on the leaked Facebook Papers indicate
that Facebook was certainly aware that changes that they made
to their news feed configurations spread dangerous content more
rapidly. Each of our witnesses have mentioned this in one form
or another already. Yet, company leaders repeatedly argue that
they are investing in trust, safety, and civic integrity
efforts. I am certainly struck, as I think you are, by this
fundamental conflict between efforts to take down hateful and
violent content and company algorithms that seem to amplify
extremist views simultaneously.
Mr. Sifry, can you speak to this apparent conflict you have
mentioned in your opening comments but a little more in depth
if you would, please? How successful can these social media
companies be at content moderation if they continue to design
their own platforms to amplify extreme content?
Mr. Sifry. Mr. Chairman, thank you so much for that
excellent question. I think it really cuts to the heart of the
conversation here, that at its core, what we are talking about
is the incentive systems that drive this business model, a
business model that is based around getting you engaged with,
unfortunately, the natural human biases that we have toward
engaging with controversial, with polarizing, with content that
makes us afraid, with incendiary content.
What happens? All of those indicators that were talked
about earlier, right, likes, shares, retweets, you name it,
that those indicators go up. And so these platforms, Senator,
are working as designed. What ends up happening is that we end
up spending more time on those platforms and they end up
tracking us more and then they get to send us more hyper
targeted ads.
This will not change until there is a clear shift in the
incentive systems that they use to be able to do this business,
and that is where Congress must act. In creating systems that
actually bring about a change in their incentive systems, that
is how we end up getting them to behave rationally in this
sense, right, and move toward those different incentive
systems.
Chairman Peters. Professor Kornbluh, just a follow-up
basically on that question, in your opinion, can investments in
trust and safety ever overcome a business model that
prioritizes advertising revenues based on extremist content and
the desire to keep people on a platform as long as possible?
Ms. Kornbluh. I think that goes right to the heart of the
question. The algorithm is this machine that is pushing
misinformation or extremist content into a user's feed. It is
recommending these extremist groups. It is targeting small
groups of people with ads designed to agitate them. It is the
mechanics of the system that are working.
Then these poor human content moderators, or even
outmatched AI systems as Dr. O'Neil was talking about, do not
stand a chance. So their own data shows that less than 1
percent of content that depicts violence or could incite
violence is caught, and this is content that violates their own
terms of service.
This is true; there are new revelations today about even
something that was a high priority from Mark Zuckerberg, which
is catching COVID misinformation. Apparently, they are
unmatched at the content moderation even there. An
international vaccine expert that I talked to told me he was so
frustrated because he felt that the disinformation purveyors
were working with the motor of social media and that the public
health officials were fighting against the engine of social
media.
I always think of that ``I Love Lucy'' skit when she and
Ethel are on the candy conveyor belt and they are desperately
trying to go as fast as the conveyor belt. These content
moderators just do not have a chance.
Chairman Peters. I want you to continue on this line of
questioning, Ambassador. Earlier this year, a report found that
Facebook was posting ads for body armor, gun holsters, and
other military equipment right next to content promoting the
2020 election misinformation and news about the attack on the
Capitol on January 6, so getting this content and ads for
military equipment. What do we know about how Facebook and the
platform targets their advertising and this kind of link? Can
you talk to me a little bit more about what you are finding?
Ms. Kornbluh. Yes. There is so much here, and it is
overwhelming. It is hard to get our heads around this
advertisement of these kinds of weapons, but there are three
ways in which ads in general drive extremism.
First, it is very different than what we think of in terms
of cable and broadcast ads, where they are micro-targeted
specifically to the people who would be most moved by them and
other users who might object, who might say, ``Oh, that is
violent'' or ``Oh, that is wrong,'' they do not get to see
them. The ad content is not reviewed by humans before it is
placed as they are on broadcast and cable. Third, users cannot
really find out who is paying for these ads. Even political ads
can just list a dark money group instead of their true sponsor,
which could be a foreign government. So in 2016, remember,
there were ads paid for in rubles that got through because
there is no human monitoring this and there is very little
transparency.
Then as we have talked about, the algorithms are designed
to maximize this ad revenue. The ads are driving the whole
thing. They are keeping you online to be fed the ads, and the
incendiary content keeps users on the platforms longer.
Then there is this third element that I think is really
important to think about. The algorithms that are trying to
keep you online to feed you these ads, they are creating a
marketplace that values extremism. The more extremist the
content the better the distribution will be, the cheaper the
ads will be actually. It is a cheaper ad if it is more
incendiary per unit because it gets wider spread distribution.
So it is creating this marketplace.
Political parties apparently came to Facebook and said,
your algorithm is driving us to put out more incendiary content
because it is the only way we can get distributed. If we just
put out our 10-point plan, it does not get distributed online.
So the ads are really the heart of this matter.
Chairman Peters. Thank you. The buzzer you heard is a vote.
So we are in a series of votes. You will see Members running in
and out, and that is what I will do. I will go and vote. I will
recognize Ranking Member Portman for his questions, and Senator
Hassan will chair the hearing in my absence.
Ranking Member Portman.
Senator Portman. Thank you, Mr. Chairman, and thank you to
the witnesses. This is a really complicated area, and I am glad
that Dr. O'Neil gave us a little one-on-one on algorithms and
how they work. Ultimately, she came down to the conclusion that
this is about money. This is about what works, kind of along
the lines of what Ambassador Kornbluh just talked about, that
you know, what works is what sells more advertising and these
algorithms are at the core of that. In other words, they are
determining what we want to hear as online participants and
amplifying that.
I will say two things that I just want to try to stipulate
here at the start. Some may not agree with me, but it seemed to
me, Ambassador Kornbluh, in your comments in particular, you
were focused on right-wing extremism as being the problem. It
is not. It is everything.
I hope we can agree to that because, again, the work that
we did early in this Committee with regard to what ISIS was
doing online, and in terms of recruiting and spreading
violence, and in terms of what happens today even--I mean,
there is, as you know, a lot of concern about these platforms
not allowing what happened at the Wuhan lab to leak, to come
out, or you know, concerns about what Hunter Biden is doing or
not doing being blocked, or other things that lead me to
believe that, whether it is Antifa on one side or whether it is
White Supremacy on the other side, we need to look at this as a
problem that is impacting ideologies across the spectrum.
I am just going to stipulate that because I want to get to
some questions. Some of you may not agree with me on that, but
I think that is really important, for us not to make this a
partisan exercise.
Second is the First Amendment. This is impossible. How do
you figure out what is speech that is peaceful expressions of
points of view that we should be encouraging and what is
content that ought to be filtered in some way? And there is
lots of examples of this.
Recently, parents at a school board meeting by, I guess it
was, a National School Boards Association (NSBA) said these
were domestic terrorists. They are not domestic terrorists.
These are parents trying to--and I think they later apologized
for saying that. But, parents expressing their strongly held
views about their kids' education.
You have to be sure that we are not taking content which is
people expressing political views that are peaceful and somehow
filtering those out. Any thoughts on that as we start, for any
of the panelists, either with us or virtually?
Ms. Kornbluh. I just want to agree with you. The algorithms
are not partisan. The algorithms are economically driven, as
Dr. O'Neil said, and they are trying to keep us all online. I
think it is really important to keep that in mind.
I think for the First Amendment, the freedom of speech,
freedom of association, it is extremely important that the
government not be in the business of deciding what is true and
what is not true and that instead--that is why I think some of
these revelations from the whistleblower are so important,
because she focuses upstream of the content, at the mechanics
of the platform, and how it is driving this content just to
service itself and to service its ad revenues.
If we focus on those design elements, and we focus on
transparency especially, that furthers First Amendment
concerns. It furthers freedom of speech and freedom of
association. Transparency is such an important principle, if we
let consumers know who is behind what they are seeing, what
interest is it serving, that----
Senator Portman. That leads me to my core question today
really, again, just stipulating this is a hard area. We have
talked about a couple of those issues.
But it seems to me that, as I said earlier about getting
under the hood and looking into what these design elements are,
that you talk about the transparency issue. What you talk about
is really important because--again, it is proprietary
information. I understand that. These are private companies.
Again, this is not an easy issue for government to be
involved with, but everybody is talking about regulation right
now. I mean, everybody. Facebook is talking about it. Google is
talking about it. Twitter is talking about it. We are talking
about it. Everybody has a different view what that regulation
might be, but shouldn't it be based on better data because we
really do not know what we are trying to regulate, if there is
a lack of transparency, as to what that design is or how these
algorithms are derived.
So you know, we talked a little again in your testimony
about this. Dr. Persily, in particular, you talked about your
thoughts on how to give access to impartial researchers to be
able, I assume, to publish about what is actually behind all
this. What is the content-directing mechanism and how does it
work. I am intrigued by that. I do not know if that is the
answer. You mentioned that the FTC could play a role in this.
But can you amplify that a little bit and talk about what
you think could lead us to more transparency and better
understanding?
Mr. Persily. Thank you for that question. The model that I
put out there is that the FTC, working with the National
Science Foundation (NSF), would vet researchers who would not
have the permission of the company, but the company would have
to basically develop research portals for outside, independent
research to study all of these societal problems that we are
saying are caused by social media. The key features of this are
simply that the firms have no choice in who has access to the
data and we have some way of vetting the researchers to
prevent, say, another Cambridge Analytica and the like. We need
to have some process in place so that someone other than those
who tied to the profit-maximizing mission of the firm get
access to this data.
As you mentioned, it is proprietary data, but it is data
about us. Right? It is essentially data about everything that
is happening in American politics and, frankly, around the
world. We need to figure out some way for the firms to be
opened up so that outsiders can see it.
My view is that they should not turn the data over to a
government agency, that that would pose real privacy and
surveillance problems. We want to make sure that there is a
vetted, independent third party that is able to do this kind of
research, on all of these questions that have come up today, so
that we can get to the answers to some of the questions that
you asked earlier about the propensity on the left and the
right to engage in hate speech or engage in violent content and
the like, as well as potential bias in the content moderation
platforms and the like. Only if we have access to the data can
we really answer those questions.
Senator Portman. Otherwise, very hard to come up with
regulation, which is what, again, everybody is talking about. I
know there are different views on what that means, but it seems
to me that there should be a consensus that if we are going to
try to regulate this we need to have better information as to
what the actual design and what the intentions are and the
impact is. So this hearing is helpful, I think, in that regard.
I am at the end of my time. Hopefully, we will come back
for a second round. I have so many other questions for this
team. But again, I thank you for your expertise today.
Senator Hassan [presiding]. Thank you, Senator Portman.
OPENING STATEMENT OF SENATOR HASSAN
it is now my turn for a round of questions. I want to thank
Senators Peters and Portman for holding this hearing and to all
of our witnesses, both in the room and virtually. I really
appreciate your testimony. This is an excellent panel, and I
appreciate your work very much.
I want to start with a question to Dr. O'Neil and Dr.
Persily. I am concerned that extremist groups, including ISIS,
continue to develop and refine online radicalization techniques
that make it easier and quicker to radicalize people. At this
Committee's annual threats hearing in September, leaders from
the Department of Homeland Security (DHS), the Federal Bureau
of Investigation (FBI), and the intelligence community (IC)
expressed similar concerns. Extremists take advantage of the
speed of social media platforms but also their algorithms,
which often facilitate the quick spread of extreme content.
Dr. O'Neil and Dr. Persily, in your view, what are the
weaknesses in the algorithms used by large social media
platforms that can make it possible for anyone, including
violent extremist groups, to expand and capture their audience?
What steps can social media platforms take to curtail extremist
efforts? Why aren't these companies taking these actions
already? I will start with you, Dr. O'Neil.
Ms. O'Neil. Thank you for the question, yes, and it is an
important one. I want you to think of the filters, these things
that are trying to catch content like that is hateful or
otherwise not allowed, as sort of nets that fishermen use in
the ocean. They sort of pull the net, and they see what they
have got. They have some fish there, and they count the fish.
They say, oh, we got a lot of fish. What they are not counting,
of course, is the fish that got through the holes of the net.
What I am talking about are the people who are paid,
actually, to put hateful content on Facebook and elude the
nets. The thing about is that they can tell when they have been
caught, and then they will double, redouble their efforts to
change their content somewhat so that it gets through the net.
It is kind of like if you think about, the early spam on
Viagra that would filter into your e-mail, and then you know,
the spam filters got rid of the things that said Viagra. Then
they started saying Viagra, but it was spelled--instead of with
an ``I,'' it was spelled with a ``1.'' Those got through for a
while until they did not get through.
Spam filters work pretty well to remove Viagra ads in part
because Viagra is the same word over many years. But in the
case of social media, the stuff that they are putting on social
media changes very quickly, and like the spam filters
essentially cannot keep up.
I think you should think of it as an arms race. It is the
filters on the one hand owned by the social media, and of
course, the people that get sent the high scored, like high
risk content----
Senator Hassan. Right.
Ms. O'Neil. Then they have to decide whether it is in fact
against policy, and then on the other hand, all the
propagandists who are actively trying to evade the filters. The
simple truth is that the propagandists are winning that war.
To the extent that social media companies can combat it
more, it would require much more expensive work, and they
simply do not want to do it. So their policy has been, we are
going to count how many fish we got, we are not going to count
how many fish we did not get, and we are going to hope that it
sounds good enough for you guys to stop asking questions.
Senator Hassan. Thank you.
Dr. Persily.
Mr. Persily. So this is one of those areas where I wish I
had the answer, and in order to get the answer we do need to
have access to the data. I will say that, having talked to many
of the integrity teams at these companies, I mean, these are
pretty sophisticated operations----
Senator Hassan. Right.
Mr. Persily [continuing]. That they have stood up in the
last five years, and there is more that they could do.
As bad as things may be in the United States, by the way,
they are a lot worse around the world, and that is something
that I think naturally we are focusing on problems unique to
the United States. But if they do not have the competencies in
the local languages around the world, especially if you are
dealing with terrorist content and the like, then they are
going to be hindered in their ability to really attack these
problems.
Again, if you look at the work that we have done at
Stanford, if you go to the Stanford Internet Observatory, we
have been trying on the outside to flag this kind of content,
the violent content, terrorist content, and foreign election
meddling and the like. We have been trying to do on the outside
what they do on the inside, but it is extremely difficult.
Senator Hassan. OK. Thank you.
Dr. Franks, the proliferation of nonconsensual intimate
imagery, sometimes called ``revenge porn,'' is a pervasive
problem on the Internet. There are a number of truly despicable
sites dedicated to hosting that material, but often users also
share these private images and videos on large social media
platforms with absolutely devastating consequences for those
whose images are shared without their consent.
Congress and the States have taken notice of the tremendous
harms from these situations, and there is some work going on to
address the problem. But, what more should social media
companies be doing to prevent this content from being hosted on
their platforms and to more immediately remove it when found?
What additional tools could these companies give to people to
help ensure that their images are not shared on social media
platforms?
Ms. Franks. Thank you. Social media companies--I will say
this about some of the more dominant companies--have been
trying to tackle this issue, Facebook among them. There are
teams at all of these various companies that are quite
concerned about these issues and have worked with organizations
like mine, the Cyber Civil Rights Initiative, to think about
ways to impose their content policy restrictions and to
encourage people not to participate in this kind of abuse and
to try to empower victims to be able to remove the content.
That being said, those kinds of measures are essentially
putting victims at the mercy of these companies. They may
choose to make this priority. They may choose to impose some
sort of policies. But there is not any necessary reason for
them to continue doing so, and tomorrow they could simply stop.
I think it really is important for the work that is being
done by the State legislatures and by Congress. The SHIELD Act
is included in the Violence Against Women Reauthorization Act
of 2021. That would federally criminalize nonconsensual
pornography. That would be a real incentive, I think, for these
companies to take it seriously.
This is connected to what I have suggested about Section
230, and it also goes to the broader question of incentives,
about transparency or about regulation, about any of these
questions. So long as these companies enjoy essentially blanket
immunity for these harms, there is no real incentive for them
to do anything, and therefore, it would be very important for
there to be changes in Section 230 to take away some of that
preemptive immunity.
Senator Hassan. Thank you very much. Again, I appreciate
the testimony of all of you.
I am going to recognize Senator Johnson now for his round
of questions.
OPENING STATEMENT OF SENATOR JOHNSON
Senator Johnson. Thank you, Madam Chair. First of all, let
me say anybody who has watched the documentaries, The Social
Dilemma and The Creepy Line, I think has to understand that
this can do great harm, particularly to children. I think as a
result I think our first line of defense is with parents. I
think that they need to really do everything they can to make
sure that their children are not affected by this in a negative
fashion.
Professor Franks, you made the comment that it seems like
from your standpoint most of the bad content leans toward
conservative and that type of misinformation. I would just ask
you, what is your assessment in terms of the leadership, the
people who work in these companies? How do they lean
politically in your assessment?
Ms. Franks. I am not sure that it is relevant. I think
the----
Senator Johnson. Pardon?
Ms. Franks. I am not sure that that is relevant. The
particular individual leaning----
Senator Johnson. OK. I will just determine what is a
relevant question or not, but let me just ask you again, what
political affiliation do you think most people in leadership
positions and that work in whether it is Facebook or Google or
Twitter? Are they right-wing extremists? Are they conservative?
Are they liberal? Are they radical leftists? I mean, what part
of the political spectrum do you think they fall on?
Ms. Franks. Perhaps you could explain to me the relevance,
and then I could answer the question.
Senator Johnson. Can you speak a little bit louder?
Ms. Franks. Sure. Perhaps you could explain to me the
relevance of political affiliations of individual employees.
Senator Johnson. I was saying, just listening to the
testimony here, it seems like the big concern here is about
right-wing extremism, which I completely condemn, or right-wing
misinformation. So again, I would just argue when you take a
look at what Mark Zuckerberg has done through his Center for
Tech and Civic Life, a couple hundred million dollars spent,
pretty well took over the election system--I think in violation
of Wisconsin law--of Green Bay and three or four other cities,
it does not seem to me that the impact or the intent of their
manipulation of data would tend to favor conservative groups or
conservative thought. It seems to make more sense that they
would tend more to push a liberal ideology.
Ms. Franks. As other experts have testified, the main issue
for most of these companies tends to be profit, and profit is
usually going to be built around engagement. Engagement is
usually going to be built around outreach, misinformation,
half-truths, things that provoke people into thinking that they
are under attack, that they are being victimized.
Senator Johnson. OK. We heard that in testimony. Let me ask
you, have you heard of the work of Dr. Robert Epstein?
Ms. Franks. I am not sure.
Senator Johnson. So I saw one of you shake your head. Is
that Mr. Persily?
Mr. Persily. Yes. I am familiar with it.
Senator Johnson. You are familiar with it?
Mr. Persily. Yes.
Senator Johnson. Can you summarize it, or should I
summarize it for you?
Mr. Persily. Whichever you prefer, but, yes, that he makes
the argument that the platforms favor the Democrats in his----
Senator Johnson. He makes the argument that through
manipulation of search, Google, as it ramps up toward
elections, starts manipulating the search to push users of
Google into the type of information that is going to tend to
have you vote or decide to vote for a Democrat, delivering,
according to him, millions of votes to Hillary Clinton,
millions of votes to congressional Democrats in elections in
2018.
Mr. Persily. Yes, that is what he says.
Senator Johnson. OK. I think the point I am trying to make
here is I think manipulation is potentially going to swing both
ways. Who is going to be the arbiters of truth? You know?
Professor Franks, you said that the reckless endangerment
of democracy. Now I would happen to think that if one of these
platforms is utilizing their awesome power of manipulating a
search to turn votes toward one political party versus the
other, if one of these tech giants uses--spends hundreds of
millions of dollars to turn out voters in Democrat-leaning
areas and regions, that certainly impacts our democracy.
I will give you another example in terms of Facebook. I
provided a forum for people who were vaccine injured to just
tell their story. Following that, there were about 2,000 people
involved in groups supporting each other. Some of these women
have such severe vibrations from vaccine injuries that they
committed suicide. So this Facebook group was a support group.
It was literally helping people prevent suicides. Within a
week, their group grew from 2,000 to 5,000 people, at which
point Facebook dismantled the groups. These individuals who
were counseling and helping prevent suicides lost contact with
the people who were suicidal.
So who is going to regulate this? How is free speech
different when it is on a platform versus when it is just
spoken in the town hall?
Ms. Franks. I would suggest----
Senator Johnson. Who is going to be the unbiased arbiter of
truth? I do not think it exists. I am certainly questioning the
Section 230 liability protections when you have these platforms
acting as publishers, which is what they are doing when they
censor primarily conservative thought. I have been censored
myself, repeatedly.
So again, I am just pushing back and challenging the fact
that this is something that is fomenting right-wing
conspiracies and highly advantageous to the conservative
movement. I would say, if anything, it is more likely it is,
from a political realm, advantaging left-wing ideology.
But again, I will come right back to we have a constitution
that protects free speech. Who is going to regulate that
fairly, in an unbiased fashion? It is just not possible. Along
the way we are violating people's constitutional rights.
Anybody want to just take a stab at that one?
Ms. Franks. I would be happy to respond to that. Yes, we do
have a First Amendment. We do have right to free speech.
But we also know, of course, that private companies are not
obligated under the First Amendment to take all comers. They
are allowed to make their own decisions about what is
considered to be high quality or low quality content. They can
make any number of decisions, and I think we would applaud them
in many cases to make those decisions. As we were talking about
just before, in terms of nonconsensual pornography, I, for one,
am very happy that Facebook has made the decision to say that
that is not welcome on their platform.
When it comes to the questions of conservative versus
liberal bias, this is not a preconceived notion that I am
suggesting here. This is not about intuitions or impressions
although I know that those can go in many different directions.
This is about what the data actually suggest. The data actually
do indicate that right-wing content is more amplified on these
social media platforms than left-wing content and that right
wing content is more disproportionately associated with real-
world violence, not hurt feelings, not people being upset, but
in fact actual violence, actual armed insurrections, actual
notions of terrorism and anarchy.
Senator Johnson. Thank you, Mr. Chairman.
Chairman Peters. [presiding]. Thank you, Senator Johnson.
Senator Ossoff, you are recognized for your questions.
OPENING STATEMENT OF SENATOR OSSOFF
Senator Ossoff. Thank you, Mr. Chairman.
Ms. O'Neil, based upon your experience reviewing the
algorithms underpinning many of these platforms and similar
products, can you please connect the dots for the Committee,
the link between the scale that these companies manage to
achieve and the algorithms that they use to feed content to
targeted users?
Ms. O'Neil. Thank you for your question. I am going to make
a confession. I have not audited these algorithms because these
companies have not welcomed me or invited me to--inside their
systems. They would not want me there.
For that matter, I do not think I would take that job for
the very reason you are asking the question. As an algorithmic
auditor, I want to consider who would be harmed, who are the
stakeholders, and what kind of harm would come to those
stakeholders or might come to those stakeholders. I do not go
in assuming there is harm, but I do go in thinking about who
are the stakeholders and what could harm them, and then making
and developing tests and experiments to see the extent to which
these harms are actually occurring.
For example, the kind of research we learned about from the
whistleblower around teenage girls and suicidal ideation would
be the stakeholder, teenage girls, and the harm would be
suicidal ideation caused by their experiences on Instagram.
That would be an example of the kind of stakeholder and harm
that I would examine.
But if I were to actually be given the job of auditing the
Facebook algorithm or the Instagram algorithm or any of the
other algorithms, it would just be too large. There would be
too many stakeholders. For example, it is very clear to me that
if I had been given that job four years ago I never would have
imagined the stakeholder that would be the Rohingya Muslims who
were going to be there was going to be a genocide against them
in Myanmar. That simply would not have occurred to me.
So it is just too big a job to do that, and it is because
of scale, because it is international and because even within
the United States it is too large to imagine who are the
stakeholders and what kind of harm could befall them.
Having said all that, there are specific stakeholders that
you can imagine right now that are interested--that are the
focus of this particular Committee, that you could be saying,
well, wait a second. Are these stakeholders having these harms
that are actually illegal or a national security threat? That
is a kind of algorithmic audit I would be happy to do. It would
not be something that Facebook would invite me to do. You would
have to somehow subpoena the data for me on my behalf, but that
is something I could do.
To summarize, auditing for me is a stakeholder and a harm
and an experiment to see the extent to which that harm is
falling on that stakeholder. I would be happy to do that, but
you would have to choose a few of them because there is just
almost an infinite number to consider a priori.
Senator Ossoff. OK, Ms. O'Neil. Thank you.
Dr. Franks, having observed this hearing, paying close
attention to the broader discourse in politics, culture,
society on these issues, what do you think is being missed by
policymakers, and how is the nature of our debate perhaps
overlooking key considerations or facts relevant to the policy
discussion?
Ms. Franks. Thank you. I think a lot of what is being
missed is--or I should say the focus is sometimes on purposes,
intentions, listening to the companies say, we are working on
this; we wish we had caught that. I really think it is long
past time that we look beyond what the companies are saying
they care about and what they are intending to do and simply
look at what is happening, that the question of intentions just
be something we leave in the past. This is why I suggest, for
instance, that Section 230 is ripe for reform because it gives
too much deference to the idea that this industry will be able
to regulate itself.
I think the other important issue is to recognize that
certain types of changes to Section 230 need not raise or
settle all of the First Amendment concerns that have been
brought up, that are, of course, legitimate to bring up and be
concerned about, that modest reforms to Section 230, for
instance, denying immunity when it comes to harm that is
caused, that is foreseeable, and to which intermediaries have
exhibited deliberate indifference.
All that would do would be to allow people to be able to
sue these companies if they had a theory. It does not mean that
they would be vindicated. It does not mean that some of that
speech would not ultimately found to be First Amendment
protected.
It means that the industry would not continue to have this
kind of preferential treatment that rarely any other industry
really has. They would be called to account. They would have to
reveal documentation. They would give us some transparency
about the inner workings of what these companies actually do.
They would allow, at least in some cases, for people who have
been harmed to prevail and to actually get some kind of
compensation for their injuries.
Senator Ossoff. Thank you, Dr. Franks. What information
about the business practices of these firms that may not
currently be public do you think would be of the greatest value
to Congress as we weigh potential statutory revisions?
Ms. Franks. There are any number of things that I would
particularly be interested in hearing what these companies are
doing, but just to take a few examples, when companies
implement policies against certain types of harms and say that
they now have removal policies, let's say, with nonconsensual
intimate imagery: What is the data in terms of what kinds of
reports they are getting? What is the data in terms of whether
they are taking those requests seriously? How quickly are they
responding to those requests? How often are they aware that
those types of material are flourishing on the platform and
increasing engagement? How many times are they willing to,
nonetheless, hold to their principles and take it down as they
have said that they would?
In essence, are they actually fulfilling the promises that
they are making to the public?
Senator Ossoff. Thank you, Dr. Franks. Thanks to the panel.
Mr. Chairman, I yield back.
Chairman Peters. Thank you, Senator Ossoff.
Senator Rosen, you are now recognized for your questions.
OPENING STATEMENT OF SENATOR ROSEN
Senator Rosen. Thank you, Chair Peters, of course, Ranking
Member Portman, for holding this very timely hearing.
To our witnesses, I appreciate you being here and sharing
your expertise with us.
I want to talk a little bit about algorithms because the
tools of violent extremism, like conspiracy theories and
disinformation, they frequently begin online as building blocks
of hate. As we have seen time and time again, hateful online
words can morph into deadly online--offline actions, excuse me,
and then amplified once again online. This is a vicious cycle.
This was the case three years ago yesterday at the Tree of Life
Synagogue shooting, and it is true in far too many cases of
hate.
As we have been discussing today, what often enables
extremist groups and individuals to disseminate hate messaging
are social media algorithms. Platforms generate algorithms that
promote content, even if it is harmful, to keep people engaged.
And engagement is what drives advertising revenue.
We have learned recently that social media platforms have
known for some time--and I am going to quote here--``hate
speech, divisive political speech, and misinformation,'' on
their apps are having a negative impact on society and that--
again I am going to quote--the core product mechanics, such as
virality, which means how viral someone goes or some post goes.
Those recommendation algorithms optimizing for engagement, they
are a significant part of why these types of speech flourish.
So as a former computer programmer, I know that platforms
have the capability to remove bigoted, hateful, and incendiary
content that leads to violence, and I also know that they have
an obligation to do so.
So to Mr. Sifry and then Ms. O'Neil, when a platform
announces a new policy banning hate content, do you know how
often do they or should they adjust their algorithms to reduce
this content and, more importantly, to reduce and remove the
recommendation algorithm for hate content so they do not
continue to spread like we see that?
Mr. Sifry. Senator, that is an excellent question. Thank
you so much. The core issue, right, that these policies that
they are creating--and many of the large tech companies have,
on the face of it, admirable policies against hate, against
incitement to violence, and against harassment. However, the
issue is enforcement at scale, and what we have seen over and
over again is that the platforms are falling down at being able
to enforce at scale.
For example, for nine years, Facebook did not have an
official policy on Holocaust denial. They allowed this content
to stay up on their site with no policy. Last year, after years
of advocacy by civil society groups, including ADL, they
finally changed their policy and said, OK, this terrible,
nefarious content needs to be off the platform.
Yet, in January of this year, just 3 months later, when you
would expect they had time to actually enforce said policy, we
went and tested that policy. We were able to find groups with
tens of thousands of members that were still advocating
Holocaust denial, and we were able to find these kinds of
nefarious content still being pushed to the tops of people's
feeds.
So you are so right in the sense that not only the policies
are, of course, important, but it is how do you enforce these
policies at scale and how these algorithms, because of these
engagement mechanisms, will then push them to the top of our
news feeds right next to content from our friends and family.
Senator Rosen. Thank you. Mr. Sifry, what do you think we
can do to reduce or remove these algorithms? I am sorry, I
meant Ms. O'Neil.
Ms. O'Neil. Yes. Thank you.
Senator Rosen. I was looking at Mr. Sifry but seeing you on
the screen. I knew what I meant. Sorry about that.
Ms. O'Neil. Thank you, Senator Rosen. It is a really good
question and an important question. I personally, and probably
you as well--when I heard Mark Zuckerberg say a few years ago
that AI is going to solve this problem, I knew that was a lie,
and he knew that was a lie. AI does not have a notion of truth.
It does not understand the English language or never mind other
languages. It simply looks for keywords. It is like a little
bit of a gussied-up version of a keyword search.
So the point is that if we cannot decide what is true,
right, AI can definitely not decide what is true. Its track
record is so very weak. We have been hearing about less than 1
percent of certain types of violent content being caught by
this particular type of AI. I hesitate to call it AI because
that is giving it too much credit. It is just an algorithm that
is a filter.
I also want to say that I do not have any confidence that
this will work better in the future. The reason is that,
whereas the propagandists are working with the recommendation
engine, should get more attention because they want more
attention--so you can say that it is like they are working with
that algorithm--they are working directly against the filtering
algorithm. They are trying to bypass it, and they are very good
at that. The filter is just a very weak thing, and it will
never work as far as I am concerned.
Senator Rosen. I want you to go on. We think about
enforcement at scale, some of the things you are talking about.
What can we do here in Congress to really be sure that we can
propose guardrails there for this misinformation, hate, things
that promote violence in the real world and then amplify that
violence and celebrate it after it happens? What can Congress
do to help stop this?
Ms. O'Neil. I think I am going to go back to the first
person who spoke--I believe her name is Karen--that it is about
incentives. Right? So the thing that I suggest should happen
would be for you guys to have a specific definition of the kind
of harm that you want them to keep an eye out for and that you
have a way of making sure it is happening, a monitor, and every
single time their monitor fails they get charged money because
it is always going to be about money.
So in other words, you have to make it more expensive for
them to let it happen than the profit that they gain by letting
it happen. You have to really counter the profit motive, and
that is the only way you can do it.
Of course, it will be very expensive for them to stop these
kinds of things. It will not work with AI. They will have to
put humans on it, and they do not want to do that.
Senator Rosen. Right. Thank you.
[Simultaneous discussion.]
I was going to say, Mr. Sifry, do you have something to
add? Because I really want to work to stop this escalation and
celebration and this vicious cycle. So, please.
Mr. Sifry. Absolutely. What is so critical here is it
starts with transparency. So No. 1, what are the actual
policies that these platforms actually have? Facebook had a
policy that they never reported on, called Cross Check, where
over five million people were essentially absolved from any
kind of machine review, and this included a virtual whitelist
where celebrities, politicians, and others could say whatever
they wanted without review.
Then for each of the policies they have, what are the
enforcements that they have done? We expect these kinds of
reports when companies go public and go to the FCC, and we
expect it, and we penalize them when they lie. We should be
doing the exact same things with these companies that are so
vital to the future of our democracy.
Senator Rosen. Thank you. I appreciate both.
I have some other questions for the record. Mr. Chairman, I
will be submitting those. Thank you for your time today.
Chairman Peters. Thank you, Senator Rosen.
Senator Lankford, you are recognized for your questions.
OPENING STATEMENT OF SENATOR LANKFORD
Senator Lankford. Mr. Chairman, thank you very much.
Thanks for all the testimony. This is an ongoing national
conversation, obviously. We do not have an anticipation we are
going to solve it all today in this hearing, but we do have an
anticipation we are going to narrow down what are the key
things that we are going to engage in and that we have to be
able to find a way to be able to solve. So it is helpful to be
able to have your dialog in the conversation.
Five years ago, I was with an executive from a Silicon
Valley company, which I will leave out the name of the company,
and as we were chit-chatting I made just some random comment
about his social media page. he looked at me point-blank and
said: Oh, I do not do social media at all, and my children do
not do social media at all. I know better because I know what
it is and what it is designed to do. I would never do that.
It was stunning to me to hear someone in the middle of
Silicon Valley to say: Oh, I know how toxic this is. There is
no way I am going to allow my children or my own family to be
affected by this.
For some reason, that message has not gotten out to a lot
of other folks, and for some reason there are folks that are in
social media companies that have no personal challenge with
being a part of that. The original sale of this seems to be we
are going to swap pictures of our children and what we are
eating for dinner, and it has moved to something very
different.
So saying all that, I want to talk a little bit about ways
that it could get better, practical aspects. Some of them were
just mentioned online, in some of the conversation just now.
What are some practical things that could be done to make this
platform, this type of platform, better to be able to engage?
Anyone who wants to jump in with a practical idea, you are
welcome to.
Mr. Sifry. Senator, if I may, so we have talked a lot about
changing the incentive systems that then get these companies to
change their algorithms. I think, No. 1, we have been looking
at Section 230 reform in particular targeted ways. But this is
a complex issue. We need to be careful about how we do so and
how to, in fact, incentivize companies to maintain--they will
keep the 230 protections if they actually reform and behave
more responsibly.
Second, we have talked about transparency reporting, a
number of us have, and I think that this is something that
again has real First Amendment considerations here. This is
something that we should be asking these companies to do tout
de suite.
Third, ADL has been bringing forward the idea of a resource
center, an independent resource center much like the National
Center for Missing and Exploited Children (NCMEC), where we
have for the exploitation of a child and sexual imagery. It is
a public-private partnership to help to track what is going on
in the world of extremism as well.
Most importantly as well, academic and other good faith,
third-party researcher access to this data so that they can
actually be looking at what is going on and report back on what
is happening beyond what is just mandated.
Senator Lankford. OK. Other specific issues?
Ms. Franks. I do think that Section 230 reform is going to
be one of the most important planks of this. Changing
incentives for the industry is of the utmost priority. If
companies are essentially given a blank check, told that they
can engage as much as they want, as often as they want, and
take profits from that, that they will face no repercussions
for this, that they are immune sort of preemptively from suit,
that is something that gives them a terrible incentive to do
anything responsibly. So that would be one way.
Senator Lankford. Is this 230 reform, or is it 230
enforcement because the issue about not being an editor or
publisher? Clearly, multiple of these entities are already
trying to be able to alter content and to be able to engage in
something that looks like a violation of 230 already. If there
is a 230 issue, is it enforcement, or is it changing the way
that it is written?
Ms. Franks. The 230 issue I think can be thought of in
either way. There are ways to read 230 more in a limited
fashion, but I do think that the primary problem with Section
230 currently mostly has to do with the (c)(1) provisions which
provide immunity for things that the companies are leaving up.
Senator Lankford. Right.
Ms. Franks. I think if we were to open the door to that
they would be subject to the same kinds of litigation that
other industries are a part of, we would see more safety and
better standards in addition to, I think, far more FTC
interventions against what some of these companies are doing,
recognizing that there is a fundamental misalignment of
incentives here because these companies, their customers are
not the users; their customers are advertisers.
Therefore, we have a big problem when it comes to how much
those incentives are being thought of as advertiser issues as
opposed to user issues----
Senator Lankford. Right.
Ms. Franks [continuing]. How often users believe somehow
that they are getting a free service. There needs to be more
oversight from the FTC about what that means for consumers and
what it means for their consent.
Senator Lankford. With that, how do you balance out the
issue of censorship? Because Focus on the Family and the
Heritage Foundation and countless other conservative groups or
faith based groups would tell you they put up content and
immediately it gets pulled down and blocked or they get their
account suspended. They are saying, ``we are trying to keep,
hateful content away, and typically, hateful content is
conservative content''. And so they are saying, ``we are trying
to abide by that''. How do you balance that out?
Ms. Franks. I think there again we have to underscore the
fact that these companies do have a First Amendment right of
their own to exclude content and not associate with content
that they find objectionable for any reason.
Senator Lankford. Right.
Ms. Franks. Therefore, that is the part of Section 230, the
extent to which it gives them procedural rights as far as that
enforcement goes, that I think we probably should leave alone.
Senator Lankford. We are, on the other side of this then,
dealing with antitrust violations, where you have an issue
where a company intentionally goes in and blocks out
competition to be able to make sure there are no other voices
than them in the marketplace and only their point of view gets
out. That leaves the First Amendment issue, which I agree with,
and moves into an antitrust issue that we still have to be able
to resolve for several things.
Mr. Persily, do you have something to add to that?
Mr. Persily. Certainly. Thank you. Let me begin where you
left off, which is so we were thinking about the antitrust
problem in a very classic way, and we realize that these firms
do not fit the normal model of antitrust. Let us talk about it
in terms of competition law or the like. There are a lot of
measures on interoperability and trying to break up their
monopoly on data that I think Congress should explore.
Also, there are sort of non-speech-related reforms that I
think will have an impact on these speech-related issues,
right, in a beneficent way. So privacy legislation, right, is
tied to advertising regulation because the amount of data that
they hoover up from all of us that is what enables certain
types of targeting, certain types of messaging which a lot of
us here today were complaining about.
Then on transparency, everybody is in favor of
transparency. What I want to emphasize--and Mr. Sifry mentioned
it before--is researchers or just getting some third party in
there to figure out what is happening, whether it is on the
political bias, on content moderation or on the actual nature
of the platforms, that this will change their behavior if they
know that they are being watched. Right?
It is not just about providing a subsidy to outside
researchers to figure out something for their publications. It
is about making sure someone is in the room to figure out sort
of what is going on and that the researchers are not the only
ones in the firms who are tied to the profit-maximizing mission
of the firms, who then have access to that information.
Senator Lankford. Yes. Mr. Chairman, it is interesting. I
brought to Facebook several years ago now just an idea, to say,
if you want to turn down the volume in some of these pages
where there is people attacking other people on the page, allow
the opportunity for the page owner, quote-unquote, to be able
to just say, the comments just come to me, but they are not
public. If you want to comment to me, you can comment to me.
But then that discourages people from attacking each other back
and forth.
Their response was, well, we like the interaction.
It is interesting. The preference for them was very clearly
we like people hating on each other on this, even on pages
where they know that is the dominant theme, because it helps
with advertising dollars. It helps with the income side. These
platforms are very aware of methods to be able to turn down
volume, turn down hateful rhetoric, but they choose to be able
to leave it up for advertising dollars.
I do think that is something we need to continue to be able
to engage in, to be able to show how this could be better, that
you can get back to a point of sharing pictures with loved ones
and what you had for dinner, though I have no idea why people
care what picture you had for dinner. But there is a way to be
able to turn down volume, and we expect platforms to be able to
step up and do that.
Mr. Chairman, thank you for holding this hearing.
Chairman Peters. Thank you, Senator Lankford. Clearly, a
lot of work ahead for us. I agree with you.
Senator Romney, you are recognized for your questions.
OPENING STATEMENT OF SENATOR ROMNEY
Senator Romney. Thank you, Mr. Chairman. I, like all
Americans, are very concerned about what we are seeing in
social media, the impact it is having on our democracy, the
disinformation that we are seeing. I have grandkids whose
parents are wisely telling their kids not to get on social
media. They are not giving them smartphones until they get a
good deal older.
But frankly, as I have listened to the conversation this
morning, it strikes me that this is a story we have heard
before in a different context. The idea that there are
companies that are trying to give people what they want to
read, well, that is what newspapers do. There are companies
that are trying to let people hear what they want to hear; that
is what radio stations do. There are TV stations that have
something known as Nielsen, which tells them what people watch,
and they found more salacious things, I presume, get more
eyeballs.
I have heard on some of the nightly shows that they can
look at which guest comes in, what that guest is wearing, and
they can see their ratings going up based upon the gender of
the guest and what they are wearing.
I mean, this is not something that is just particular to
social media. It is something that is part of our entire media
system. The idea that media companies, like social media
companies, are trying to maximize their profitability, so is
TV, radio, newspapers, magazines. They are trying to stay in
business and maximize their revenue.
Disinformation. Have you ever read the National Enquirer or
the Star? I mean, those are out there. They are at the
newsstand. People can pick them up. There is all sorts of
disinformation.
How about political bias? How about Fox, MSNBC, New York
Times, and Washington Examiner? They have a point of view. They
have a bias.
And we say, people only have access to those sites.
Actually, there are a lot of social media sites out there, and
more are being introduced, and there will be more introduced
over time.
The idea that saying, hey, we are going to subject them--we
are going to break them up for antitrust reasons. Do not
forget, data and social media is international. It is not like
it is just a U.S. enterprise. TikTok is owned by the Chinese.
We are not going to break them up, even though some of us might
prefer that we could. We have to think about the competition
coming from around the world.
I must admit I have a hard time seeing how we are going to
change the algorithms. If we are going to change the algorithms
for social media, how about changing the algorithms for Fox,
MSNBC, the New York Times, NBC, and CBS? I mean, each of them
have their own bias. They have their own way of assessing how
they get the most eyeballs, how they get the most clicks, if
you will, even if it is just the channel changer. I do not see
how that works.
Likewise, to say we are going to set the standards for what
is truth and not truth, I do not know how government does that
consistent with the First Amendment.
I understand why a CBS decides what they are going to show
and why a New York Times decides what they are going to print.
That is their standards. By the way, their people can respond
to it, and people can boycott or stop subscribing as a result
of what they are saying.
This is an area, in my opinion, that is extraordinarily
fraught for government action. What we can do that does not
violate the freedom of expression, of which we are intent on
having in our country, is something which I think we have to be
careful in considering.
I would note that what is unusual about social media, among
many things, is the precision with which data is able to be
gathered and people are able to be targeted. The New York Times
may say, I want to get a left-of-center subscriber base, but
they cannot get it literally down to the home to print the
articles that you might want to watch although as they go
online, and increasingly they are an online forum, they will be
able to provide the articles that you want to watch and will
compete on that basis.
Anyway, I guess I raise a couple of things as possibilities
in a topic I know very little about but care very deeply about
because of my kids and grandkids. That is, one, to require
social media companies to certify that a person who is posting
or commenting is an actual person, so very simply saying you
have a responsibility as a social media company to determine
if, ``JAB123'' is actually a human being and also what country
they are from perhaps but whether it is a government or a
corporation but whether it is human being. That would be No. 1,
and that is probably the easiest constitutionally.
The other is something I would like to do. I would like to
require actually people to be identified as to who they are,
what their name is, if not their address, at least their name
so people are responsible for what they post and what they
comment on. I think that would fly in the face of the First
Amendment and the right to be anonymous.
Perhaps having more emphasis on ``blue checks,'' if you
will, verifying people and really insisting that social media
companies encourage more people to be identified and perhaps
even having an alternative site, which is, if you will, there
is Facebook where there is no blue check, but then there is
another portion of Facebook which is all blue check. To comment
or to post, you have to have a blue check on that particular
wave. That would give people some confidence that there is a
real human being that is willing to stand behind that comment.
Those are the only two ideas I could come up with that I
thought would be helpful here and not fly in the face of the
same challenge we have with all of our media, that there is
disinformation that we do not make illegal, that there is bias
that we do not make illegal, that there is profit motive that
we do not make illegal, that there are algorithms that draw
more articles that people want to read that we do not make
illegal.
I do not see the course that you all are talking about as
being consistent with what the rest of our media system is
providing.
Do any of you have any comments about those ideas that I
suggest? I am concerned about the topic, but I just cannot seem
to find a way to resolve some of the problems we describe.
Yes, Dr. Persily?
Mr. Persily. Yes.
Ms. Kornbluh. Can I just try real quick? Just because I
worked at Nielsen, so you are absolutely right, and I think it
is great that you are pulling us back to this history of
newspapers and broadcast. What is different, I think, is that
in the newspaper, first of all, speakers did not have this much
power. That is true in broadcast, too. We had ownership
restrictions. You could not own a newspaper and a broadcast in
the same market. You could not own----
Senator Romney. But you could have the whole country, like
Sinclair does.
Ms. Kornbluh. But you could not in the past. We tried to
deal with some of these issues but mostly through transparency.
Right? So there is a masthead.
Senator Romney. Yes.
Ms. Kornbluh. There are codes and standards that you hold
yourself to if you are a newspaper. If you are a broadcaster,
we had the payola rules so that the DJ would have to reveal if
he was being paid to play that record. The political ads that
you run, you had to be transparent and say, I paid for these
ads.
What we are seeing on social media, I think, is we have not
worked through those systems. We do not have that transparency.
That buyer cannot beware because they just do not know who is
pushing this at me, why are they pushing it at me. The codes
and standards that they proclaim they are not really honoring.
I have seen----
Senator Romney. Yes, the transparency, that is the
direction I am going is transparency. Yes.
Ms. Kornbluh. I think what you are talking about is really
interesting. I might tweak it a little bit and say, if you are
Facebook and you say you have a real name policy, then you
better have a real name policy. If you have a blue checkmark
and it says it is a human, it better be a human. If it is a
bot, it should be labeled as a bot.
Senator Romney. Yes.
Ms. Kornbluh. If it is a deepfake, you need to let people
know it is a deepfake.
Senator Romney. Totally agree.
Ms. Kornbluh. That is really consistent, I think, with the
way we approach--with the way the industry had norms and
standards in the journalist industry and in broadcasting, what
the public interest standard required. In fact, those kinds of
user empowering----
Senator Romney. That course makes sense.
Ms. Kornbluh. Exactly.
Senator Romney. Yes, sir?
Mr. Persily. I agree with that. Just in terms of the
differences with TV, which is that we know what Tucker Carlson
or Rachel Maddow is saying every night. There is a record of
it. We know who was able to see it. We do not know that with
speakers on the internet. Right?
That is where the transparency legislation would come in--
and I attached some to my testimony--because we need to figure
out sort of who saw what, when, and how. Right? With television
and other kinds of broadcast media and even some print media,
we can figure that out, but not with social media platforms.
Senator Romney. Yes. Thank you.
Mr. Chairman, thank you.
Chairman Peters. Thank you, Senator Romney.
Senator Padilla, you are recognized for your questions.
OPENING STATEMENT OF SENATOR PADILLA
Senator Padilla. Thank you, Mr. Chair.
Let us just jump right into it. When I was Secretary of
State of California, prior to joining the U.S. Senate, I saw
more than our fair share of bad actors seeking to discourage
communities, particularly communities of color, from exercising
their right to vote by gaming social media and exploiting gaps
in trusted sources and data voids.
Like many of you, I am alarmed by the recent Facebook
Papers and, in particular, their absolute failure at Facebook
to invest in integrity systems responsive to non-English
languages and cultures around the world and right here in the
United States as well.
While the focus of today's hearing is on social media, I
hope we can keep in mind the broader information ecosystem. We
need to equip our kids and neighborhoods with media and
information literacy skills. We need to address the collapse of
local journalism--and I will be asking a question about that
here in a minute--that are expanding news deserts across the
country.
Now this hearing is about platform design choices, and it
would be an oversight not to reflect on user data. User
personal data is indeed what fuels targeting on social media,
informing the content users see and how ads are targeted to
them. It is all driven by their data.
Now strong data privacy protections may help address some
of the unhealthy dynamics that we see online. We have been
talking about a national privacy law for a long time, and I
think it is time that Congress finally gets it done.
Question for Dr. Persily: How can a strong privacy law
reduce the risk of echo chambers, micro-targeting of
disinformation or exploitative advertising which targets
specific individuals or groups based on profiling?
Mr. Persily. Thank you. Thank you, my Senator, for that
question. It is good to see you. I hope to see you back in
California.
As I mentioned before, we tend to think of internet
regulation in two domains: One is on speech, sort of
explicitly, and Communications Decency Act (CDA) 230 reform is
one of them. Then kind of structural or infrastructure design
questions, and privacy would be one of those. But I actually
think that we need to start recognizing that they sort of bleed
over into each other.
I think you are right to point out that through national
privacy legislation and regulating the kind of data that the
firms can collect, that we will be able to get at some of these
problems because if you think that part of the problem here is
the micro-targeting of messages that necessarily select out
audiences for manipulation and persuasion and the like, that it
is only enabled because of the amount of data that the firms
have. If we had rules on what particularly the big platforms
could do in terms of collecting data, I think it would go a
long way in addressing some of these speech problems as well.
Senator Padilla. Thank you. Now on an extended topic
because I mentioned reporting of the Facebook Papers reveals an
abdication of responsibility to meet the needs of non English
speaking Facebook users around the world, and it is happening
here in the United States as well. We are blessed with a very
diverse population. Spanish-language disinformation about how
to vote, where to vote, when to vote, et cetera, ran rampant on
platforms in 2020 as compared to similar content in English.
It is not limited to just election information, by the way.
I saw it significantly when we were doing census outreach and
assistance at that critical time in 2020. We continue to see it
in regards to the COVID-19 pandemic, the safety of vaccines, et
cetera. It is absolutely unacceptable.
Question for Dr. O'Neil: Why do you think platforms are
failing even more for non-English language speakers, and in
what ways can Congress be helpful in this space?
Ms. O'Neil. Thank you for the question. You are absolutely
right that they are failing in non-English language spaces. In
India, which is a huge problem as we have read about in the
last few days, but for many months and years we have known
this, there are just too many language dialects. Facebook just
does not want to hire people to know those languages. It would
be very expensive.
It is a cost issue on top of the fact that I already
mentioned, that the filters for hateful or extreme content are
essentially keyword searches, so you need to understand what
keywords to search for. You need a lot of experts working full-
time on this, and they just simply do not want to pay for that.
It would be very expensive. So it is clear.
But I want to make it also clear that there is no simple
solution. I am not suggesting that they are avoiding doing
something simple to solve these problems. This is actually
really hard. Their mistake is not that they are not doing it
well. It is that they are pretending they can do it. They
simply cannot do it because of what I have said before, that AI
does not understand truth so it is just simply looking for
keywords.
To the extent that Facebook cares about looking good, they
care much more about looking good to English-speaking Americans
and to people like you.
I would say one quick story. I gave a talk in the Ukraine
recently, and one of the audience members was a parliament
member of the Ukraine. She said, what can we do here in the
Ukraine about Russian propaganda that undermines people's trust
in our elections?
I was like, wow, I really do not know. I mean, you have
even less power over Facebook than the Senators in the U.S.
Senate. It is a really important question.
Senator Padilla. I will just add to your commentary about
looking good in front of Americans or looking good in front of
people like us, members of the U.S. Senate. Sadly, looking good
in front of investors and Wall Street seems to trump it all.
My final question in the time remaining: We know today's
information ecosystem is complex. In addition to facing
organized propaganda campaigns, social media users encounter
more content at higher speeds. Right? The innovation technology
has a role to play here.
I worry that efforts to help communities critically engage
with information is not keeping pace. It is also not lost on me
that we have seen an explosion of propaganda campaigns aimed at
manipulating and intimidating communities online while we are
in the midst of a collapse of local journalism and independent
media.
My final question is for Ambassador Kornbluh. I welcome any
thoughts you may have on how the shuttering of local news
outlets has impacted how users engage with content that they
consume online.
Ms. Kornbluh. That is such an important question. As you
say, it creates this vacuum, and people are served things
online. Again, I think we have to underscore that so much of
what happens online is manipulation. People do not know what
they are being served or who is behind it.
There are these pretend local outlets that they see online
that seem to have a name that suggests that they are local, but
they in fact are often controlled centrally. The news stories
can even be constructed by AI. They think they are getting
local news, but they are actually being fed information that
serves a political interest or a financial interest, and they
are not aware of it. There is no alternative, so they do not
have access to the civic information that they need to be a
citizen.
The Secretary of State of Colorado just made a really
interesting point a couple of weeks ago. She said, if I am
standing up at a podium and having a press conference, and the
voters in my State are reading about, something completely
different, SharpieGate or whatever it is, online, I am not in
conversation with them because she is communicating to them
over these social media platforms and that is sort of a
funhouse mirror of what is going on.
I think we really have to think about how is the civic
information, public health information, election administration
information, how is it going to get to citizens at a time when
local news is so undermined.
I should say part of the reason local news is undermined is
because it was supported by advertising and all those
advertising dollars have now gone to the platforms. There is no
revenue base for local news. So this is a problem.
It is really a fundamental democracy problem. The press is
mentioned in the Constitution. It is something we really have
to address.
Senator Padilla. Thank you very much.
Thank you, Mr. Chair.
Chairman Peters. Thank you, Senator Padilla.
I would like to take this opportunity to thank once again
our witnesses for joining us here today. I think I speak for
the entire Committee when we say we appreciate your very unique
insight and expertise and helping us examine this critical
issue and navigate the tough challenge today.
Today's hearing provided an opportunity for us to learn
about the role social media platforms play in the amplification
of extremist content, including White Nationalist and anti
government ideologists or ideologies. We heard expert testimony
about how their algorithms and recommendation tools are driving
extreme content users, how that exposure to harmful content can
translate to real-world violence, and how their business models
built on user engagement and targeted advertising appear to
prioritize profits over safety.
The connection between extremist content on social media
platforms and domestic terrorist attacks in our communities is,
without question, a national security threat and one this
Committee will continue to examine. Our next steps will be to
include a hearing from the social media companies themselves,
and we will work to bring more transparency to this pressing
issue. I am also seeking information from both the Department
of Homeland Security and the FBI on this threat and will be
looking for needed reforms.
The record for this hearing will remain open for 15 days,
until 5 p.m. on November 12th, 2021, for the submission of
statements and questions for the record.
This hearing is now adjourned.
[Whereupon, at 12:12 p.m., the Committee was adjourned.]
A P P E N D I X
----------
[GRAPHICS NOT AVAILABLE IN TIFF FORMAT]
[all]