[Senate Hearing 117-782]
[From the U.S. Government Publishing Office]
S. Hrg. 117-782
ALGORITHMS AND AMPLIFICATION: HOW
SOCIAL MEDIA PLATFORMS' DESIGN CHOICES
SHAPE OUR DISCOURSE AND OUR MINDS
=======================================================================
HEARING
before the
SUBCOMMITTEE ON PRIVACY,
TECHNOLOGY, AND THE LAW
OF THE
COMMITTEE ON THE JUDICIARY
UNITED STATES SENATE
ONE HUNDRED SEVENTEENTH CONGRESS
FIRST SESSION
__________
APRIL 27, 2021
__________
Serial No. J-117-14
__________
Printed for the use of the Committee on the Judiciary
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
www.judiciary.senate.gov
www.govinfo.gov
______
U.S. GOVERNMENT PUBLISHING OFFICE
54-197 WASHINGTON : 2026
COMMITTEE ON THE JUDICIARY
RICHARD J. DURBIN, Illinois, Chair
PATRICK J. LEAHY, Vermont CHARLES E. GRASSLEY, Iowa, Ranking
DIANNE FEINSTEIN, California Member
SHELDON WHITEHOUSE, Rhode Island LINDSEY O. GRAHAM, South Carolina
AMY KLOBUCHAR, Minnesota JOHN CORNYN, Texas
CHRISTOPHER A. COONS, Delaware MICHAEL S. LEE, Utah
RICHARD BLUMENTHAL, Connecticut TED CRUZ, Texas
MAZIE K. HIRONO, Hawaii BEN SASSE, Nebraska
CORY A. BOOKER, New Jersey JOSH HAWLEY, Missouri
ALEX PADILLA, California TOM COTTON, Arkansas
JON OSSOFF, Georgia JOHN KENNEDY, Louisiana
THOM TILLIS, North Carolina
MARSHA BLACKBURN, Tennessee
Joseph Zogby, Chief Counsel and Staff Director
Kolan L. Davis, Republican Chief Counsel and Staff Director
SUBCOMMITTEE ON PRIVACY, TECHNOLOGY, AND THE LAW
CHRISTOPHER A. COONS, Delaware, Chair
SHELDON WHITEHOUSE, Rhode Island BEN SASSE, Nebraska, Ranking
AMY KLOBUCHAR, Minnesota Member
MAZIE K. HIRONO, Hawaii LINDSEY O. GRAHAM, South Carolina
JON OSSOFF, Georgia JOSH HAWLEY, Missouri
JOHN KENNEDY, Louisiana
MARSHA BLACKBURN, Tennessee
Sophie Brill, Democratic Staff Counsel
William Payne, Republican Staff Counsel
C O N T E N T S
----------
OPENING STATEMENTS
Page
Coons, Hon. Christopher A........................................ 1
Sasse, Hon. Ben.................................................. 2
Durbin, Hon. Richard J........................................... 3
WITNESSES
Bickert, Monika.................................................. 6
Prepared statement........................................... 43
Questions submitted with no response returned 75
Culbertson, Lauren............................................... 9
Prepared statement........................................... 48
Questions submitted with no response returned 84
Donovan, Joan.................................................... 13
Prepared statement........................................... 52
Questions submitted with no response returned 89
Harris, Tristan.................................................. 11
Prepared statement........................................... 59
Questions submitted with no response returned 91
Veitch, Alexandra................................................ 8
Prepared statement........................................... 61
Responses to written questions............................... 93
ALGORITHMS AND AMPLIFICATION: HOW
SOCIAL MEDIA PLATFORMS' DESIGN CHOICES
SHAPE OUR DISCOURSE AND OUR MINDS
----------
TUESDAY, APRILl 27, 2021
United States Senate,
Subcommittee on Privacy, Technology, and the Law,
Committee on the Judiciary,
Washington, DC.
The Subcommittee met, pursuant to notice, at 10:03 a.m., in
Room 226, Dirksen Senate Office Building, Hon. Christopher A.
Coons, Chair of the Subcommittee, presiding.
Present: Senators Coons [presiding], Klobuchar, Hirono,
Ossoff, Sasse, Hawley, Kennedy, and Blackburn.
Also present: Senators Durbin, Grassley, and Blumenthal.
OPENING STATEMENT OF HON. CHRISTOPHER A. COONS,
A U.S. SENATOR FROM THE STATE OF DELAWARE
Chair Coons. This hearing will come to order.
Thank you to all of our witnesses for participating today,
and a particular thank you to Mr. Harris, who I understand is
joining us from Hawaii where it is 4 o'clock in the morning. I
would also like to thank Ranking Member Senator Ben Sasse for
working with me to put this hearing together. I am truly
grateful we have been able to work together on this important
topic. It is too important to let it fall victim to the trap of
the typical partisan gridlock here in Washington. Thank you as
well to Chairman Durbin for attending today.
Generally, when people hear the term ``algorithm,'' you
might think of some very complicated mathematical formula or
piece of computer code. As many of us have become increasingly
aware, algorithms impact what literally billions of people read
and watch, and impact what they think every day. Facebook,
Twitter, YouTube, the three major tech companies represented in
today's hearing, use algorithms to determine what appears on
your screen when you open and engage with their applications,
and there is nothing inherently wrong about that.
With billions or even trillions of pieces of content to
choose from on each platform, it makes sense that they should
have a way to help us sift through what they think their users
are looking for and what we are actually seeking. Advances in
machine learning that made this technology possible have led to
enormous good in other contexts. Machine learning has driven
innovation across many industries from medical science to
public transportation, and has allowed companies to deliver
better services. Many have also recently argued this advanced
technology is harnessed into algorithms designed to attract our
time and attention on social media, and the results can be
harmful to our kids' attention spans, to the quality of our
public discourse, to our public health, and even to our
democracy itself.
What happens when algorithms become so good at
amplification, at showing you content that a computer thinks
you will like, that you or your kids or your family members end
up spending hours each day engaged, staring at the screen? What
happens algorithms become so hyper tailored to you and your
habits and interests that you stop being exposed to ideas you
might find disagreeable, or even so different from you--yours
as to be offensive? What happens when they amplify content that
might be very popular, but is also hateful or just plain false?
As I noted, Ranking Member Sasse and I worked on this
hearing, and one of the main reasons for that is because we
truly do not see these as partisan questions and do not come to
this hearing with a specific regulatory or legislative agenda,
but this is an area that requires urgent attention. As Mark
Zuckerberg himself recently put it, and I quote, ``When left
unchecked, people will engage disproportionately with
sensationalist and provocative content, which can undermine the
quality of public discourse and lead to civic polarization.''
If we are so polarized and angry and we can no longer hear each
other's points of view, then our democracy itself suffers.
As quaint as some might think it, Ranking Member Sasse and
I plan to use this hearing as an opportunity to learn about how
these companies' algorithms work, what steps may have been
taken to reduce algorithmic amplification that is harmful, and
what can be done better so we can build on that knowledge in
considering a potential path forward, whether voluntary,
regulatory, or legislatively.
I look forward to hearing from the representatives of
Facebook, Twitter, and YouTube, who have agreed to testify.
Each of these platforms has taken a number of measures in
recent years to curb some of the harms that algorithmic
amplification can cause. It is also my hope that these
platforms can build upon good practices, learn from each other,
and make a significant difference. We will also hear from two
outside experts, who can help us ask some bigger-picture
questions and to narrow in on some of the strategies and
tactics we could or should follow moving forward, including
whether and how legislation might improve the practices that
these and many other platforms use.
Thank you, and I am now going to turn to my Ranking Member,
Senator Sasse, for his opening remarks.
OPENING STATEMENT OF HON. BEN SASSE,
A U.S. SENATOR FROM THE STATE OF NEBRASKA
Senator Sasse. Thank you, Chairman Coons. Congratulations
on having a gavel for the first time in 6 years. Hopefully you
do not get to keep it long, but I have enjoyed the preparation
for this hearing with you and with your team. They have been
thoughtful to deal with, and I appreciate your opening
statement. I guess I should acknowledge the witnesses, too.
Thank you to all four of you. Mr. Harris, it is actually 85
degrees this afternoon in DC, so you did not have to avoid us
in Hawaii and have to testify at 4 a.m., but thank you for
participating in the pre-dawn hours there, nonetheless.
Chris, I want to applaud your opening statement. It is too
easy in DC for us to take any complicated issue and reduce it
immediately to heroes and villains, and whatever the regulatory
or legislative predetermined tool was, to then slam it down on
the newly to-be-defined problem. I think--I think you
underscored a number of really important points. The simplest
one is that algorithms, like almost all technologies that are
new, have costs and benefits. Algorithms can make the world a
better place. Algorithms can make the world a worse place.
One of the most fundamental questions before us as a people
is not, first and foremost, governmental, or legislative, or
regulatory, though those issues do exist. The first one is, in
the new digital economy or the attention economy, the old adage
holds that if a product is free, you are probably the product.
The American people need to understand, we, parents and
neighbors need to understand that we are being given access to
these unbelievably powerful tools that can be used for lots and
lots of good. In most cases, because it is free, there is
somebody who would really like to capture our attention,
shorten our attention spans, and drive us into often poisonous
echo chambers. Algorithms have great potential for good. They
can also be misused, and we, the American people, need to be
reflective and thoughtful about that first and foremost.
To the tech companies who showed up today and to those of
you who are also adjacent to the Silicon Valley conversation,
thank you for your interest and attention to this conversation.
I think it is very important for us to push back on the idea
that really complicated, qualitative problems have easy,
quantitative solutions. In some hearings that are not narrowly
on this topic, but other technology-related Big Tech hearings
that we have had over the course of the last 2 or 3 years in
this Committee, sometimes really hard, nettlesome problems we
have wrestled with, we have been told that as soon as the
supercomputers were better, they would solve these problems.
The truth is we need to distinguish between qualitative and
quantitative problems.
I appreciate the Chairman's perspective on the way we are
beginning this hearing, which is this is not a rush to pretend
politicians know a lot more about these problems than we really
do. It is an acknowledgement that there are some big problems
and challenges in this area, and prudence, and humility, and
transparency are the best way to begin. I am grateful for the
Chairman's leadership of this Committee and this particular
hearing.
Chair Coons. Thank you, Senator Sasse. I will now turn to
Chairman Durbin for his opening remarks.
OPENING STATEMENT OF HON. RICHARD J. DURBIN,
A U.S. SENATOR FROM THE STATE OF ILLINOIS
Chair Durbin. Thanks. I will be brief, and I appreciate the
opportunity to join you and Senator Sasse and make a statement.
Congratulations, Senator Coons, for taking the reins as Chair
of the Privacy, Technology, and Law Subcommittee, which I was
pleased to reconstitute in this Congress. You have already
demonstrated significant leadership. I look forward to your
work and the cooperative efforts of Senator Sasse in bringing,
I hope, some policy and legislation before the Full Committee.
This country stands at a crossroads as we grapple with the
role of technology and social media in our lives and culture. I
think Senator Sasse summarize it: it is plus and it is minus.
We are trying to look to the minus side, but should never
overlook the plus side. For example, the right to privacy,
especially for children, is one of the persistent concerns I
share with many Members of this Committee. Every day, internet
companies collect reams of personal data on Americans,
including kids, but we cannot expect children to fully
understand the consequences of their internet use and this
collection process.
Kids deserve, I believe, a chance to request a clean slate
once they are old enough to appreciate the nature of internet
data collection. That is why later this week, I will be
reintroducing the Clean Slate for Kids Online Act, which would
give every American an enforceable legal right to demand that
website companies delete all personal information collected
from or about the person when he or she was a child under the
age of 13.
The right to privacy and access to one's data could keep
this Subcommittee completely occupied. There is a lot more to
explore, including the subject of today's hearing, which will
examine how social media platforms use highly targeted
algorithms to captivate and persuade us in our every aspect of
our life. Algorithms influence what we read, watch, buy, and
how we engage, and they do not just affect our personal lives.
They infect--they affect us on a global basis.
For example, an independent civil rights audit last year
found that Facebook is not sufficiently attuned to how its
algorithms ``fuel extreme and polarizing content,'' and can
drive people toward self-reinforcing echo chambers of
extremism. Following the recent release of that audit, Chairman
Coons wrote a letter to Facebook, which I was proud to join,
that called on the company to do more to mitigate the spread of
anti-Muslim extremism and bigotry on their platform. Last
November, when Facebook CEO, Mark Zuckerberg, testified in this
Committee, I asked him about recent incidents where hate and
conspiracy groups use Facebook to plan and recruit, including
the organizer of the conspiracy to kidnap Michigan's Governor,
Gretchen Whitmer, and the so-called Kenosha Guard Militia,
which posted a ``call to arms'' on Facebook in the aftermath of
the shooting of Jacob Blake in Kenosha, Wisconsin.
That call to arms spread widely and was read by a 17-year-
old vigilante named Kyle Rittenhouse, who traveled from
Illinois to Wisconsin, where he allegedly shot and killed two
people on the streets of Kenosha on August 25th, 2020. That
militia page was reportedly flagged at least 455 times to
Facebook. However, Facebook found the page did not violate
standards, so it was left up. The response from Mr. Zuckerberg
at the hearing was, and I quote, ``It was a mistake. It was
certainly an issue, and we are debriefing and figuring out how
we can do better.''
Unfortunately, it is clear that they did not figure out how
to do better quickly enough. Not even two months later, a mob
of domestic terrorists and violent extremists stormed this
Capitol Building in the January 6th coup attempt, fueled by
widespread lies and conspiracy theories that claimed the
election had been stolen from the former President. While the
efforts to overturn a free and fair election were ultimately
unsuccessful, the trauma of that harrowing day lingers on.
After January 6th, the consequences of rampant hate and
misinformation on social media platforms has never been
clearer. We need social media companies to finally take real
action to address the abuse and misuse of their platforms and
the role that algorithms play in amplifying it. I look forward
to hearing from the witnesses, and I am hopeful that this
Subcommittee can accomplish under Chairman Coon's leadership
what we are expecting as an opportunity for this country to
move in the right direction.
Chair Coons. Thank you, Mr. Chairman. I will now briefly
introduce our witnesses for today and then swear them in.
Up first is Monika Bickert, Facebook's vice president of
content policy. She originally joined Facebook in 2012 as lead
security counsel, advising the company on child safety and law
enforcement. Prior to joining Facebook, Ms. Bickert served as
resident legal advisor at the U.S. Embassy in Bangkok where she
specialized in Southeast Asia rule of law, development, and
response to child exploitation and human trafficking. She also
served as a prosecutor with the Department of Justice for 11
years in Washington.
Lauren Culbertson is Twitter's head of U.S. Public Policy
based in Washington, DC, leads the company's Federal and State
public policy teams and initiatives. Serves as Twitter's global
lead for intermediary liability policy, and spearheads the
company's efforts to help combat the opioid crisis. Previously,
Ms. Culbertson worked in the U.S. Senate for my friend, Senator
Johnny Isakson of Georgia. She also founded a business,
Millennial Bridge, to promote public policy.
Alexandra Veitch leads YouTube's government affairs and
public policy for the Americas where she advises the company in
public policy issues around online and user-generated content.
She previously served as special assistant to President Obama
and as Deputy Assistant Secretary for the Department of
Homeland Security. Before that, she served as a member of
Speaker Pelosi's senior staff and began her career working for
Senator Barbara Mikulski of Maryland. Ms. Veitch's private
sector experience also includes leading North American
government affairs for Tesla and CSRA.
Tristan Harris has spent his career studying today's major
technology platforms and how they have increasingly become the
social fabric by which we live, and think, and communicate. Mr.
Harris is the cofounder and president of the Center for Humane
Technology, which aims to catalyze a shift toward humane
technology that operates for the common good. Mr. Harris was
the primary subject of the Netflix documentary, ``The Social
Dilemma.'' Mr. Harris also led the Time Well Spent movement,
which sparked product changes at Facebook, Apple, and Google.
Dr. Joan Donovan is a leading public scholar and
disinformation researcher specializing in media manipulation,
critical internet studies, and online extremism. She is the
research director at the Harvard Kennedy School Shorenstein
Center, and director of the Technology and Social Change
Project. Dr Donovan is a cofounder of Harvard Kennedy School's
Misinformation Review. Her research can also be found in peer-
reviewed academic journals, such as Social Media Plus Society,
the Journal of Contemporary Ethnography, Information,
Communication and Society, and Social Studies of Science. She
is a columnist at MIT Technology Review.
You are all virtual, which makes this next step just a
little different or novel for me. Would our four witnesses
please stand to be sworn and raise your right hand? Since I
cannot see you, I cannot affirm that you are doing that.
[Witnesses are sworn in.]
Chair Coons. Thank you. We will now proceed with witness
statements. Each of you has 5 minutes to make an opening
statement to this Subcommittee. Ms. Bickert, please proceed
with your testimony.
STATEMENT OF MONIKA BICKERT, VICE PRESIDENT
FOR CONTENT POLICY, FACEBOOK,
MENLO PARK, CALIFORNIA
Ms. Bickert. Thank you. Chairman Coons, Ranking Member
Sasse, and distinguished Members of the Subcommittee, thanks
for the opportunity to be here with you today. I am Monika
Bickert, and I lead content policy for Facebook.
Facebook uses algorithms for many of our product features,
including enforcing our policies. However, when people refer to
Facebook's algorithm, often they are referring to our content
ranking algorithm that helps us order content for people's News
Feed, so I will just dive into that one briefly.
Chairman Coons, as you pointed out, the algorithm ranks
content because we have--when people come to Facebook, they
have so much potential content they could see. The average
Facebook user has thousands of eligible posts every day that
she could see in her News Feed, and they are all there, but
what we do is we try to save them the time of sorting through
all of that to find what is most meaningful to them, by instead
using a ranking algorithm that ranks each post and tries to put
at the top the content the person will find the most
meaningful. The algorithm looks at many signals, including
things like how often the user typically comments on or likes
content from this particular source, how recently that content
was posted, and whether the content is in a format, such as a
photo or a video, that that user tends to engage with. The
process results in a News Feed that is unique to each person.
Naturally our users do not see the underlying computer code
that makes up the algorithms, but we do publish information
about how the ranking process works, and that includes
describing the inputs that go into that ranking process. Also,
we have a blog post that we put out whenever we have
significant changes to how we are ranking content in the
algorithm. Additionally, people can actually click on any post
in their News Feed and then go to--toggle the menu and go down
where it says ``why am I seeing this post,'' and they will see
the factors and explanation for why the algorithm put that
piece of content where it did in their News Feed. This helps
people understand what the algorithms are doing and why they
are doing it.
I do want to underscore that people can opt out of this
ranking algorithm. They can toggle over to a most recent News
Feed, which basically means that all of that eligible content
that you could see is simply ordered in reverse chronological
order. They can also choose from a--an option that we call
favorite speed, which basically allows you to select pages or
accounts that are favorites of yours, and then those will be
the only things that will be ranked in your News Feed. We
recently released a feature that allows people to toggle among
those different options.
As we work to bring more transparency to the algorithm and
also give people more control over how it works for them, we
also are working to improve the way that the ranking system
itself works. We announced last week that part of that includes
expanding our surveys to understand what is meaningful to
people and what is most worth their time, and also making it
easier for them to give us feedback on individual posts. That
is feedback that we will take from them, and we will build into
the ranking algorithms in a hope that, as we make this process
better and better, people will leave Facebook feeling more
inspired.
News Feed ranking is not the only thing that determines
what people might see when they come to Facebook. We also have
a set of community standards that says this is--there are
certain categories of content that simply are not allowed on
our service, and those are public standards that we have had
for years, and we publish a quarterly report on how we are
doing at finding that content and removing it. As the report
shows, we have gotten better and better, made significant
strides over the past years. If content is removed for
violating those standards, then it does not appear in our News
Feed at all. There are other types of clickbait--there are
other types of content that do not violate the standards, but
nevertheless people do not want to see them, like clickbait or
borderline content, that the algorithms down rank.
The reality is, it is not in our interest financially or
reputationally to push people toward increasingly extreme
content. If we--if we do something like that to keep somebody
on the site for a few extra minutes, but it makes them have a
worse experience and be less likely to use our products, then
that is self-defeating. Our long-term interest is to make sure
that people want to value our products for years down the road.
The algorithms are a key part of how we help people connect and
share and how we fight harmful content and misinformation on
our site, and we will continue to do more to help people
understand how the systems work and how they can control their
experience.
Thanks. I look forward to your questions.
[The prepared statement of Ms. Bickert appears as a
submission for the record.]
Chair Coons. Thank you very much, Ms. Bickert. Ms. Veitch,
would you please proceed with your testimony?
STATEMENT OF ALEXANDRA VEITCH, DIRECTOR OF
GOVERNMENT AFFAIRS AND PUBLIC POLICY FOR THE
AMERICAS AND EMERGING MARKETS, YOUTUBE,
SAN BRUNO, CALIFORNIA
Ms. Veitch. Chairman Coons, Ranking Member Sasse, and
distinguished Senators of the Subcommittee, thank you for
inviting me to appear before you today. My name is Alexandra
Veitch, and I am the director of government affairs and public
policy for the Americas and emerging markets at YouTube. I
appreciate the opportunity to explain how algorithms and
machine learning support YouTube's mission to give everyone a
voice and show them the world.
Through the adversity and uncertainty of the last year,
YouTube has helped bring people together as we have stayed
apart. More viewers than ever have come to YouTube to learn new
skills, to understand the world more deeply, and to be
delighted by stories that cannot be found elsewhere. YouTube's
business relies on the trust of our users, our creators, and
our advertisers. That is why responsibility is our number one
priority.
Our approach is based on what we call the four R's. We
remove content that violates our community guidelines, we raise
authoritative voices, we reduce the spread of borderline
content, and we reward trusted creators. Our written submission
explains each pillar in detail, but I want to focus my comments
today on how machine learning supports this responsibility work
when it comes to recommendations.
Recommendations on YouTube help users discover content that
they will enjoy and on key subjects. We want to recommend
content to our users that is authoritative. Recommendations are
based on a number of signals, including, if enabled, a user's
watch and search history. We also consider factors like country
and time of day, which help our system show relevant news
consistent with our efforts to raise authoritative voices. We
also give our users significant control over how their
recommendations are personalized. Users can view, pause, edit,
or clear their watch or search history at any time. We also
give users the opportunity to provide direct feedback about
recommendations so they can tell us if they are not useful.
We also believe we have a responsibility to limit
recommendations of content that is not useful or may even be
harmful. That is why in January 2019, we launched more than 30
changes to our recommendation systems to limit the spread of
harmful misinformation and borderline content, which is content
that comes close to, but does not cross the line of violating
our community guidelines. As a result we saw a 70-percent drop
in watch time of such content from non-subscribed
recommendations in the U.S. that year. This borderline content
is a fraction of one percent of what is watched on YouTube in
the U.S., but we know that it is too much, and we are committed
to reducing this number.
We know there is interest in the quality of the content we
recommend to our users. Researchers around the world have found
that YouTube's recommendation systems move users in the
direction of popular and authoritative content. Our efforts to
raise up content from authoritative sources and reduce
recommendations--excuse me--of borderline content and harmful
misinformation outweigh other recommendation signals, even if
the net result is decreased engagement. We are proud of our
record here, but we also work continuously to improve.
Because responsibility and transparency go hand-in-hand, I
would like to close with three recent transparency efforts we
have undertaken to facilitate a better understanding of our
platform. First, in May 2020, we collaborated with Google to
launch the first threat analysis group bulletin. It regularly
discloses actions that we have taken to combat coordinated
influence operations from around the world. Second, in June
2020, we launched a website called How YouTube Works to answer
frequently asked questions. It explains our products and
policies in detail and provides information on critical topics,
such as child safety, harmful content, misinformation, and
copyright. Third, earlier this month, we added a new progress
metric to our quarterly Community Guidelines Enforcement
Report. Our violative view rate estimates the percentage of
views on content that violates our policies. Last quarter, this
number was .16 to .18 percent, meaning that out of every 10,000
views on YouTube, only 16 to 18 come from violative content.
This is down by over 70 percent compared to the same quarter of
2017, thanks in large part to our investments in machine
learning.
As we work to balance the open nature of our platform with
our important work to be responsible, we appreciate the
feedback we receive from policymakers. We will continue to do
more. Thank you again for the opportunity to appear before you
today. I look forward to your questions.
[The prepared statement of Ms. Veitch appears as a
submission for the record.]
Chair Coons. Thank you, Ms. Veitch. Ms. Culbertson from
Twitter, if you would now present your opening statement, your
testimony, that would be wonderful. Thank you.
STATEMENT OF LAUREN CULBERTSON, HEAD OF U.S.
PUBLIC POLICY, TWITTER, SAN FRANCISCO, CALIFORNIA
Ms. Culbertson. Thank you, Chairman Coons, Ranking Member
Sasse, and Members of the Subcommittee for the opportunity to
testify on behalf of Twitter today on the role of algorithms
and amplification of content.
Twitter's purpose is to serve the public conversation. In
the early days, we were where you could go to share 140
character status updates. Our service has become the go-to
place to see what is happening in the world and to have
conversations about a wide range of topics, including current
events, sports, entertainment, and politics. While much has
changed since the company was founded 15 years ago, we believe
our mission is more important than ever. While many of the
challenges we grapple with today are not new, the creation and
evolution of the online world have affected the scale and scope
of these issues. Moreover, we must confront these issues amidst
increasing global threats to free expression.
We believe that addressing the global challenges that
internet services, like ours face requires a free and open
internet. We are guided by the following principles as we seek
to build trust with the people we serve. This includes
increasing transparency, providing more consumer control and
choice, and improving procedural fairness. Let me expand on the
principle of consumer control and choice, as it is particularly
relevant to today's discussion on algorithmic choice.
In 2018, we introduced a feature to give people on Twitter
control over the algorithms that determine your home timeline.
Through the sparkle icon you see on the top right corner of
your screen, you can choose to see your tweets ranked or toggle
to view tweets in reverse chronological order. When we
implemented this, some suggested it would be bad for our
business. We thought it was the right thing to do for our
users, and it has been a core feature ever since.
Further in line with our commitment to choice and control,
Twitter is funding BlueSky, an independent team of open-source
architects, engineers, and designers, to develop an open and--
open and decentralized standards for social media. It is our
hope that BlueSky will eventually allow Twitter and other
companies to contribute to and access open recommendation
algorithms that promote healthy conversation and ultimately
provide individuals greater choice. These standards could
support innovation, making it easier for startups to address
issues like abuse and harmful content at a lower cost. We
recognize that this effort is complex, unprecedented, and will
take time, but we are currently planning to provide the
necessary exploratory resources to push this project forward.
As we make investments to provide more transparency and choice,
we have also launched our Responsible Machine Learning
Initiative to conduct an in-depth analysis and studies to
assess the existence of potential harms in the algorithms we
use. We plan to implement our findings and share them through
an open process to solicit feedback.
Finally, as policymakers and Members of Congress here
debate internet regulation, I urge you to consider the ways
algorithmic choice and machine learning make Twitter and other
services a safer place for the public conversation. Technology
is essential for rooting out harmful content, like terrorism
and child sexual exploitation content. We also rely heavily on
machine learning tools to surface potentially abusive or
harmful content for human moderators to review. Simply put, we
must ensure that regulations enable companies to tap technology
to help solve some of the problems that technology itself
poses.
In summary, we believe that moving toward more open systems
will increase transparency, provide more consumer control and
choice, and increase competition in our industry. This will
ultimately lead to more innovation to solve today's and
tomorrow's challenges. We appreciate the enormous privilege we
have to host some of the most important conversations in the
world. We are committed to working with a broad group of
stakeholders to get this right for the future of the internet
and for the future of our society.
Again, thank you for the opportunity to be here with you
today.
[The prepared statement of Ms. Culbertson appears as a
submission for the record.]
Chair Coons. Thank you, Ms. Culbertson. Mr. Tristan Harris
of the Center for Humane Technology, if you would now please
give your opening statement.
STATEMENT OF TRISTAN HARRIS,
COFOUNDER AND PRESIDENT, CENTER FOR
HUMANE TECHNOLOGY, SAN FRANCISCO, CALIFORNIA
Mr. Harris. Thank you, Senator Coons, Senate Sasse, and
Chairman Durbin. It is an honor to be here with you today.
My background is I used to be a design ethicist at Google.
That was before recently featuring in the film, ``The Social
Dilemma,'' which many of you might have seen, which really had
the insiders who understood how these technologies were built
in the first place and have affected society. My friends in
college were some of the people who ended up working at these
companies in the very early days, including my friends Mike and
Kevin, who actually started Instagram.
What we really are missing in this conversation is a focus
on the business model and the intrinsic nature of what these
platforms are about, not because they are evil, and none of the
people who are here with us today are, you know, intentionally
causing any harm. Neither do I believe that the tech companies
who created these systems have intentionally wanted any of
these harms to happen. We are now in a situation where if we do
not diagnose the problem correctly, we are going to be in a bit
of trouble.
While you are hearing, you know, from the folks here today
about the dramatic reductions in harmful content, borderline
content, hiring tens of thousands more content moderators, et
cetera, it can sound very convincing. At the end of the day, a
business model that preys on human attention means that we are
worth more as human beings and as citizens of this country when
we are addicted, outraged, polarized, narcissistic, and
disinformed, because that means that the business model was
successful at steering our attention, using automation. We are
now sitting through the results of 10 years of this
psychological deranging process that has warped our national
communications and fragmented the Overton Window and the shared
reality that we need as a Nation to coordinate to deal with our
real problems, which are existential threats, like climate
change, the rise of China, pandemic, education, and
infrastructure.
Long as these companies profit by turning the American
conversation into a cacophony, into a kind of Hobbesian war of
all against all, because that is the business model, again, of
not the advertising, but the model of everyone getting a chance
to speak and have it go viral to millions of people. As long as
that is the promise with personalization, we are each going to
be steered into a different rabbit hole of reality, which Joan
will do such a good job of talking about. If you care about or
believe that masks work, you will see infinite evidence that
masks work. If you click on a couple articles that say masks do
not work and here is the, you know, stats in Florida showing
that the data was different, you will see infinite evidence
that masks do not work. Then we are pitted against each other
with this sort of infinite virality where anything that
somebody said can go viral.
Fundamentally, this is breaking many different aspects of
the Nation's fundamental life organs. For children, increased
cyber bullying leads to an increase in suicide. It takes
momentary drama and it turns it into drama snowballs that drown
out the effects of teachers and classrooms, who have to spend 2
hours on Monday morning clearing up all the drama that occurred
on social media over the weekend. It can reverse huge progress
that we have made in civil rights and not perpetuating racial
stereotypes by increasing online harassment and rewarding the
presentation of minorities in ways that are demeaning. It can
increase--it can inhibit our progress on climate change because
climate disinformation has gone viral on these platforms.
It can--it can--it can pose a threat to national security
in the sense that if Russia or China try to fly a plane in
United States, they would be shot down by our Department of
Defense, but if they try to fly an information bomb in the
United States, they are met by a white-gloved algorithm from
one of these companies that says exactly which zip code would
you like to target. It is the opposite of national security.
What a cannon was to a castle, social media is to the nation
state because it removes the power asymmetries of the millions
and billions of dollars that we have spent on F-35s, on
passport controls, and the Department of Homeland Security.
Once your society becomes virtual, all those protections go
away. Most importantly, if we are not coordinated as a society,
if we cannot even recognize each other as Americans, we are
toast. That is the only thing that matters. If we do not have a
truth that we can agree on, then we cannot actually change
any--do anything on our existential threats. We are really
sitting at a moment in history where we are transitioning into
becoming a digital society, and we kind of already have a
neural link brain implant for our society.
Right now we have two options. We have the Chinese brain
implant, which leads to kind of an Orwellian control of
thought, mass behavior modification, and surveillance, or we
have the Western brain implant that is built on this business
model that turns us into a performative culture. You know,
there is the Orwellian dystopia or the Aldous Huxley and
``Brave New World'' dystopia in which we fall into a kind of
development of amusing ourselves to death, constantly immersed
in distractions and unable to focus on our real problems.
What I really encourage us to think about is someone is
going to be controlling the 21st century. Will it be open
societies or closed societies? Either we beat China at becoming
China in a digital way, which we do not want to do, or we
figure out how to be a digital open society that does not
actually lose to that. That is our task. Either we figure it
out, or the American experiment may be in question.
[The prepared statement of Mr. Harris appears as a
submission for the record.]
Chair Coons. Thank you very much, Mr. Harris. Dr. Donovan,
if you would now give your opening statement, please.
STATEMENT OF JOAN DONOVAN, PH.D.,
RESEARCH DIRECTOR, SHORENSTEIN CENTER
ON MEDIA, POLITICS, AND PUBLIC POLICY, AND
LECTURER IN PUBLIC POLICY AT JOHN F. KENNEDY
SCHOOL OF GOVERNMENT AT HARVARD UNIVERSITY,
CAMBRIDGE, MASSACHUSETTS
Dr. Donovan. Great. Thank you to the esteemed Members of
the Subcommittee, Chairman Coons and Ranking Member, Senator
Sasse, for inviting me, and thank you to your staff as well. I
appreciate the opportunity to talk about how algorithms and
amplification shape public discourse. I am Joan Donovan, the
research director of the Shorenstein Center at Harvard Kennedy
School, and I study the internet.
I want to remind everyone that the internet is a truly
global technology requiring massive amounts of international
labor. Whatever policy ends up coming from the U.S. will
undoubtedly become the default settings for the rest of the
world. I also want to begin by saying that I believe a public
interest internet is possible, and I have to believe that in
order to do the heinous job of researching hate, incitement,
harassment, and disinformation on these social media products.
What a public interest internet means practically is
crafting policy that draws together the best insights across
many different professional sectors, matched with rigorous,
independent research into how automation and amplification
shape the quality of public life. We should begin by creating
public interest obligations for social media timelines and news
feeds, requiring companies to curate timely, local, relevant,
and accurate information, as well as providing robust content
moderation services and options.
Today, let us try to name the problem of misinformation at
scale and its impacts. In the U.S., when we talk about
politics, we are really talking about media about politics, and
when those news and information flows get laced with strategic
misinformation, then a simple search for something like
``coronavirus origin'' or ``mail-in ballots,'' can lead people
down the rabbit hole of medical misinformation or political
disinformation. In October 2020, I testified about
misinformation at scale having similar harmful societal impacts
as secondhand smoke, and it took a whole-of-society approach to
address the burden of disease caused by secondhand smoke, and
which led us to clear the air in workplaces, schools, and
airports.
When I say ``misinformation at scale,'' I am not
complaining that someone is wrong on the internet. What I am
pointing to is the way that social media products amplify novel
and outrageous statements to millions of people faster than
timely, local, relevant, and accurate information can reach
them. Post-2020, our society must assess the true cost of
misinformation at scale and its deadly consequences.
Disinformers, scammers, and grifters use social media to sell
bogus products, amplify wedge issues, impersonate social
movements, and push conspiracies. What I have learned over the
last decade of studying the internet is that everything open
will be exploited. Moreover, misinformation at scale is a
feature of social media, not a bug.
What do I mean when I say that? For example, because of
what I study, I often joke nervously that my computer thinks I
am a white supremacist. For researchers, going down the rabbit
hole means getting pulled into an online subculture where key
words, slang, values, and norms are unfamiliar, but,
nevertheless, the content is plentiful. There are four aspects
of the design of social media algorithms that can lead someone
into the rabbit hole. Coincidentally, they are also four R's.
Repetition relates to seeing the same thing over and over
on a single product, which, you know, likes, shares, retweets
do that. Redundancy is seeing the same thing across different
products; that is, you see the same thing on YouTube that you
see on Twitter. It tends to produce a feeling that something is
more true. Responsiveness is how social media and search
engines always provide some answer, even if it is wrong, unlike
other forms of media. Then last, reinforcement refers to the
ways that algorithms work to connect people and content so that
once you have searched for a slogan or a keyword, algorithms
will reinforce these interests time and time again.
Nowhere, of course, is this more prevalent than on YouTube
where any search for conspiracy or white supremacist content,
using the preferred keywords of the ingroup, will surface
numerous recommendations, and even offer up--offer up direct
engagement with these communities and influencers. If you have
recently searched for contentious content, like
``Rittenhouse,'' ``QAnon,'' ``Proud Boys,'' or ``Antifa,'' you
are likely to enter a rabbit hole or extracting yourself from
reinforcement algorithms ranging from the difficult to the
impossible. The rabbit hole is best understood as an
algorithmic economy were algorithms pattern the distribution of
content in order to maximize growth, engagement, and revenue.
I have a few things that companies could implement if we
want to talk about that later, but I think tackling a problem
this big will require Federal oversight for the long-term. We
did not build airports overnight, but tech companies are flying
the plane with nowhere to land at this point. Of course, the
cost of doing nothing is nothing short of democracy's end.
Thank you.
[The prepared statement of Dr. Donovan appears as a
submission for the record.]
Chair Coons. Thank you very much for your thoughtful
testimony, to all of our witnesses. Given the limited number of
Members, we may get several rounds of questioning, which is
exciting to me.
I just want to say to Ms. Bickert, Ms. Culbertson, Ms.
Veitch, your efforts to down rank borderline content to improve
transparency and empower users are all positive steps, and we
need to continue to find ways to preserve the positive benefits
of algorithms in showing content to people that is meaningful
to them, while addressing the very clear threats and
challenges, the very real potential for the harmful impacts of
algorithmic amplification. The questions I have today are meant
to get a better understanding of how one might further build on
your efforts and strike the right balance.
Some have proposed that social media platforms create
virality circuit breakers--we are all familiar with the phrase
``blowing up on the internet''--to detect content that is
rapidly gaining widespread viewership so that humans can review
whether it actually complies with platform policies before it
racks up tens or hundreds of millions of views. Professor
Donovan, could you just briefly, concisely explain why this
kind of mechanism might be particularly valuable?
Dr. Donovan. Yes, I think one of the things that we know
now from decades of tracking flagging, especially in
communities--conspiracist communities, hate communities, they
only tend to flag things as a result of trying to get
retribution on one another. They do not--they search for this
content and they enjoy it, and so systems that are built in do
not tend to work when it comes to particular kinds of strategic
misinformation, especially hate or harassing content as well.
As a result, what you need to do as a corporation is really
look for it. I know that there have been a couple of different
instances recently where corporations have found and rooted out
some really heinous stuff, but it obviously has to be part of
the business process and the process of content moderation to
seek out content that is, essentially, out of skew with signals
from the past.
Chair Coons. Thank you.
Dr. Donovan. That is one of the ways that they could
incorporate this.
Chair Coons. Thank you, Professor. Ms. Bickert, Facebook
said last fall it was piloting this very concept. What did you
find through this experience, and do you expect to further roll
this out more broadly? Please explain briefly, if you might.
Ms. Bickert. Senator, thank you for the question. We do
look at virality of content as a signal in when we should
assess proactively, as Dr. Donovan is suggesting, whether or
not something in it violates our policies or should be referred
to our fact checkers. The fact checkers, as you may know, are
more than 80 independent fact-checking organizations that we
work with. They can proactively rate content or we can send it
to them. Either way, or--and that could be based on user
reports, too. Either way, if they rate something ``false,''
then that is when we will put on that label saying this content
is false, directing people to the fact check, and we will also
reduce the distribution of that content in our News Feed.
Yes, we are seeing that those efforts are paying off. In
fact, we see that when we put one of those informational labels
on top of a piece of content, people are far less likely to
actually click through and see the content than they would if
we did not have that label.
Chair Coons. Ms. Bickert, I appreciated several steps
Facebook announced it was taking just in advance of the Derek
Chauvin verdict. One of these steps was limiting the spread of
content. This is a quote from Facebook, ``that systems predict
is likely to violate our community standards in the areas of
hate speech, graphic violence, and violence and incitement.''
Facebook's statement also noted the company had done this in
other emergency situations in the past. My question for you is
why Facebook would not always limit the rapid spread of content
likely to violate these standards? Could you help us understand
that?
Ms. Bickert. Senator, yes. What we are doing, and I put
that blog post out. What I meant by that was we use systems to
proactively identify when content is likely to violate or is
maybe borderline. Often what that can help us do is send that
to our reviewers and have them assess whether or not it
violates. In extreme situations, because, of course, not all of
that content will violate, you know. There will be some false
positives in that.
There is a cost to, you know, for instance, taking action
on that content with about half of it being real people who
look at it. What we do is, generally, we use those measures to
find content that we can send to reviewers. In situations where
we know that there is extreme and finite, in terms of date,
risk, such as an election in a country that is going through
civil unrest or the situation in Minneapolis with the Chauvin
trial, we will put in place a temporary measure where we will
deemphasize content that the technology--that the algorithms
say it is likely to violate.
Chair Coons. Let me ask a last question of the three social
media representatives before I turn this to my Ranking Member.
Ms. Bickert, Facebook has said, and I think you said in your
opening statement, it is not in your long-term interest to
pursue maximum engagement if it comes at the cost of spreading
polarizing and sensationalized content, that it is not really,
long term, in the financial interest of the company, let alone
its reputational interest, to have algorithms that amplify
harmful or divisive content. I agree with this. I am concerned
about what the underlying incentives are at all three of your
platforms for those who have to make decisions day in and day
out about exactly how your companies operate.
The MIT Technology Review reported last month that pay
incentives at Facebook for employees broadly are still tied to
growth metrics and engagement metrics. If I am a Facebook
employee who works on its News Feed, are the metrics the
company has set up to measure my performance directly related
simply to engagement and growth metrics, or is there some way
that these broader, more positive social objectives are
incorporated? If you could, all three, just answer briefly. Ms.
Bickert, for Facebook, Ms. Veitch and Ms. Culbertson, do you
provide pay incentives in terms of algorithms teams, directly
or indirectly, based on engagement- and growth-related metrics?
Thank you.
Ms. Bickert. Senator, the engineers are not specifically
goaled or given pay incentives simply to increase time on the
site. The focus is really on making sure that the products are
services that people find useful and will want to use for years
to come.
Ms. Culbertson. Senator, for Twitter, a top priority for
our company across our teams is to serve a healthy public
conversation, and I would love to share with you our transcript
from our latest analyst day, which is what we share with our
investors and our advertisers. All of the concerns and
priorities that we have talked about thus far today, you will
see that what we are telling you is exactly what we tell our
investors and our advertisers because they have the same
concerns. We have no incentive to have a toxic or unhealthy
conversation on the service.
Chair Coons. Thank you, ma'am.
Ms. Veitch. Similarly, Senator. Responsibility is our
number one priority, and when we set goals, we set those goals
around what we define as responsible growth. We may set a goal
to encourage adoption of a feature, but also we want to take
into account how that feature may be used or misused, and how
we can ensure it is adopted responsibly.
Chair Coons. Mr. Harris, if you could just provide a brief
comment on your understanding of the incentives of employees
and how it aligns with responsible growth versus growth at all
cost.
Mr. Harris. Yes. My understanding is, even to this day, I
think there was a brief experimentation at Facebook with non-
engagement-based performance incentives for social impact, but
that those have largely gone away, and it is actually still a
measure of engagement. This is things like not time onsite, but
sessions, 7-day active users, growth, and that is still the
focus. Everything else we are going to be talking about today,
it is almost like having the heads of Exxon, BP, and Shell
asking about what are you doing to responsibly stop climate
change.
Again, their business model is to create a society that is
addicted, outraged, polarized, performative, and disinformed.
That is just the fundamentals of how it works. While they can
try to skim the major harm off the top and do what they can--
and we want to celebrate that, we really do--it is just
fundamentally if they are trapped in something that they cannot
change.
Chair Coons. Thank you all. Let me turn to my Ranking
Member, Senator Sasse.
Senator Sasse. Thanks, Chris. My first question is actually
building--based on exactly--pardon me--on where the Chairman
just finished, and I really do think that constructive
engagement in these Committees is better than people trolling
for sound bites. I am not trying to get you all to fight, but
the truth of the matter is this hearing would work a lot better
if we were in the same room so we did not have to try to bring
you all into dialogue. The last three answers from the social
media companies and Mr. Harris' answers are just ultimately not
reconcilable, I do not think.
I want to go back to--I will start with Ms. Bickert as
well. Saying that you aspire to healthy engagement as opposed
to just more quantity, I agree with Mr. Harris' line that you
definitely aspire to skim the most destructive habits and
practices off the top of digital addiction, but the business
model is addiction, right? I mean, money is directly correlated
to the amount of time that people spend on the site. I guess
what would be useful for me is to hear each of the three of you
say what you think is wrong with Mr. Harris' argument, because
right now I think we are talking past each other.
I know that there is bad content and there is
disinformation content that you all, well intentioned as your
companies surely are, want to curtail. His argument is really
more broadly about the business model, and the business model
is addiction, is it not? Ms. Bickert, can we start with you?
What is--what is Mr. Harris missing?
Ms. Bickert. Senator, thanks for the question. You know, I
will say--I will say two things that I hope will be helpful.
One is, for us, the focus is always on the long term, and I
will give one concrete example of that. In January 2018, we put
out a post announcing that we were going to be prioritizing
content from family and friends over, say, news content. It was
called ``meaningful social interactions.'' We suspected that it
would lead to less time spent on the service and it did. It led
to people spending tens of millions of fewer hours on Facebook
every day. That was something that we did because we thought
that, longer term, it was more important for people to see that
sort of content because they would find it meaningful, and they
would want to continue to use this site. It is not--it is a
long-term picture.
The other thing I would say is, the teams that I work with,
who include the engineers who are focused on safety issues,
removing content, say, bullying content or hate speech, and the
engineers who are focused on the way that we reduce, for
instance, misinformation that has been labeled on the site, a
key statistic for those engineers is prevalence. Violating
content, that is their goal, and we put out public reports on
their prevalence. You know, that is an example of how we are
focused on the long term and making sure that we are stopping
abuse and maintaining a healthy environment.
Senator Sasse. I want to be clear that I am not--I am not
targeting the three of you because I--my opening statement is
very sincere. I think that there is a danger in politics, in
governance, where if you agree that there is a problem, then
there must be a definitive regulatory solution that can come
real fast and easy. On the other hand, if you are not persuaded
there is a regulatory fix right away, then you have to deny
there is a problem. I am sort of a heterodox tweener on this in
that I do not have clarity about what the regulatory fixes
would be, but I think society-wide, we should admit that there
is a problem in the last 12 or 14 years as we have consumed
more and more digital stuff that seems to be correlated with
some benefits, but also some very real costs.
I do not think it is just your companies. I mean, there
have been reports out of the New York Times about their own
internal deliberations about how they would like to have more
Americans engaging in healthy content, and they are just
printing money right now over the course of the last 4 or 5
years, but engagement is much higher when they are angry. When
the content is angry, it leads to more engagement.
I do not think any of you are really going to dispute that,
but I would like to stay where I--where the question was 2
minutes ago which is, I would love it if, Ms. Culbertson, will
you tell me what you think is wrong with Mr. Harris's argument?
Ms. Culbertson. We are really focused on serving the public
conversation, and that includes having controls in place so
people can also control their experience. I think as we are
talking about algorithms today, you know, Twitter really does
one thing. We do tweets. We have a home timeline. As we are
talking about algorithms, we have a ranking algorithm. That is
designed to show you what might be most relevant to you. Then
also if we are talking about screen time or how much time do
you spend on a service, I think that is really relevant because
I know as a user of Twitter myself, I rely on that so I can
kind of see what happened in the day, what people are talking
about, and then I kind of log off, move on with my day.
I think it is important to look at this in a nuanced view
and recognize that algorithms can also be helpful in terms of
cutting down on screen time or providing more valuable
experience for people.
Senator Sasse. Sure, but the reality is the loop between
the products that are being produced and the way, we as a
narcissistic Senators, consume it is--maybe I will ask if this
is the right question. Is it or is it not true that when
somebody tweets something that is really anger-invoking and
outrageous and it goes viral, but then 2 hours later, they
realized they were wrong and they correct it, is the correction
not usually, like, 3 percent of the traffic of the original
outrageous, but false, thing? I mean, so it seems to me that
what we know is that people are pretty good at short-term rage,
and the product capitalizes on that, does it not?
Ms. Culbertson. I think when looking at Twitter, it is
important to remember that it is an open, public conversation,
and so everything that happens is in the open in the public,
and, typically, you know, you will see these debates play out.
You know, I am a firm believer that connection and connectivity
is key to solving problems, and that is what our service does.
Of course people have robust and spirited debates, but I think
you have to look at the greater picture there.
Senator Sasse. We are basically at time, so I will not
indulge the--get the Chairman to indulge me much longer. Ms.
Veitch, do you have anything to say that you think Mr. Harris
is wrong about? It would be useful, but right now we are not
getting much direct engagement with that. He is making a big
argument, and I think we are hearing responses that are only
around the margins. Ms. Veitch, do you have any criticism of
Mr. Harris' argument?
Ms. Veitch. Yes. Thank you, Senator. I think in your
opening statement, you call these nettlesome problems. I agree
they are, but I would just make two quick points. First,
misinformation is not in our interest. Our business relies on
the trust of our users, but also our advertisers, who, on our
platform, advertise on single pieces of content. We want to
build these relationships for the long term. That is why we
bake user choice, user control right into the product with
things like timers and the opportunity to turn auto play off,
take a break reminders of which we have sent over one billion.
Again, those exist so we can build this relationship with our
users for the long term.
Senator Sasse. Thank you. I do think a lot of those user
preference tools to manage our level of engagement. Engagement
bridging over to addiction is really an important innovation,
so I applaud those tools.
Chair Coons. Thank you, Senator Sasse. Chairman Durbin?
Chair Durbin. Thanks a lot, Chairman Coons, and it is a
pleasure to be with you. First two disclaimers. I am a liberal
arts lawyer, not nearly as tech savvy as I should be for this
hearing. Point number two, my experience in Government, which
has been over several decades, suggests that we are slow to
recognize issues that are fast-breaking and have a very spotty
record when it comes to responding to them in a thoughtful way.
I hope this is an exception. If I could address Tristan Harris
first. Aloha. Then may I ask you this question? I have been
reading and trying to understand why the European Union is
taking such an apparently bold and innovative approach to this
subject and we are so slow to respond. Any thoughts?
Mr. Harris. I do not--in this country, obviously we value
free speech above other values, and so that makes it, I think,
harder to regulate an environment where the composition of what
constitutes speech in our society is a Frankenstein monster
that spins out blocks of attention virally to different people
on a personalized basis and outrages them. I do not think that
we have had a framework.
To your earlier point, one of the quotes we reference often
is from E.O. Wilson: ``The fundamental problem of humanity is
we have Paleolithic emotions, medieval institutions, and then
accelerating godlike technology.'' It is not meant as an insult
to any government body. It is just to say the challenge you
pose, which is how do we deal with the first derivative of
these issues, we are still talking about mostly conversations
that we had, you know, 4 years ago about spread of
misinformation, things like that. The rate and acceleration of
new kinds of threats, new kinds of issues, the growth rate of
that is going far--growing far faster than the growth rate of
our capacity to mitigate or respond to those threats.
I was speaking with someone in the fact-checking network
who said, you know, there are now 200 billion messages a day
going through WhatsApp, 15 billion going through Facebook. They
get about 100 fact checks per day. If you think about a bank
being overleveraged and how much risk are they--how far over
their skis are they, we have got about 200 billion-to-100 in
terms of scale event of information that is running through a
system without moderation. With--as Senator Sasse said, the
decentralized incentive for yellow journalism is, if it bleeds,
it leads. It wants to make each of us into yellow journalists
because we each get more rewarded the more extreme things that
we say in an inadvertent way.
In a subtle way, in this is very slow climate change and
culture kind of way, just by 2 percent increasing the kind of
outrageous sense of what you are saying, the sensationalism of
what you are saying, you got a few more likes, you get a few
more hits, and so you keep doing it. Suddenly, that heats up
the global conversation and has us not even recognizing our
fellow Americans as fellow Americans.
Chair Durbin. I want to take you down a different path, if
I can, for a moment as I try to read and absorb the European
Union's risk-based approach to this AI issue. They say two
things they find unacceptable: use examples, manipulating human
behavior. I think that is at the heart of it, as I understand
the explanation. I have heard people from Facebook talk about
making your Facebook experience more meaningful, and folks from
YouTube and Twitter talking about healthy dialogue. The bottom
line is, it appears that there--like it or not, there is a
factor here where our human behavior is being affected by what
we are seeing, what we are reading, what we are experiencing,
and that seems to violate the basic premise of the EU
regulation.
The second one in the extreme is this social credit
scoring, which they use as an example, which apparently is
rampant in places like China, and takes the manipulation, and
analysis, and algorithms to the point where they disqualify
people from being able to get on a fast train in China because
their social credit score does not merit it. Give me your
thoughts on those two elements.
Mr. Harris. Yes, on the manipulation front, you know, that
would disqualify just about all of the three companies that are
sitting in front of you, including TikTok, by the way, which is
not getting nearly as much attention and it is actually
dominating children's minds on a daily basis. You know, I think
if you have seen the film, ``The Social Dilemma,'' we speak
about my background in a class called the Stanford persuasive
technology class and lab, and that these technologies are
designed to be persuasive. When you hear Twitter talk about the
healthy conversation, there is still a persuasive technology
that manipulates human behavior. They are trying to do it in as
healthy a way as they can. When you hear Facebook talk about
meaningful social interactions, they are still creating this
sort of digital addiction dopamine loop, getting you to invite
your friends, create social obligations, dripping out
notifications one at a time instead of in batches. They are--
but they are doing it in ``the most meaningful way'' that they
can.
They are trying to do the best they can, but to Senator
Sasse's point earlier, it is almost like listening to a hostage
in a hostage video. Nothing they are saying kind of makes much
sense until you realize there is a gun off stage holding--you
know, their business model--held to their heads, and it is
causing them to say the things that they are saying. Again,
these are really good people. They are just--we cannot talk
about the actual underlying issue because the business model is
based on this manipulation. Sorry. I think I ran out of time
for your second part.
Chair Durbin. In my last 5 seconds, I am plugging for the
Clean Slate for Kids Online Act. It is a small, but very
important, part of this conversation. Thank you, Mr. Chairman.
Chair Coons. Thank you, Chairman Durbin. Thank you for
joining us today. Senator Hawley.
Senator Hawley. Thank you, Mr. Chairman, and thanks to the
witnesses for being here. Mr. Harris, I would just like to come
back to you, if I could, because I think you just said
something in response to Senator Durbin that is very, very
important. You talked about the business model of the
companies, although the companies that are up before us today
are, in fact, all of the dominant platforms. The point--your
point I think is, and I would like you to elaborate on it, the
business model of these companies is addiction. The business
model of these companies is advertising, which takes place by
getting more and more users to spend more and more time online,
so that these companies can gather more and more personal
information about them to sell them more and more stuff, right?
It is an attention treadmill. It is an addiction economy. That
is the design. That is by design. They did not wander into it.
It was not an accident. It was not, oops, how did we end up
here. They designed it this way. Addiction is the design.
You have written a lot about this. You produced, as you
have referenced, a documentary about this. Can you just tell us
more about this core business model of these dominant
platforms?
Mr. Harris. Yes, thank you. I would actually say that there
are so many of the ways that these platforms' work actually
comes from that original business model. I literally remember
in the early days of Instagram when they were trying to figure
out a way to get people to kind of come back, and they decided
to borrow Twitter's follow model. The idea that, you know, you
can follow any user. Why did they invent this follow model? Let
us just ask it that way. Why are we following each other here
and there? Because what it does is it creates this treadmill
where every day or two you see you have got two more people who
followed you, and that creates what is kind of like a viral
bait, right? It is like a clickbait that gets you to come back
into the service to see, oh, who is that person? What do they
want to follow me for? They are preying on each other's social
validation, and then they have this AI that sits on top that
tries to predict, well, which users could we get to follow
whom.
If you want to say the problem has gone away, Facebook
right now--right now--is testing the increase of suggested
users you may know--by the way, this is in the film, ``The
Social Dilemma''--when one of the AIs sort of twirls the
mustache and says, could we invite them--could we get them to
invite more friends. That is actually literally what they are
doing right now, which is they are saying, here are some
channels, here are some people you might want to follow, and
they are very good at predicting that next person.
I have even done this myself because I have got a
supercomputer pointed at my brain saying, yes, this is your old
friend that you might actually know. It creates, again, this
treadmill that is all about getting us to come back. It is
really almost like a digital drug lord because if you are a
child and you saw ``The Social Dilemma,'' you say, hey, I am
going to uncheck--you know, pull out of these services for a
while. Have you ever tried taking down your Facebook account or
not using Instagram for a week? You will notice they dial up
how aggressively, like a digital drug lord, to show you more
notifications and more emails. They will send you more and more
emails, each of the services will. They would not do that if
their business model was not preying on addiction.
These are all techniques, again, from the kind of
persuasive framework that all of the people that I came up with
in the tech industry really learned, again, not because these
people are evil. Just little bit by bit, you do what works, and
then it keeps turning into this treadmill, and it sort of turns
us all into attention vampires that want attention from other
people.
Senator Hawley. Yes, ``attention vampires'' is a great
term. The amount of control that this business model then gives
these companies over our lives is absolutely unbelievable.
There is that infamous experiment that Facebook ran on its
users in 2014, half a million users, to see if it could depress
them or change their moods by tweaking the algorithm that then
would refer the content that they saw. Of course, the amazing
and extremely scary thing is that they could. They could, in
fact, directly influence their users' moods. They could, in
fact, change how their users felt about the day, or felt about
a particular story, or felt about a particular event by
tweaking their algorithm, because they control--``they,''
Facebook--control what their users see. They control the
interactions. They increasingly control how much time people
spend online. You know, and really these companies say that
they are about social media, but really they once were, right?
I mean, they used to be social media networks.
Back in 2006 when Facebook first introduced the News Feed,
there is this great post that Mark Zuckerberg wrote called
``Calm Down. Breathe. We Hear You,'' in which he assured users
that the introduction of this new called the News Feed would be
not a very big deal. He said, ``We have been getting a lot of
feedback about mini-feed and News Feed. We think they are great
products, but we know that many of you are not immediate
fans.'' There is an understatement. ``For those who are worried
about privacy,'' he went on, ``nothing you do is being
broadcast.'' Right. ``Rather, it is being shared with people
who are about what you do, your friends.'' Of course, what
turns out to be true is it is not your friends who he is
concerned about. It is advertisers, and who they are sharing
this personal information with is advertisers who are in the
business of trying to manipulate the user.
Let me just ask you this, Mr. Harris. These companies have
been able to do this. They have been able to manipulate
content. They have been able to push particular content to
users. They have been able to try and interfere with their
user's own moods, in large part, because they get a special
blanket immunity from this Government, from the Federal
Government. Here is my question. Why should any platform that
engages in algorithmic amplification or behavioral advertising,
why should they get the Section 230 immunity? Why should we not
just remove Section 230 immunity for any platform that engages
in behavioral advertising or algorithmic amplification?
Mr. Harris. Section 230 is a--is a--is a difficult--it is a
double trade, and so it is going to be a complex one to get
into there. It is important to say that whether the companies
want to or not, if they took their hand off the steering wheel,
they are still manipulating people's emotions. In fact, the
more they take their hand off the steering wheel, the more
outrageous, values-blind engagement, which means that,
literally, the most outrageous stuff, the most child
trafficking stuff, the most sexual pornographic stuff would
rise to the top, and that would also be a form of manipulation.
If you compare side-by-side how much restrictions we do in an
IRB study in a psychology lab at a university where, if you are
going to experiment on 14 people, you got to file an IRB
review, Facebook, Twitter YouTube, TikTok, are, on a regular,
daily basis, tinkering with the global brain implant of 3
billion people's daily thoughts with no oversight.
I think what we need to do is actually compare side-by-side
what are the regulations and protections, as you are talking
about, that we apply in one domain and we are not applying in a
different domain. I think the focus on content in 230 is
problematic. I think it is more about the design and
fundamental oversight about the way these platforms
fundamentally operate.
Senator Hawley. I will just say in closing, Mr. Chairman,
that I think Senator Durbin put a very good question, which is
why are we so slow. Why has the United States been so slow to
confront this kind of manipulation and these kinds of
deleterious effects? I think part of the answer to that,
frankly, is money, and we all know that is true. These
companies spend enormous sums of money trying to influence this
body, trying to influence our regulators, trying to influence
the Federal Government. It is time that this Congress did
something about it to show who is really in charge. It is not
them. It should be the people. It should be us. I am at the
point I think we probably should repeal 230 completely, but we
certainly have to take action to stop this kind of rampant
manipulation for profit, which is what these companies do.
Thank you, Mr. Chairman.
Chair Coons. Thank you, Senator Hawley. Senator Hirono is
joining us next by Webex.
Senator Hirono. Thank you, Mr. Chairman. I would like to
ask Ms. Bickert to respond to three questions that I have for
her very briefly. A recent report from the MIT Technology
Review found that Facebook's ad delivery algorithm
discriminated based on gender when surveying ads for jobs.
There are numerous other examples of Facebook's system
discriminating on the basis of age, gender, and race, and
delivering housing, job, and financial services ads. Facebook's
help center describes what factors it uses to target ads to
users, including, ``other information about you from your
Facebook account, such as your age, your gender.'' Facebook
also allows ads to be targeted based on things like zip codes
that can be used as a proxy for race.
Question one. Are you concerned that Facebook's reliance on
these factors in targeting ads results in discrimination? Can
you give a ``yes'' or ``no'' answer, please, Ms. Bickert?
Ms. Bickert. Senator, thank you. Making sure that the--that
our ads are served in a fair way without discrimination is
always a priority for us, and we do have policies in place to
prevent discriminatory targeting. I am very happy to follow-up
with you in the interest of time on some of the specifics of
those policies.
Senator Hirono. How does Facebook ensure that it does not
violate, for example, civil rights laws when targeting ads for
housing, employment, and financial services?
Ms. Bickert. Senator, thank you. We have policies around
when people can use certain targeting criteria. For instance,
we do not allow some of the more sensitive targeting criteria
that you have mentioned. We do not allow that for certain types
of advertisements, such as financial services advertisements or
housing advertisements. I can follow-up with you on some of
those specifics on how we ensure that the ads are served in a
fair way.
Senator Hirono. The--you have addressed the concerns raised
by the recent report from the MIT Technology Review that found
discrimination on the basis of gender, for example, that you
have addressed those concerns raised by this report?
Ms. Bickert. Senator, I am not familiar with that specific
report, so I cannot comment on that. I can tell you in my years
of being in this job, making sure that we do not have
discriminatory ads has been a priority for us. We have worked
on that for years, made a number of improvements, and I can
follow up with those details.
Senator Hirono. Yes, I hope you can follow-up after you
have read the MIT Technology Review, and I do not think they
are the only ones who raise those concerns. Third question.
When Facebook has been sued for discrimination by its ad
targeting algorithm, it has often hidden behind Section 230.
Earlier this year, I joined Senators Warner and Klobuchar in
introducing the Safe Check Act, which would remove Section 230
immunity for violations of civil rights laws. We are not
talking about total removal of Section 230, but as referencing
civil rights laws. Do you agree that Facebook should not be
immune under Section 230 when it discriminates when delivering
ads?
Ms. Bickert. Senator, thank you for the question. I agree
that there should be regulation to hold social media companies
accountable. I think that there is there is a lot to consider
when crafting that regulation. I put out a white paper on how I
think regulation could work, and I look forward to having those
conversations, and I know our team does as well with your
office and other offices.
Senator Hirono. For Mr. Donovan and Mr.--Dr. Donovan and
Mr. Harris, do you think that regulation is appropriate to
prevent Facebook ads from discriminating, or should we just
eliminate the immunity from lawsuits for civil rights laws'
discrimination? Briefly, Dr. Donovan and Mr. Harris.
Dr. Donovan. Yes, this is Dr. Donovan. I think that we do
need to have some carve-out related to civil rights violations,
especially those that would require oversight. One of the main
problems here that we are addressing is that when automation is
matched with amplification--that is, there is no review of
these ads--we do not know who we are doing business with. Not
only does Facebook not know who they are doing business with
directly, but--and we see a bunch of different shell games with
some of the stuff that has been implemented related to
advertisers disclosing who they really are. Overwhelmingly,
over the pandemic, we have seen all kinds of scams, grifts, and
hoaxes that violate people's civil rights, and so we do need
this.
Senator Hirono. I appreciate that, that you think there
should be a carve-out. Mr. Harris, do you think there should be
a carve-out for civil rights violations?
Mr. Harris. I am not familiar the exact way the legislation
is written, but that is--I would be sort of in support of that.
I think the important thing to recognize here, as Joan was just
saying, is that the companies make money by not having human
oversight discernment.
Senator Hirono. Thank you. I----
Mr. Harris. The premise is that more of this--yes.
Senator Hirono. I am dying to ask this one last question,
Mr. Chairman. Apple released a software update this week that
gives users control over whether apps are able to track them
when they use other apps and surf the web. Giving users greater
control over their online privacy strikes me as a positive
thing. This is for Mr. Harris. What do you think the impact of
this change will be, both on the issue of misinformation and
division we are talking about today, and the system of
surveillance capitalism that companies, like Facebook and
Google, rely on more broadly?
Mr. Harris. Yes. I applaud Apple for making this small step
in--I think of it almost like a carbon tax on micro-targeted
advertising. If we are left with this sort of extractive
business model that treats us as the product and not the
customer, then removing micro-targeted advertising, sort of the
hyper-personalization because apps can track you across
applications. Think of it as going more close to the 1970s
model of billboards, which are de-personalized, as opposed to
this micro-targeted model. That is not completely true because
advertisers can still micro-target you from within the Facebook
ecosystem, so it is not going to affect things that much, and
it will not address problems like misinformation or
polarization, which will continue. It just is almost like a
subtle carbon tax on the advertising business model.
Senator Hirono. Thank you. Thanks, Mr. Chairman.
Chair Coons. Thank you very much, Senator Hirono. Senator
Grassley?
Senator Grassley. Thank you, Mr. Chairman. I use Twitter
regularly. Facebook, Twitter, and YouTube are also popular
social media platforms where users disseminate their views and
opinions for billions of users around the world. Just here in
the United States, in 2019, an estimated 72 percent of
Americans use at least one social media site. People can make
their voices heard, share their opinions, and interact.
Increasingly, however, these Big Tech companies are deciding
what we can and cannot say and infringing on Americans' freedom
of speech.
I constantly hear from Iowans about their concerns, with
control that Big Tech has over the discourse in this country as
well as the biases that these platforms have against
conservative voices in Middle America. I have heard numerous
stories about posts being deleted, businesses removed, and
creditors silenced. Many times, this happens without warning
and very little, if any, due process. These platforms have
monopoly powers with very few competitors and are not
constrained by market forces, and consumers have no
alternative. Big Tech is also immune from liability under
Section 230. This immunity, combined with monopoly, allows them
to censor, block, and ban whatever they want. We must look at
the power and control that a handful of companies have over
speech and their silencing voices with which they disagree.
My question is to Ms. Bickert, Culbertson, and Mr. Veitch--
Ms. Veitch. When you decide to remove current content from
platforms, do you believe that you do that consistent with
First Amendment free speech principles, such as viewpoint
neutrality? If you believe that you are doing that, then why is
it that conservative voices are consistently the ones who are
being censored?
Ms. Bickert. Senator, thank you for the question. We are a
platform for ideas across the political spectrum. I do believe
that we enforce our policies without regard to political
affiliation. I do hear questions from both sides of the aisle,
if you will, about whether or not we are fair in our content
policy enforcement, but I can tell you that we enforce our
policies without regard to political ideology.
Ms. Veitch. Senator, I also appreciate the question here.
We want YouTube to be a place where diversity of viewpoints are
heard. We do have public-facing community guidelines that
govern what is allowed on our platform and what is not. We do
enforce these consistently without regard for political
viewpoint. You did mention due process, so I wanted to call out
that when content is removed from a creator, a creator does
receive an email explaining that and is given an opportunity to
directly appeal. We make public the data around our appeals. In
the last quarter of 2020, we did have 223,000 appeals and
83,000 reinstatements, showing we do not always get this right,
but we certainly want to apply our policies evenly.
Ms. Culbertson. As for Twitter, and, Senator, thank you for
the question, and we love to see your tweets on Twitter. You
are one of my favorite follows. As you probably appreciate,
Twitter would not be Twitter if everyone had the same
viewpoints, and we welcome diverse perspectives. It is what
makes our service Twitter. We have rules in place. We enforce
them impartially. I know people have concerns and they believe
that companies like ours should be more transparent. That way--
that is why we have put forth three core solutions which we
think would go a long way to addressing some of these concerns.
The first is increased transparency, the second is more user
control and choice over algorithms, and the third is enhanced
due process. If we do make a mistake, that users have the
ability to appeal and have their decision reviewed against our
terms one more time.
Senator Grassley. Yes. There are countless examples of
material being removed by a platform stating that it is
misinformation, but it is actually just viewpoints that
liberals might disagree with. What are your platforms doing to
ensure that they are not using pretextual reasons to censor
differing opinions? Then that is my last question.
Ms. Culbertson. I am happy to take this one, and Twitter
has taken a very narrowly scoped focused on misinformation at
this time. We have three categories that govern our policies.
The first is synthetic and manipulated media, the second is
civic integrity, and the third is COVID-19 misinformation. We
are piloting a program called Bird Watch that would crowd
source annotations to potential misinformation. This is
something to address all forms of misinformation, and it would
also bring more voices in to help us with that work.
Ms. Veitch. Senator, we do have robust community guidelines
on YouTube. Those exist to keep people safe. To your point, it
is important to note that those community guidelines are public
facing and can be reviewed by any of our users.
Senator Grassley. I guess, Mr. Chairman, nobody--the third
person did not want to comment, so you can--I will give up my
time. Go ahead.
Chair Coons. Thank you, Senator Grassley. I appreciate
that. Senator Klobuchar.
Senator Klobuchar. Hard act to follow there. Okay. Thank
you, Senator Grassley, for your interest in this issue. Mr.
Harris, you and I were on a panel together in March, and good
to see you again. Could you explain more about how companies'
market power exacerbates problems of disinformation, extremist
contact, and bias?
Mr. Harris. Yes. It is great to see you again, too. Thank
you. Thank you for the question, Senator. You know, if there is
anyone with an alternative model to the current problems that
plague us in misinformation, disinformation, and morality, can
they succeed in the marketplace? There is something in, you
know, the literature called Metcalfe's Law, right, where the
power of a network grows exponentially with the number of
participants. Really what we have between social media
platforms is a race to Metcalfe. Once you have a dominant
platform, it is very hard for there to be an alternative.
Market concentration means that even if there are
alternatives that are trying to do any--and solve any of the
problems we are talking about today differently, they are going
to get bought up by the existing platforms. If you are a
venture capitalist, the only way you are going to fund an
existing company is by knowing that there is an exit pathway,
and we kind of all learned the lesson as all the sort of
competing platforms and things that have come out have just
been acquired by the existing companies. We also----
Senator Klobuchar. Yes. Could I--I think that point just
cannot be lost because there are regulations we can put in
place--that is one way to do it--and you can do both things at
once. If you have a company that buys out everyone from under
them--in the words of Mr. Zuckerberg, they would rather buy
than compete--and buys companies like Instagram and/or
WhatsApp, we are never going to know if they could have
developed the bells and whistles to help us with misinformation
because there is no competition. Do you want to comment more on
that, Mr. Harris?
Mr. Harris. Yes. I mean, just as you said, there--if
WhatsApp were to remain independent, and let us say we are
living in some alternative reality where now WhatsApp was
separate, and we solve these problems and WhatsApp decided they
are going to spend billions more dollars on content moderation
because they want to actually be the platform that people can
trust. They cannot make that choice because Facebook bought
them, and now they are sort of integrated in how much they are
working on these problems, and it is a race to sweep the
garbage under someone else's rug.
What we have seen, unfortunately, is instead of
collaboration between all these platforms, in some cases, we
have seen, hey, look how bad their problems are because we do
not want to pay attention to ours, not, again, because they are
evil. It is just game theory happening between the companies.
It really does----
Senator Klobuchar. Okay. Thank you. Dr. Donovan, in your
research, you have looked at medical misinformation at scale
and the role of social media platforms. Could you please
comment on how the sheer size of a few powerful platforms
affects the problems that we should be addressing?
Dr. Donovan. Yes, thank you, Senator Klobuchar, and I
really look forward to reading your book, ``Antitrust.'' The
problem of medical misinformation, of course, is one that was
exacerbated by the pandemic, but anti-vaccination activists
have a long history of using social media in order to attack
the public understanding of science. During the pandemic, of
course, the way in which the tech companies have turned to
medical misinformation is really--it is like putting a band-aid
on an open wound. Right now what we need is a comprehensive
plan for ensuring that people have access to timely, local,
relevant, and accurate information, like public interest
obligations, but instead what we have is a very slapdash
approach to, you know, whatever the breaking news event is of
the day. I do think that the size of the platform and the way
in which medical misinformation scales much more quickly than
any intervention is probably the most pressing public health
issue of our time.
Senator Klobuchar. Okay. Thank you. Ms. Bickert, a recent
poll found that nearly 1 in 4 Americans said they will not get
the coronavirus vaccine. Meanwhile, a recent report from the
Center for Countering Digital Hate identified 12 specific
content producers as the original source of an estimated 65
percent of coronavirus disinformation online. Recently, Senator
Lujan and I, after he conducted a hearing, sent a letter to
Jack Dorsey and Mark Zuckerberg calling on them to remove these
individuals from the platforms. Do you agree that more action
needs to be taken? What is your response to our letter? I guess
I would start with you, Ms. Bickert and then go to you, Ms.
Culbertson.
Ms. Bickert. Senator, thank you, and thank you for the
letter as well. I know that we have assessed that content and
removed those accounts that were violating, and I can--I can
follow-up more with you on the specific details of that.
More broadly, and I think this is a really important issue,
we know that we have to get it right when it comes to
misinformation around COVID. One of our goals is to help 50
million people get vaccinated. We are doing that both
proactively through partnerships with local and national health
authorities, making sure that we are directing people to
authoritative health information, including where they can get
vaccinated. We have now directed--we have connected more than 2
billion people with those authoritative health resources. We
also, since the very beginning, have been partnering with the
CDC to remove content that contradicts CDC guidance that could
lead to an increased risk that people could contract or spread
COVID. That includes removing over 12 million pieces of safety-
related COVID-19 misinformation.
Senator Klobuchar. Okay. In my role on the Commerce
Committee, of course, Senator Cantwell is leading a bill on
privacy. Do you agree that consumers should have the ability to
access their data and control how it is used, including what
data is used in social media company algorithms? Do you give
customers that ability now for both content and advertising
algorithms, Ms. Bickert?
Ms. Bickert. Senator, thank you. We do give people a number
of controls. That includes everything from the ability to
download your own information, remove it, control who can see
your posts, see what type of--you can opt out of our algorithm.
You can see who can see your content at any time, and you can
change those----
Senator Klobuchar. Is the company then supportive of our
bill on privacy Senator Cantwell and I----
Ms. Bickert. I would have to have our U.S. public policy
team follow up with you on the specifics of that.
Senator Klobuchar. Okay.
All right. Okay. Thank you. I appreciate that. I am going
to--I will ask the--I can see Senator Coons over his mask
raising his eyebrows at me. That is his sign enough is enough.
Chair Coons. No, no, the Chairman welcomes additional
questions from the celebrated author of an outstanding book I
need to----
[Laughter.]
Senator Klobuchar. Okay. I was just going to ask--I will
ask one more question then of Ms. Culbertson from Twitter, just
the original question that I had asked Ms. Bickert about the
``Disinformation Dozen,'' as we call them, the accounts online.
Of course, some of these issues that I have had--I am talking
about with market power, is not as applicable to Twitter, which
I appreciate, but--as a competitive--a competitive platform.
Could you at least answer the question here about this
disinformation dozen?
Ms. Culbertson. Certainly. Thank you for the question. We
have and are continuing to review this particular group of
individuals against our policies, and we have taken enforcement
action on several of these individuals. Our team will be
following up this week with all the details around that. Also,
I just wanted to note that while we are competitors, we are
partners to address a lot of really harmful content--content
issues. We have collaborated on COVID. We work together on
terrorism, child sexual exploitation, opioids. I take issue
with the premise that was mentioned earlier. There is
collaboration across industry to address some of the most
harmful content on the internet.
We also invest heavily in our partnerships with experts,
especially around COVID. We worked very closely with the CDC,
HHS, the White House, to not only enforce against our rules,
but to also ensure that people try to have access to credible
information on our service.
Senator Klobuchar. Were you saying you take issue with
something that I had said or was it something----
Ms. Culbertson. No, no. No, Senator. No, Senator. One of
the other panelists----
Senator Klobuchar. Oh, okay.
Ms. Culbertson [continuing]. Suggested that we have a
competitive edge to compete on addressing these harms where we
actually collaborate in a lot of these areas.
Senator Klobuchar. Okay. Thank you very much. I appreciate
it.
Ms. Culbertson. Thank you.
Chair Durbin [presiding]. We now go to Senator Kennedy,
remote. Can you hear us, Senator Kennedy?
Senator Kennedy. I can hear you, Mr. Chairman. Can you hear
me?
Chair Durbin. Yes, I can. The time is yours. Take it away.
Senator Kennedy. Thank you. It seems to me that in the
guise of giving consumers what they want, a lot of our social
media platforms first use surveillance to identify a person's
hot buttons, and then they use algorithms to show that person
stuff that pushes those hot buttons. This is called, as you
know, optimizing for engagement. The social media platform
wants a person to visit its platform early and often. That is
how it makes more money advertising. In any event, when that
person that we are talking about, as a result of those
algorithms, gets all revved up with no place to go, he posts
something outrageous. Not every time, but quite frequently, and
that is why you can still find kindness in America, but you
have to go offline to do it.
Mr. Harris, I would like a straight answer from you. Would
you--I have a bill--others have a similar bill--a bill to say
that--that would say that Section 230 immunity will no longer
apply to a social media platform that optimizes for engagement.
Would you--if you were a Senator, would you vote for it?
Mr. Harris. I would have to see the way that the bill is
written, Senator.
Senator Kennedy. Do not do--do not do that to me, Mr.
Harris. Give me a straight answer. We all want to read the
bills. Would you vote for it or not?
Mr. Harris. I would--I would be in support of a bill that
had technology companies not measure as their primary mode of
success any of the engagement metrics--time spent, clicks
shared, et cetera.
Senator Kennedy. That is swell, but if the bill said--I do
not like to waste time in these hearings. If the bill said no
Section 230 immunity if you optimize for engagement, would you
vote for it? If you do not want to--if you do not want to
answer, just tell me.
Mr. Harris. It sounds like a very interesting directional
proposal. I just--I would have to know the details, but I am
sorry for not being more clear.
Senator Kennedy. You are being very clear. You are dodging
the answer. Ms.--Dr. Donovan, would you vote for it?
Dr. Donovan. Yes. When it comes to bills, the reason why I
am in research is so I do not have to make those decisions. I
would say that when we are talking about what these companies
optimize for and the way in which it is optimized----
Senator Kennedy. Doc. Doc. Doc. Doc.
Dr. Donovan. Yep. Please.
Senator Kennedy. Doc. Doc, would you vote? Would you vote
for the bill?
Dr. Donovan. I would--I would vote for some form of bill
that required oversight of these algorithmic systems.
Senator Kennedy. All right. I mean, we have these hearings
and I appreciate them, but we never get down to it. Everybody
just wants--you know, we all talk. I am as guilty as anyone
else. At some point you got to get down to it, and if you say--
and that is where I am coming from. I am not trying to be rude.
I am just trying to get an answer out of you. You have both
been very critical of what we have today. I am, too. I am
looking for solutions. I am not just looking to--for us all to
show how intelligent we are----
Dr. Donovan. I think that one of the things----
Senator Kennedy [continuing]. Or not.
Dr. Donovan [continuing]. That we could address, Senator--
--
Senator Kennedy. I appreciate it, Doc. I am going to run--I
am going to run out of time. Let me ask--I am thinking about
introducing a bill--in fact, we are working on it--to take the
principles of the general data protection regulation in the EU.
I never thought I would do something like this, but take the
principles and the general data protection regulation in the EU
and have that--those principles apply here in the United
States. Ms. Bickert, would you support that bill?
Ms. Bickert. Senator, I focus on content, but there are
people at our company, and we can have someone follow-up on
that.
Senator Kennedy. That is a dodge. Ms. Culbertson, would you
vote for it?
Ms. Culbertson. We certainly comply with GDPR. There are
some tensions with the First Amendment in the U.S., but we
would welcome a longer conversation about this. Generally, yes.
Senator Kennedy. Yes?
Ms. Culbertson. Yes, Senator.
Senator Kennedy. Oh, God bless you. God bless you. Thank
you for an answer. Ms.--I am sorry, but I am mispronouncing
your name. ``Vytch?'' ``Veetch?''
Ms. Veitch. Senator, it is ``Veetch,'' yes.
Senator Kennedy. I am sorry, Ms. Veitch. I apologize. Will
you vote--would you--if you were a Senator, would you vote for
it?
Ms. Veitch. Senator, I am not an expert on GDPR. I can tell
you on privacy, what we want to do is give our users security--
--
Senator Kennedy. I know. I know. You want--you want
privacy, but your whole model is built around finding out
everything you can about me, other than my DNA, and you may it,
for all I know. I am not trying to be rude, but I cannot tell
you the number of these hearings I have been to, and there--I
learn something every time. When we get down to it, what are we
going to do about it? Nobody wants to answer, and you are
supposed to be our experts. I would strongly encourage you to
come to these hearings with positions, firm positions, on
behalf of yourselves, or on behalf of your companies, that you
are ready to take. Do not just word whip us. We are trying to
solve a problem here. What----
Chair Durbin. Senator Kennedy.
Senator Kennedy. Yes, sir.
Chair Durbin. I have to ask you for a ``yes'' or ``no''
answer. Do you realize you have gone over time?
Senator Kennedy. I realize that--yes, and I realize
everybody else has gone over time.
[Laughter.]
Chair Durbin. Take another minute, and then please wrap it
up.
Senator Kennedy. I am done.
Chair Durbin. Thank you, sir. Senator Ossoff, remote?
Senator Ossoff. Thank you, Mr. Chairman. Thank you to the
panel. Ms. Bickert, much of the public discussion is focused on
Facebook's moderation practices, but there is a compelling
argument that the real problem is not the quality of your
moderation policies or the nature of the algorithm, but the
underlying business model, your scale, and your power. While
you clearly have an obligation to remove certain content, for
example, incitement to violence or hate speech, I am not at all
enthusiastic about huge multinational tech companies becoming
the arbiters of legitimate speech and expression, especially
when the decisions about what you may boost or suppress
algorithmically are often made in secret and under heavy
pressure from politicians, and advertisers, and public opinion.
On the subject of your scale and your power, I would like
to ask, does Facebook anticipate that it will embark on further
acquisitions of competitor services in light of the suit that
you are already facing from the FTC and a number of State
attorneys general alleging that your acquisitions of Instagram
and WhatsApp constituted anticompetitive activity?
Ms. Bickert. Senator, thank you for the question. Of course
I cannot comment on any litigation. I can tell you, because I
am responsible for our content policies and a lot of what we do
around moderation, that we do take very seriously both the
balance between expression and safety, but also the need for
transparency. So with, for instance, our algorithm, over the
past few years, we have put out a number of blog posts and
other communications where we have actually given the inputs
for what goes into the ranking algorithm. We have explained any
significant ranking changes. We have introduced this tool where
on any post on Facebook, you can click on it and go under ``why
am I seeing this post,'' and it will tell you why that is
appearing in your News Feed where it is. Then, significantly,
we have made it more visible how you can opt out of that News
Feed ranking algorithm.
If people just want to see their content in reverse
chronological--reverse chronological order--excuse me--they
can----
Senator Ossoff. Ms. Bickert, yes, respectfully, and I
greatly appreciate your response, and I heard some of these
points earlier in the hearing, and I am not asking you to
comment on any specific litigation. To be clear, my point is
actually that everything you just said about improving the
quality of your moderation practices, disclosing some of the
decisions underlying the algorithm, are not the root issue. The
root issue is that Facebook has too much power, and one company
perhaps should not be such a massive gatekeeper that determines
what ideas prosper and what ideas do not. That is why the
question that I asked was, does Facebook anticipate that it
will embark on any further acquisitions of competitor services.
Ms. Bickert. Senator, this is--acquisitions is really not
my area at all. I am focused on content. I can tell you,
though, from where I sit, from my perspective, it is a highly
competitive space, and I know that not only from, you know,
being the--being an executive working on content at Facebook,
but also being the parent of two teenage daughters, both of
whom use social media, and there are a lot of services out
there that people use. Nevertheless, I do think it is really
important that we recognize that these content modernation
rules are really important, and we have to be very transparent
about what they are so people can make informed choices about
whether or not they want to use our services.
Senator Ossoff. Thank you, Ms. Bickert. Ms. Bickert,
Apple's recent iOS update will require apps to seek additional
explicit authorization from users in order for those apps,
presumably some of your products included, to continue tracking
users across the internet. Tracking cookies and other
technologies allow Facebook and other entities to monitor
virtually all of their users' web browsing activity. I want to
commend Apple for taking this step and ask whether you will
take significant steps in the short term to reduce your
tracking--your ubiquitous tracking of your users' web activity,
location data, the technology that they use, and whether you
will consider extending the feature that allows the removal of
personal data from Facebook, to include the removal of personal
data not just from Facebook, but from any entities to whom
Facebook sold such data, and including in your contracts with
those to whom you sell data, a provision that they must delete
all data that they have purchased from Facebook at the command
of the user.
Again, it is two questions. Will you follow Apple's lead in
ceasing tracking of users across the web, and will you include
in contracts with those to whom you sell data a provision
requiring them to permanently delete and verify the deletion of
all data you have sold to them about any user who activates the
Facebook feature to remove their data from Facebook? Thank you
so much.
Ms. Bickert. Senator, thank you for the question. First,
let me be really clear. We do not sell user data. That is not
the way our advertising works. The way that it works is an
advertiser selects from among different targeting criteria, and
then we deliver that ad to a relevant audience. We can follow-
up with more details on how that works.
With respect to controls, I know we have introduced
controls around people's off-Facebook experience. I am not an
expert in that area. There are those in the company who are, so
I can get that information and follow-up with you.
Senator Ossoff. Thank you, Ms. Bickert, and thank you, Mr.
Chairman.
Chair Durbin. Thank you, Senator. Senator Blackburn, are
you available by remote?
Senator Blackburn. Yes, I am. Thank you, Mr. Chairman. I
appreciate the witnesses and the hearing today, and I think all
the witnesses are hearing that Americans are pretty much fed up
with the arrogance of Big Tech. You are seeing it from all
sides, and certainly Twitter's CEO, Jack Dorsey, was--had his
contempt for Congress on full display in the House Energy and
Commerce Committee hearing last--I think it was last month. He
tweeted out a poll on possible answers to the questions,
basically treating the hearing as a joke. Ms. Culbertson, do
you agree it is unacceptable for Twitter's CEO to tweet while
he is testifying before Congress? Yes or no.
Ms. Culbertson. Certainly, he is the CEO and creator of
Twitter, and he likes to tweet, and that is the way he
communicates.
Senator Blackburn. Okay. I asked for a ``yes'' and ``no,''
but I will say I am pleased you are looking and appearing more
presentable than your CEO in his testimonies before us. When he
behaves disrespectfully in a congressional hearing and before
the American people, he embarrasses Twitter. It is just such
proof of how out of touch Big Tech is with the rest of the
country. Big Tech is, in my opinion, destroying news, free
speech, competition, original content. It is responsible also
for much of our children's minds. This is something that
bothers me as a mom and a grandmom, the power of Facebook and
YouTube's algorithms to manipulate social media addiction. We
are even reading that it is among babies, toddlers, kids,
tweens, and teens, and this is something that should terrify
each of us.
YouTube deploys algorithms to breed this addiction,
clickbait in children, and they do it because it pays well. Our
children's brains are being trashed so if you Silicon Valley
CEOs can pocket billions of dollars in ad revenue. YouTube
algorithms create an unpoliced automated reward system. Videos
with little educational content are amplified to unsuspecting
toddlers and kids and to their unsuspecting parents. Senator
Thune mentioned that we are reintroducing the bipartisan Filter
Bubble Transparency Act to force Big Tech to disclose if their
secret algorithms are manipulating customers.
Ms. Veitch, YouTube has a history of exploiting children to
harvest and profit off of their viewing history. Is it not true
YouTube has illegally collected data on kids under age 13 in
violation of COPA, and marketed that data to companies? Ms.
Veitch.
Ms. Veitch. Thanks for the question, Senator. I am familiar
with COPA that you are referring to. That was a novel
interpretation of COPA. We worked directly with the FTC to
reach an agreement about how we treat made-for-kids content on
YouTube main. We do----
Senator Blackburn. Okay. Ms. Veitch, Yes, you reached an--a
settlement in 2019. You were fined a record $170 million. Do
you recall that?
Ms. Veitch. Yes, ma'am.
Senator Blackburn. Okay. The FTC order does not require
YouTube to police the channels that deceive by mis- designating
their content. However, Commissioner Slaughter said YouTube
should have to take the extra step of creating an algorithmic
classifier to better police YouTube content for kids. I know
your engineers are capable of designating and designing
algorithms for all sorts of purposes, good and evil. Let me ask
you this. Is the YouTube engineering team capable of designing
an algorithm that can identify a designated and child-directed
content and turn off behavioral advertising?
Ms. Veitch. Senator, they are capable of that, and they
have done that. We do require creators to designate their
content as made for kids or not, but we also run classifiers,
as you mentioned, to check that system and to determine what
content is appropriate to be made for children and serve to
children. We also, just to be clear, Senator, do not allow
personalized advertising on made-for-kids content.
Senator Blackburn. Are you prioritizing profit over
children?
Ms. Veitch. No, Senator. Child safety on our platform is
our top priority. We build our product with parental controls
baked right in, things like timers----
Senator Blackburn. The FTC is prioritizing children and
taking steps to safeguard them. Under the settlement, you
promised to stop illegally marketing targeted ads to children.
Videos now have been labeled as made for kids, as you just
mentioned. So, and made-for-kid videos will no longer include a
comment section or in screens that allow viewers to subscribe
to children. Are you allowing this behavioral advertising to be
turned off?
Ms. Veitch. Yes, Senator. We do not serve personalized
advertisements on made-for-kids content.
Senator Blackburn. Okay. I am over my time. Ms. Bickert, I
have a question for you I will submit for the record. Thank
you, Mr. Chairman.
Chair Coons [presiding]. Thank you, Senatr Blackburn.
Senator Blumenthal.
Senator Blumenthal. Thank you, Mr. Chairman. Thank you to
all of our witnesses for being part of this hearing and to the
Chairman for holding it. It is a very, very critically
important topic and hearing, and I apologize that I am late
coming here because I was chairing a Subcommittee of Commerce
on Consumer Protection dealing with COVID scams.
I am very proud that last week, the United States Senate
approved the bipartisan Jabara-Heyer NO HATE Act, which I led
alongside Senator Moran. We have known for a long time that
hate crimes are on the rise. They are exploding in this very
polarized and vitriolic time. Viral videos of individual crimes
posted by--on Facebook, Twitter, and YouTube, no matter how
horrifying or stomach-turning, really tell only part of the
story. The NO HATE Act will improve hate crime reporting
because so many of them are invisible and unreported, and it
will expand assistance and resources for victims of hate crimes
as well as for law enforcement, and hopefully will enable us to
understand the full scope of the problem so that we can take
more effective action against hate crimes.
We know that the tech platforms play a role in hate crimes
and hate speech online and off. The Anti-Defamation League
recently found that as many as one in three Americans
experience hate crimes and harassment online. Following the
ADL's concern--very concerning findings, I teamed up with
Representative Raskin to request a Government Accountability
Office study specifically on the prevalence of online hate
crimes and hate speech in the United States. During the 2020
election, Facebook spoke about the break-the-glass measures it
was taking to ``dial down'' the hate, incitements to violence,
and misinformation on its platform. Last week, Ms. Bickert, you
wrote a blog post about turning the dial down on hate speech,
graphic violence, violence and incitement, as the country was
anticipating the verdict in the Chauvin trial.
If Facebook does, in fact, have a dial for hateful content,
can the company dial it down now? Why does not it dial it down
already? To all of the representatives who are here today from
YouTube, Twitter, as well as Facebook, can you commit to
providing access data to independent researchers to help us
better understand and address the scourge of hate and
harassment online?
Ms. Bickert. Senator, thank you, and let me start by saying
I completely agree that the rise of hate speech and hate crimes
is very concerning and needs to be a priority for us and is a
priority for us. I will just point to one quick example, which
is we have started publishing the prevalence in our quarterly
reports that we put out, our community standards enforcement
reports. We now publish the prevalence of hate speech, which
means we go through with a fine-tooth comb and see what we
missed for a significant--statistically significant subset of
content. The prevalence of hate speech on our service is very
low--less than a 10th of 1 percent--but it is something that we
are really focused on finding. I am happy to say that more than
95 percent of the content that we removed for hate speech
violations we find ourselves before anybody reports it to us,
so we are making strides.
To respond to your point about why we do not--the measures
we took around the Chauvin trial and the election, why we do
not always do that, let me--let me sort of give you an example
of the cost of those measures because they have benefits, but
they have costs. In the--in the run-up to the election, for
instance, we took some very aggressive measures to reduce the
distribution of content that might be violating our policies.
We did that with the Chauvin trial as well. Those measures are
not perfect, and so there will be content that actually does
not violate our policies that was flagged by our technology
that really should not be reduced.
When we take those measures, we are mindful of the cost. It
is always this balance between trying to stop abuse and trying
to make sure that we are providing space for freedom of
expression and being very fair. We take those measures where
there is a risk of false positives only when there is an
additional risk of abuse.
Senator Blumenthal. Thank you. My time has expired. Thank
you very much. Thanks, Mr. Chairman.
Chair Coons. Thank you so much, Senator Blumenthal. We are
going to have a second round of questioning that may be
participated in only by the Ranking Member and myself given
that there are votes actively ongoing on the floor. Let me
thank our five witnesses again and the many Members of this
Subcommittee who have come to question.
Ms. Veitch, I understand that 70 percent of the views on
YouTube come by--are driven by its recommendation algorithm.
With 2 billion users worldwide and over 1 billion hours of
video watched each and every day, that makes your
recommendation algorithm incredibly powerful. Members of the
public can see how many times any video has been viewed, but
members of the public cannot see how many times that video has
been recommended, though I understand YouTube does collect this
information and gives it to content providers. If a video ends
up getting taken down by YouTube for violating its content
policies, we have no way of knowing how many times it was
recommended by your algorithm before it was ultimately removed.
Could YouTube commit today to providing more transparency about
your recommendation algorithm and its impacts?
Ms. Veitch. Thanks for this question, Senator. Just
generally speaking, if content violates our policies, we want
to remove it as quickly as possible. As you will see in our
Public Community Guidelines Enforcement Report, of the 9.3
million videos we removed in Quarter 4 of 2020, more than 70
percent were removed before they had 10 views. I think you have
brought up an interesting idea, and we are always looking to
expand transparency when it comes to our platform. One way we
have done this recently is by making public what we call our
violative view rate. It is the percentage of views on our
platform that violate our community guidelines. Last quarter
they were between .16 and----
Chair Coons. Ms. Veitch? Ms. Veitch, if I might, I just
want to know if you are willing to release the data. I believe
you are already collecting about how many times videos that
violate your content standards have been recommended by your
recommendation algorithm.
Ms. Veitch. Thank you, Senator. I cannot commit to really
saying that today, but it is an interesting idea. We want to be
more transparent, so let us work with you on that.
Chair Coons. Thank you. I look forward to getting an answer
as soon as is reasonably possible. Ms. Bickert, several
publications have reported that significant portions of
misinformation and polarizing content on Facebook comes from
readily identifiable, hyper-active users, or super inviters,
who generate a lot of activity on your platform. Dr. Donovan
can you comment briefly on how these hyper-active users create
problems, and then, Ms. Bickert, I want to ask about whether or
not Facebook intends to tackle this challenge. Dr. Donovan.
Dr. Donovan. Yes, you are referring to the BuzzFeed article
that reported on an internal memo from Facebook that showed
that there is a power law at play where it skews to highly--you
know, misinformation tends to be most potent when you have a
densely networked and highly coordinated small group of people
working essentially around the clock to try to get their groups
stocked with the public. What has been interesting about
reading the document internal to Facebook is that, even as they
tried to counter super inviters, their own internal systems and
teams were not able to overcome that coordinated small network.
There is a lot that the company needs to do to address
adversarial movements, and in this case was--they were looking
at the formation of Stop the Steal groups and the Patriot
Party.
Chair Coons. Thank you, Dr. Donovan. Ms. Bickert, the Wall
Street Journal reported last year that Facebook considered
seriously, but ultimately declined, to take measures that would
put limits on these users' activities. There was a proposal
reportedly called Sparing Sharing, which would have reduced the
spread of content that was disproportionately favored by these
so-called hyper-active users. Could you speak to how Facebook
is intending to approach this issue?
Ms. Bickert. Yes, Senator, and let me say we did actually
put a restriction, a limit on the number of invites that an
individual user could send out in a day to a group during the
election period. I want to speak also to the point that Dr.
Donovan raised, and I completely agree. There are--there are
networks of bad actors, who are particularly sophisticated, who
try to target--use social media to achieve their objectives.
Understanding the way that those networks work has been
something that we have really been focused on in the past few
years, building a team under Nathaniel Gleicher, who has got
expertise in this area and I know knows Dr. Donovan as well.
In terms of identifying these sophisticated actors, who are
often engaged in shell games and, you know, other attempts to
sort of obfuscate what they are doing using inauthentic
identities, we have gotten far better at that. We have removed
more than 100 such networks since the beginning of 2017. We are
public about it when we do. We publish the results of those,
and we have also gotten better generally at identifying fake
accounts. We removed more than 1 million fake accounts at the
time--up or near the time of upload every day now.
Chair Coons. Thank you. I look forward to delving into this
further with you and with other folks at Facebook. Let me just
ask two more--maybe three more, Mr. Ranking Member. A quick
just structural question. I know it is common for employees at
major tech companies to be required to sign nondisclosure
agreements as a condition of employment. When I was in the
private sector, that was a common practice in the businesses
that I knew about and practiced for. Ms. Bickert, Ms.
Culbertson, Ms. Veitch, do each of your companies generally
require your employees to sign NDAs? It strikes me as a ``yes''
or ``no'' question.
Ms. Bickert. I will go first, Senator. I do not know the
answer, but I will follow-up--I will have the team follow-up
with you.
Ms. Veitch. Senator, I want to be careful here because I am
not a lawyer--an employment lawyer, but I do believe that we
have standard agreements to protect proprietary information
with our employees.
Ms. Culbertson. I would want to come back to you with the
answer, but, of course, we have certain provisions in place to
make sure people are not sharing private data they might be
handling. I would just say generally, the Twitter-y spirit
among our employees is to share their perspectives. You will
oftentimes see our employees tweeting about our different
products and services.
Chair Coons. Thank you. In general, my concern is that if a
former employee from one of your companies wants to question or
criticize the company or its decision making, that they might
risk facing legal action. Mr. Harris, Dr. Donovan, I would
welcome some more input from you following this hearing on that
dynamic, and whether or not NDAs actually prevent some of the
most relevant information about algorithms from getting out to
the general public.
Two last questions, if I might, one on transparency. I
appreciation--I appreciate the information that has been shared
today about how algorithms work at a high level. Many
independent researchers have said it is critical to know the
details, the dials and knobs of algorithms to understand how
components that drive decisions are weighted, so how much a
metric, like meaningful social interaction, is actually
correlated with growth and engagement, which, as Mr. Harris has
repeatedly asserted, and as I fundamentally believe, the
business model of social media requires you to accelerate.
Given the immense impact of the knobs and dials of your
algorithms in potentially both positive and negative ways, I
think greater transparency about those matters, about how your
algorithms actually work and about how you make decisions about
your algorithms, is critical.
Ms. Bickert, Ms. Veitch, Ms. Culbertson, could you speak to
whether your companies are considering the release of more
details about this kind of information or other types of
enhanced transparency measures or audits about the impact of
your companies' algorithms moving forward?
Ms. Culbertson. I am happy to start. We are constantly
thinking about how we can be more transparent about any actions
that we take or our systems in place, including our algorithms.
That is why we are investing in our responsible machine
learning initiative. We would be happy to provide more details
in the interest of time.
We have an interdisciplinary group at Twitter looking at
our algorithms, our machine learning, studying these--our
machine learning. We will also be sharing some of our findings
with the open--with the public so we can be open throughout
this process.
Then just more broadly, we totally agree that we should be
more transparent. We should also provide more consumer control
and choice. We are also committed to improving procedural
fairness. To those first two points, we have invested in this
independent project called BlueSky, which is aimed at creating
open protocols, which would essentially potentially create more
controls for the people who use our services as well as
transparency.
Chair Coons. Thank you. My last comment will be just this
one. Mr. Harris spoke forcefully and pointedly about how the
business model of social media's attention harvesting, and that
after a decade of the positive and negative impacts of social
media, which has accelerated to be one of the most important
forces in our society today, that we have more than not seen
the toxic impacts of division and disinformation. Mr. Harris
has asserted that your entire business model is based on
dividing society, and that as we transition into a digitized
society in the 21st century, in order for Western open
democratic societies to survive, we have to develop model
humane standards for how social media works. It is my hope, I
will share with my Ranking Member, that the next time we
convene, it might be to consider what sorts of steps are
possible, necessary, or appropriate to make that progress that
Mr. Harris speaks about.
To my Ranking Member, Senator Sasse.
Senator Sasse. Thank you, Chairman Coons, and, again, thank
you to all five of you for being here. I do--I do want to put
another question to Mr. Harris, but before I start that second
round of questioning, I would like to just briefly address
colleagues on both sides of the aisle because both Republican
and Democratic colleagues today have said a number of things
that presumed more precision about the problem than we have
actually identified here, and then sort of picked up the most
ready tool, usually the 230 discussion.
I am--I think I am a lot more skeptical than maybe most on
this Committee to push to a regulatory solution at this stage.
I think, in particular, some of the conversations about Section
230 have been well off point to the actual topic at hand today.
I think much of the zeal to regulate is driven by short-term
partisan agendas, and I think it would be more useful for us to
stick closer to the topic that the Chairman identified from
this hearing.
I also think it is important for Members of Congress to
constantly remind ourselves that we are bound by First
Amendment constraints in our job, and many--a number of the
lines of questioning, again, on both the right and left sides
of this panel today talked as if the First Amendment is sort
of, you know, this marginal topic that we do not have to be
obsessively concerned about, and yet we need to draw a
distinction between the First Amendment and the true public
square as regulated by the powers of the government, and the
fact that the companies we are talking about--Amy Klobuchar has
raised some important topics about scale and antitrust issues,
but the companies we are talking about are private companies. I
do think there are a number of First Amendment, public-private
distinctions that we should--we should be attending to a little
more closely than maybe we did today.
Mr. Harris, can you tell us what discussions you have seen
or been a part of, either inside the extant companies or, you
know, at the VCPE environment, potential different business
models besides an ad revenue-centric business model? Can you
just give us a kind of blue sky on that question?
Mr. Harris. Yes, fantastic question. Thank you, Senator. I
mean, obviously there are subscription models. There are public
interest models, more like Wikipedia. I want to make an
additional distinction, which is not just the funding model,
but it is the design model. The engagement and advertising
model works because of the design that says, user-generated
content, we are all the unpaid journalists. Previously, you had
to pay a journalist at Fox News and New York Times $100,000 a
year to write content to get people to look at it, and that is
the cost of attention production. What if you could harvest
each of us as useful idiots to take our 5 minutes of moral
outrage, and then use that to generate attention production for
free? We are the unpaid labor, like, for the attention economy
that is sort of duped into sharing information with each other,
which reduces the costs for all these technology companies.
Then on the editorial side, instead of paying an editor at
a New York Times, at a Fox News, at a whatever, $100,000 a
year, $200,000 a year, we actually have algorithms, which we
also do not have to pay, to randomly sort that to people. This
happens in a values-blind process, which means that, in
general, you get harm showing up in all of the blind spots.
Suddenly Joan wakes up and says, hey, there is this problem,
there is this problem, this problem. The companies will respond
and say, okay, fine, we will take the whack-a-mole stick and we
will deal with those three problems.
In general, values blindness destroys our democracy faster
than people like, you know, Renee, and, Joan, and so many of
our friends in this community are essentially raising the
alarms about it, and that is fundamentally the kind of core
design model more so than the funding model. We could have
public interest technology that is funded for public interest.
We could tax these companies to put into a regenerative fund.
There are a whole bunch of models we could do. One in energy
just like--you know, energy companies have this perverse
incentive where they make more money the more energy you use,
so theoretically, leave the lights on, leave the faucets on, we
make more money. They do not--they do not do that because,
instead, they have a model that is regulated, so that after a
certain amount, they double charge you, triple charge you to
disincentivize your energy use. Then that, instead of going
into the private business model of the company, the balance
sheets, it gets put into a regenerative fund to increase the
transition to solar.
Imagine that technology companies, which today profit from,
you know, an infinite amount of engagement, only made money
from a small portion of that, let us say some small amount of
time, and the rest basically with tax to put into a
regenerative public interest fund, and this funded things like
the Fourth Estate, fact checkers, researchers, public interest,
technologists, things like that because what we really need to
do is, as we said, organize a comprehensive shift to more
humane technology, so a digital open society can compete with
digital closed societies.
Senator Sasse. Very helpful. There is a--if we had--if we
had more time, I was going to ask some questions about whether
or not you think there are--given your role as an ethicist,
whether or not there are debates inside the company about what
the optimal user time on a site is every given day, and is
there a distinction between, you know, a fully consenting, -
assenting 49-year-old, like myself, and how those platforms
think about it for a 13-year-old and a 17-year-old as well. I
know that Chris and I both need to go and vote, so I will just
echo the thanks to all five of you for the fulsome discussion
today and to be continued.
Chair Coons. Thank you, Senator Sasse. Let me conclude by
thanking all five of our witnesses for appearing today and to
my 11 colleagues who have appeared and engaged in robust
questioning. I appreciate, in particular, the willingness of
witnesses from Facebook, Twitter, and YouTube to answer some
direct and difficult questions about their platforms and their
business models. I am encouraged to see that these are topics
that are broadly of interest and where, I believe, there could
be a broadly bipartisan solution.
None of us wants to live in a society that, as a price of
remaining open and free, is hopelessly politically divided or
where our kids are hooked on our phones--their phones and being
delivered a torrent of reprehensible material. I also am
conscious of the fact that we do not want to needlessly
constrain some of the most innovative, fastest-growing
businesses in the West. Striking that balance is going to
require more conversation, and I look forward to continuing to
work with Ranking Member Sasse on these matters, whether by
roundtable or additional hearing, and whether by seeking
voluntary reforms, regulation, or legislation. That includes
exploring how best to align incentives, both within companies
and with the rest of our society, to ensure greater
transparency and user choice, and I think we have to approach
these challenging and complex issues with both humility and
urgency. The stakes demand nothing less.
Members of the Committee may submit questions for the
record for the witnesses. They are due by 5 p.m., 1 week from
today, on May 4th.
With that, this hearing is adjourned.
[Whereupon, at 12:18 p.m., the hearing was adjourned.]
[Additional material submitted for the record follows.]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
[all]