[Senate Hearing 116-609]
[From the U.S. Government Publishing Office]
S. Hrg. 116-609
MASS VIOLENCE, EXTREMISM,
AND DIGITAL RESPONSIBILITY
=======================================================================
HEARING
before the
COMMITTEE ON COMMERCE,
SCIENCE, AND TRANSPORTATION
UNITED STATES SENATE
ONE HUNDRED SIXTEENTH CONGRESS
FIRST SESSION
__________
SEPTEMBER 18, 2019
__________
Printed for the use of the Committee on Commerce, Science, and
Transportation
[GRAPHIC NOT AVAILABLE IN TIFF FORMAT]
Available online: http://www.govinfo.gov
______
U.S. GOVERNMENT PUBLISHING OFFICE
52-751 PDF WASHINGTON : 2023
SENATE COMMITTEE ON COMMERCE, SCIENCE, AND TRANSPORTATION
ONE HUNDRED SIXTEENTH CONGRESS
FIRST SESSION
ROGER WICKER, Mississippi, Chairman
JOHN THUNE, South Dakota MARIA CANTWELL, Washington,
ROY BLUNT, Missouri Ranking
TED CRUZ, Texas AMY KLOBUCHAR, Minnesota
DEB FISCHER, Nebraska RICHARD BLUMENTHAL, Connecticut
JERRY MORAN, Kansas BRIAN SCHATZ, Hawaii
DAN SULLIVAN, Alaska EDWARD MARKEY, Massachusetts
CORY GARDNER, Colorado TOM UDALL, New Mexico
MARSHA BLACKBURN, Tennessee GARY PETERS, Michigan
SHELLEY MOORE CAPITO, West Virginia TAMMY BALDWIN, Wisconsin
MIKE LEE, Utah TAMMY DUCKWORTH, Illinois
RON JOHNSON, Wisconsin JON TESTER, Montana
TODD YOUNG, Indiana KYRSTEN SINEMA, Arizona
RICK SCOTT, Florida JACKY ROSEN, Nevada
John Keast, Staff Director
Crystal Tully, Deputy Staff Director
Steven Wall, General Counsel
Kim Lipsky, Democratic Staff Director
Chris Day, Democratic Deputy Staff Director
Renae Black, Senior Counsel
C O N T E N T S
----------
Page
Hearing held on September 18, 2019............................... 1
Statement of Senator Wicker...................................... 1
Statement of Senator Cantwell.................................... 3
Statement of Senator Fischer..................................... 36
Statement of Senator Blumenthal.................................. 38
Statement of Senator Thune....................................... 39
Statement of Senator Blackburn................................... 41
Statement of Senator Scott....................................... 43
Statement of Senator Duckworth................................... 45
Statement of Senator Young....................................... 48
Statement of Senator Rosen....................................... 51
Statement of Senator Lee......................................... 52
Statement of Senator Baldwin..................................... 54
Statement of Senator Sullivan.................................... 56
Statement of Senator Cruz........................................ 58
Witnesses
Monika Bickert, Vice President for Global Policy Management and
Counterterrorism, Facebook..................................... 4
Prepared statement........................................... 6
Nick Pickles, Director, Public Policy Strategy, Twitter, Inc..... 8
Prepared statement........................................... 10
Derek Slater, Global Director, Information Policy, Google LLC.... 15
Prepared statement........................................... 17
George Selim, Senior Vice President, National Programs, ADL
(Anti-Defamation League)....................................... 19
Prepared statement........................................... 21
Appendix
Letter dated September 16, 2019 to Chairman Roger Wicker and
Ranking Member Maria Cantwell from Gretchen Peters, Executive
Director, and Dr. Amr Al-Azm, Co-founder, Director of ATHAR
Project, Alliance to Counter Crime Online...................... 63
Letter dated September 17, 2019 to Mark Zuckerberg, Chief
Executive Officer, and Sheryl Sandberg, Chief Operating
Officer, Facebook; Jack Dorsey, Chief Executive Officer,
Twitter; Sundar Pichai, Chief Executive Officer, Google; and
Susan Wojcicki, Chief Executive Officer, YouTube from The
Leadership Conference on Civil and Human Rights, Color Of
Change, Lawyers' Committee for Civil Rights Under Law, Muslim
Advocates, and NAACP Legal Defense and Educational Fund, Inc... 64
Response to written questions submitted to Monika Bickert by:
Hon. Shelley Moore Capito.................................... 66
Hon. Amy Klobuchar........................................... 67
Hon. Richard Blumenthal...................................... 68
Hon. Tom Udall............................................... 70
Hon. Jacky Rosen............................................. 75
Response to written questions submitted to Nick Pickles by:
Hon. Shelley Moore Capito.................................... 76
Hon. Richard Blumenthal...................................... 77
Hon. Tom Udall............................................... 80
Hon. Jacky Rosen............................................. 83
Response to written questions submitted to Derek Slater by:
Hon. Shelley Moore Capito.................................... 85
Hon. Richard Blumenthal...................................... 86
Hon. Tom Udall............................................... 89
Hon. Jacky Rosen............................................. 91
Response to written questions submitted to George Selim by:
Hon. Richard Blumenthal...................................... 93
Hon. Tom Udall............................................... 95
Hon. Jacky Rosen............................................. 95
MASS VIOLENCE, EXTREMISM,
AND DIGITAL RESPONSIBILITY
----------
WEDNESDAY, SEPTEMBER 18, 2019
U.S. Senate,
Committee on Commerce, Science, and Transportation,
Washington, DC.
The Committee met, pursuant to notice, at 10 a.m., in room
SH-216, Hart Senate Office Building, Hon. Roger Wicker,
Chairman of the Committee, presiding.
Present: Senators Wicker [presiding], Thune, Cruz, Fischer,
Sullivan, Blackburn, Young, Capito, Lee, Scott, Cantwell,
Blumenthal, Udall, Duckworth, and Rosen.
OPENING STATEMENT OF HON. ROGER WICKER,
U.S. SENATOR FROM MISSISSIPPI
The Chairman. Today the Committee gathers to discuss what
the technology industry is doing to remove violent and
extremist content from their platforms. This is a matter of
serious importance to the safety and well-being of our Nation's
communities. I sincerely hope we can engage in a collaborative
discussion about what more can be done, within the jurisdiction
of this committee, to keep our communities safe from those
wishing to do us harm. Today, we welcome representatives from
the world's largest social media companies and online
platforms.
We will hear from Ms. Monika Bickert, Head of the Global
Policy Management for Facebook, and Mr. Nick Pickles, Public
Policy Director at Twitter, Mr. Derek Slater, Global Director
of Information Policy at Google, and Mr. George Selim, Senior
Vice President of Programs for the Anti-Defamation League. Over
the past two decades, the United States has led the world in
the development of social media and other services that allow
people to connect with one another.
Open platform providers like Google, Twitter, and Facebook
and products like Instagram and YouTube have dramatically
changed the way we communicate and have been used positively in
providing spaces for like-minded groups to come together and in
shedding light on despotic regimes and abuses of power
throughout the world. No matter how great the benefits to
society these platforms provide, it is important to consider
how they can be used for evil at home and abroad. On August 3,
2019, 20 people were killed, and more than two dozen were
injured in a mass shooting at an El Paso shopping center.
Police have said that they are reasonably confident that the
suspect posted a manifesto to a website called 8chan, 27
minutes prior to the shooting. 8chan moderators removed the
original post, though users continued sharing copies.
Following the shooting, President Trump called on social
media companies to work in partnership with local, State, and
Federal agencies to develop tools that can detect mass shooters
before they strike--I certainly hope we talk about that
challenge today. Sadly, the El Paso shooting is not the only
recent example of mass violence with an online dimension.
On March 15, 2019, 51 people were killed and 49 were
injured in shootings at two mosques in Christchurch, New
Zealand. The perpetrator filmed the attacks using a body camera
and live-streamed the footage to his Facebook followers, who
began to re-upload the footage to Facebook and other sites.
Access to the footage quickly spread and Facebook stated that
it removed 1.5 million videos of the massacre within 24 hours
of the attack. 1.2 million views of the videos were blocked
before they could be uploaded. Like the El Paso shooter, the
Christchurch shooter also uploaded a manifesto to 8chan. The
2016 shooting at the Pulse Nightclub in Orlando, Florida,
killed 49 and injured 53 more. The Orlando shooter was
reportedly radicalized by ISIS and other jihadist propaganda
through online sources. Days after the attack, the FBI Director
stated that investigators were highly confident that the
shooter was self-radicalized through the internet.
According to an official involved in the investigation,
analysis of the shooter's electronic devices revealed that he
had consumed ``a hell of a lot of jihadist propaganda,''
including ISIS beheading videos. Shooting survivors and family
members of victims brought a Federal lawsuit against those
three social media platforms under the Anti-Terrorism Act. The
Sixth Circuit dismissed the lawsuit on the grounds that this
was not an act of international terrorism. With over 3.2
billion Internet users, this committee recognizes the challenge
facing social media companies and online platforms and their
ability to act and remove content threatening violence from
their sites.
There are questions about tracking of a users' online
activity: does this invade an individual's privacy, thwart due
process, or violate constitutional rights? The automatic
removal of threatening content may also impact an online
platform's ability to detect possible warning signs. Indeed,
the First Amendment offers strong protections against
restricting certain speech. This undeniably adds to the
complexity of our task. I hope these witnesses will speak to
these challenges and how their companies are navigating these
challenges. In today's internet-connected society,
misinformation, fake news, deep fakes, and viral online
conspiracy theories have become the norm. This hearing is an
opportunity for witnesses to discuss how their platforms go
about identifying content and material that threatens violence
and poses a real and potentially immediate danger to the
public.
I hope our witnesses will also discuss how their content
moderation processes work. This includes addressing how human
review or technological tools are employed to remove or
otherwise limit violent content before it is posted, copied,
and disseminated across the internet. Communication with law
enforcement officials at the Federal, State, and local levels
is critical to protecting our neighborhoods and communities. We
would like to know how companies are coordinating with law
enforcement when violent or extremist content is identified.
And finally, I hope witnesses will discuss how Congress can
assist in ongoing efforts to remove content promoting violence
from online platforms and whether best practices or industry
codes of conduct in this area would help increase safety, both
online and offline. So, I look forward to hearing testimonies
from our witnesses, and hope we engage in a constructive
discussion about potential solutions to a pressing issue. And I
am delighted at this point to recognize my friend and Ranking
Member, Senator Cantwell.
STATEMENT OF HON. MARIA CANTWELL,
U.S. SENATOR FROM WASHINGTON
Senator Cantwell. Thank you, Mr. Chairman, and thank you
for holding this important hearing and for our witnesses being
here this morning. Across the country, we are seeing and
experiencing a surge of hate and as a result we need to think
much harder about the tools and resources we have to combat
this problem both online and offline. While the First Amendment
to the Constitution protects free speech, speech that incites
eminent violence is not protected, and Congress should review
and strengthen laws that prohibit threats of violence,
harassment, stalking, and intimidation to make sure that we
stop the online behavior that does incite violence.
In testimony before the Senate Judiciary Committee in July,
Federal Bureau of Investigation FBI Director Chris Wray said
that the white supremacist violence is on the rise. He said the
FBI takes this threat ``extremely seriously,'' and has made
over 100 arrests so far this year. We are seeing in my state
over the last several years, we have suffered a shooting at the
Jewish community center in Seattle, a shooting of a Sikh in
Kent, Washington, a bombing attempt at the Martin Luther King
Day parade in Spokane, and over the last year, we have seen a
rise in the desecration of both synagogues and mosques. The
rise in hate across the country has also led to multiple mass
shootings, including the Tree of Life congregation in
Pittsburgh, the Pulse nightclub in Orlando, and most recently,
the Walmart in El Paso.
Social media is used to amplify that hate and the shooter
at one high school in the Parkland posting said the image of
himself with guns and knives on Instagram wrote social media
posts prior to the attack on his fellow students. In El Paso,
the killer published a white supremacist anti-immigration
manifesto on a 8chan message board, and my colleague just
mentioned this streaming of live content related to the
Christchurch shooting and the horrific incidents that happened
there. In Miramar, the military engaged in a systematic
campaign on Facebook, using fake names and sham accounts to
promote violence against Muslim Rohingya. These human lives
were all cut short by deep hatred and extremism that we have
seen has become more common.
This is a particular problem on the dark web, where we see
certain websites like 8chan and a host of 24/7, 365 hate
rallies. Adding technology tools to mainstream websites to stop
the spread of these dark websites are a start, but there needs
to be more to be a comprehensive and coordinated effort to
ensure that people are not directed into these cesspools. I
believe calling on the Department of Justice to make sure that
we are working across the board on an international basis with
companies as well to fight this issue is an important thing to
be done. We do not want to push people off of social media
platforms only to then being on the dark web, where we are
finding less of them. We need to do more at the Department of
Justice to shut down these dark websites, and social media
companies need to work with us to make sure that we are doing
this. I do want to mention, just last week, as there is much
discussion here in Washington about initiatives.
The State of Washington has passed three gun initiatives by
the vote of the people, closing background loopholes, and also
relating to private sales and extreme person laws, all voted on
by a majority of people in our state and successfully passed.
So I do appreciate, just last week representatives from various
companies of all sizes in the tech industry sending the Senate
a letter, asking for passage of bills requiring extensive
background checks.
So very much appreciate that and your support of extreme
person laws to keep guns out of the hands of people who a court
has determined are dangerous in the possession of that. So this
morning, we look forward to asking you about ways in which we
can better fight these issues. I do want us to think about ways
in which we can all work together to address these issues. I
feel that working together, these are successful tools that we
can deploy in trying to fight extremism that exists online.
Thank you, Mr. Chairman, for the hearing.
The Chairman. Thank you, very, very much. And now we will
hear oral testimony from our four witnesses. And we ask you--
your entire statements will be submitted for the record,
without objection. We ask you to limit your comments at this
point to five minutes. Ms. Bickert, you are recognized. Thank
you for being here.
STATEMENT OF MONIKA BICKERT, VICE PRESIDENT FOR GLOBAL POLICY
MANAGEMENT AND COUNTERTERRORISM, FACEBOOK
Ms. Bickert. Thank you, Chairman Wicker, Ranking Member
Cantwell, and distinguished members of the Committee. Thank you
for the opportunity to be here today, and to answer your
questions and explain our efforts in these areas. My name is
Monika Bickert and I am Facebook's Vice President for Global
Policy Management and Counterterrorism. I am responsible for
our rules around content on Facebook and our company's response
to terrorist with the intent to use our services. On behalf of
everyone at Facebook, I would like to begin by expressing my
sympathy and solidarity with the victims, families,
communities, and everybody affected by the recent terrible
attacks across the country.
In the face of such heinous acts, we remain committed to
assisting law enforcement and standing with the community
against hate and violence. We are thankful to be able to
provide a way for those affected by this horrific violence to
communicate with loved ones, organize events for people to
gather and grieve, raise money to help support communities, and
begin to heal.
Our mission is to give people the power to connect with one
another and to build community. But we know that people need to
be safe in order to build that community. And that is why we
have rules in place against harmful conduct including hate
speech and inciting violence. Our goal is to ensure that
Facebook is both a place where people can express themselves,
but where they are also safe.
While we are not aware of any connection between the recent
attacks and our platform, we certainly recognize that we all
have a role to play in keeping our community safe. That is why
we remove content that encourages real-world harm, this
includes contents that is involving violence or incitement,
promoting or publicizing crime, coordinating harmful
activities, or encouraging suicide or self-injury. We do not
allow any individuals or organizations who proclaim a violent
mission, advocate for violence, or are engaged in violence to
have any presence on Facebook, even if they are talking about
something unrelated, this includes organizations and
individuals involved in or advocating for terror activity,
domestic and international, organized hate and that includes
white supremacy, white separatism, or white nationalism, or
other violence.
We also do not allow any content posted by anyone that
praises or supports these individuals, organizations, or their
actions. When we find content that violates our standards, we
remove it promptly, we also disable accounts when we see severe
or repeated violations, and we work with law enforcement
directly when we believe there is a risk of physical harm or a
direct threat to public safety. While there is always room for
improvement, we already remove millions of pieces of content
every year for violating our policies and much of that is
before anybody has reported it to us. Our efforts to improve
our enforcement of these policies are focused in three areas.
First, building new technical solutions that allow us to
proactively identify content that violates our policies.
Second, investing in people who can help us implement these
policies. At Facebook, we now have more than 30,000 people
across the company who are working on safety and security
efforts, this includes more than 350 people whose primary focus
is counterhate and counterterrorism.
And third, building partnerships with other companies,
civil society, researchers, and Governments so that together we
can come up with shared solutions. We are proud of the work we
have done thus far to make Facebook a hostile place for those
engaged in or advocating for acts of violence, but the work
will never be complete.
We know that bad actors will continue to attempt to skirt
detection with more sophisticated efforts, and we are dedicated
to continuing to advance our work and show our progress. We
look forward to working with the Committee, regulators, others
in the tech industry, and civil society to continue this
progress.
Again, I appreciate the opportunity to be here today, and I
look forward to your questions. Thank you.
[The prepared statement of Ms. Bickert follows:]
Prepared Statement of Monika Bickert, Vice President for Global Policy
Management and Counterterrorism, Facebook
I. Introduction
Chairman Wicker, Ranking Member Cantwell, and distinguished members
of the Committee, thank you for the opportunity to appear before you
today. My name is Monika Bickert, and I am the Vice President of Global
Policy Management and Counterterrorism at Facebook. In that role, I
lead our efforts related to Product Policy and Counterterrorism. Prior
to assuming my current role, I served as lead security counsel for
Facebook, working on issues ranging from children's safety to
interactions with law enforcement. And before that, I was a criminal
prosecutor with the Department of Justice for 11 years in Chicago and
Washington, DC, where I prosecuted Federal crimes including public
corruption and gang violence.
On behalf of everyone at Facebook, I would like to express our
sympathy and solidarity with the victims, families, communities, and
everyone else affected by the recent terrible attacks across the
country. In the face of such heinous acts, we remain committed to
cooperating with law enforcement and standing with our community
against hate and violence. We are thankful to be able to provide a way
for those affected by the horrific recent attacks to communicate with
loved ones, to organize events for people to gather and grieve, and to
raise money to help support these communities as they begin to heal.
Facebook's mission is to give people the power to build community
and bring the world closer together. We are proud that more than two
billion people around the world come to Facebook every month to connect
and share with one another. But people need to feel safe in order to
build this community. That is why Facebook prohibits harmful conduct on
its platform, including hate speech and inciting violence. Our goal is
to ensure that Facebook is a place where both expression and personal
safety are protected and respected.
We are not aware of any connection between our platform and the
recent attacks, but we recognize that we all have a role to play in
keeping our communities safe. At Facebook, we have strong policies and
invest significant resources to protect our users on and offline.
II. Facebook's Policies Against Hate and Violence
Facebook is committed to protecting our community by removing any
content from our services that encourages real-world harm. Because
harmful content can take many forms, we have several policies in place
to address these issues, all of which are published in our Community
Standards, which define the content that is and is not allowed on our
platform.
When we find content that violates our standards, we remove it. We
invest in technology, processes, and people to help us identify
violations and act quickly to mitigate any impact. There is always room
for improvement, but we remove millions of pieces of content every
year, much of it before any user reports it. We outline below several
of the important steps that we take to prevent violence and keep our
users safe.
Prohibition Against Violence and Incitement: We care deeply about
our users and we want them to be safe. Therefore, it is critical to our
mission to help prevent potential offline harm that may be related to
content on Facebook. We remove content, disable accounts, and work with
law enforcement when we believe there is a risk of physical harm or
direct threats to public safety.
Prohibition of Dangerous Individuals and Organizations: In an
effort to prevent and disrupt real-world harm, we do not allow any
individuals or organizations that proclaim a violent mission, advocate
violence, or are engaged in violence to have a presence on Facebook for
any purpose, even if it appears benign. This includes organizations or
individuals involved in the following:
Terrorist activity, both domestic and international;
Organized hate, including white supremacy and white
nationalism;
Human trafficking; and
Organized violence or criminal activity.
We do not allow propaganda or symbols that represent any of these
organizations or individuals to be shared on our platform unless they
are being used to condemn or inform--for example, by media
organizations. We do not allow content that praises any of these
organizations or individuals or any acts committed by them. And we do
not allow coordination of support for any of these organizations or
individuals or any acts committed by them.
No Promoting or Publicizing Crime: We prohibit people from
promoting or publicizing violent crime, theft, and/or fraud because we
do not want to condone this activity and because there is a risk of
copycat behavior. We also do not allow people to depict criminal
activity or admit to crimes they or their associates have committed.
Policies Against Coordinating Harm: In an effort to prevent and
disrupt real-world harm, we prohibit people from facilitating or
coordinating future activity, criminal or otherwise, that is intended
or likely to cause harm to people, businesses, or animals. People can
draw attention to harmful activity that they may witness or experience
as long as they do not advocate for or coordinate harm.
Combatting Suicide and Self-Injury: We also use and continue to
develop tools and resources to proactively identify and help people who
may be at risk of suicide or self-injury. We leverage pattern
recognition technology to detect posts or live videos where someone
might be expressing an intent to harm themselves. We also use
artificial intelligence (AI) to prioritize the order in which our team
reviews reported content relating to suicide or self-injury. This
ensures we can get the right resources to people in distress and, where
appropriate, we can more quickly alert first responders. And we remove
content that encourages suicide or self-injury, including certain
graphic imagery and real-time depictions that experts tell us might
lead others to engage in similar behavior. We also work with
organizations around the world to provide assistance and resources to
people in distress.
Cooperation with Law Enforcement: Law enforcement plays a critical
role in keeping people safe, and we have a long history of working
successfully with law enforcement to address a wide variety of threats.
As a former Federal prosecutor, I know that this cooperation is vital.
When we do receive reports or otherwise find content that violates our
policies, we remove it. And we proactively reach out to law enforcement
if we see a credible threat of imminent harm.
III. Facebook's Efforts to Combat Violence and Hate
Our efforts to combat violent and hateful content are focused in
three areas: developing new technical capabilities for our products,
investing in people, and building partnerships.
Product Enhancements: Facebook has invested significantly in
technology to help meet the challenge of proactively identifying
violent content, including through the use of AI and other automation.
These technologies have become increasingly central to keeping hateful
and violent content off of Facebook.
We use a wide range of technical tools to identify violent and
hateful content. This includes hashes--or digital fingerprints--that
allow us to find secondary versions of known bad content; text parsing;
digital ``fan-outs'' to identify profiles, groups, and pages related to
those we have identified as problematic; and more holistic machine
learning that can assess all aspects of a post and score whether it is
likely to violate our Community Standards.
We also know that bad actors adapt as technology evolves, and that
is why we constantly update our technical solutions to deal with more
types of content in more languages, and to react to the new ways our
adversaries try to exploit our products. For example, in response to
the tragic events in Christchurch, we made changes to Facebook Live to
restrict users if they have violated certain rules--including our
Dangerous Organizations and Individuals policy. We now apply a ``one-
strike'' policy to Live: anyone who violates our most serious policies
will be restricted from using Live for set periods of time--for
example, 30 days--starting on their first offense. We have also updated
our proactive detection systems and reduced the average time it takes
for our AI to find a violation on Facebook Live to 12 seconds--a 90
percent reduction in our average detection time from a few months ago.
Being able to detect violations sooner means that in emergencies where
every minute counts, we can assist faster.
Investments in People: We know that we cannot rely on AI alone to
identify potentially violent content. Context often matters. To
understand more nuanced cases, we need human expertise.
One of our greatest human resources is our community of users. Our
users help us by reporting accounts or content that may violate our
policies--including the small fraction that may be related to acts of
violence. To review those reports, and to prioritize the safety of our
users and our platform more generally, we have more than 30,000 people
working on safety and security across the company and around the world.
That is three times as many people as we had dedicated to such efforts
in 2017. Our safety and security professionals review reported content
in more than 50 languages, 24 hours a day.
We also have a team of more than 350 people at Facebook whose
primary job is dealing with terrorists and other Dangerous Individuals
and Organizations. This team includes language and cultural
specialists, former law enforcement and intelligence professionals, and
academics that have studied these groups and individuals for years.
Many of them came to Facebook specifically because they are committed
to the mission of keeping people safe.
This team was previously focused on counterterrorism, and we used
our most sophisticated tools to predominantly combat ISIS, al-Qaeda,
and their affiliates, which were recognized then as posing the greatest
threats to our global community. Now, they lead our efforts against all
people and organizations that proclaim or are engaged in violence. We
are taking the initial progress we made in combatting content
affiliated with ISIS, al-Qaeda, and their affiliates, and we are
further building out techniques to identify and combat the full breadth
of violence and extremism covered under our Dangerous Organizations
policy.
Partnerships: We are proud of the work we have done to make
Facebook a hostile place for those committed to acts of violence. We
understand, however, that simply working to keep violence off Facebook
is not an adequate solution to the problem of online extremism and
violence, particularly because bad actors can leverage a variety of
platforms. We believe our partnerships with other companies, civil
society, researchers, and governments are crucial to combatting this
threat. For example, our P2P Global Digital Challenge, which engages
university students around the world in competitions to create social
media campaigns and offline strategies to challenge hateful and
extremist narratives, has launched over 600 counterspeech campaigns
from students in 75 countries, engaged over 6,500 students, and reached
over 200 million people. We're also partnering with Life After Hate, an
organization founded by former violent extremists, to connect people
who search for terms associated with white supremacy to resources
focused on helping people leave behind hate groups.
Our work to combat violence is never complete. Individuals and
organizations intent on violent acts come in many ideological stripes--
and the most dangerous among them are deeply resilient. We know that
bad actors will continue to attempt to skirt our detection with more
sophisticated efforts, and we are dedicated to continuing to advance
our work and share our progress.
IV. Conclusion
Facebook is committed to helping people build a vibrant community
that encourages and fosters free expression. At the same time, we want
to do what we can to protect our users from real-world harm and stop
terrorists, extremists, hate groups, and any others from using our
platform to promote or engage in violence. We recognize that there is
always more work to do in combatting the abuse of our site by bad
actors, but we are proud of the progress we have made over the last few
years. We know that people have questions about what we are doing to
continue that progress, and we look forward to working with this
Committee, regulators, and others in the tech industry and civil
society to continue working on these issues. I appreciate the
opportunity to be here today, and I look forward to your questions.
The Chairman. Thank you very much. Mr. Pickles.
STATEMENT OF NICK PICKLES, DIRECTOR, PUBLIC POLICY STRATEGY,
TWITTER, INC.
Mr. Pickles. Chairman Wicker, Ranking Member Cantwell,
members of the Committee, thank you for the opportunity to
appear today to discuss these important issues. Twitter has
publicly committed to improving the collective health,
openness, and civility of public conversation on our platform.
Our policies are designed to keep people safe on Twitter and
they continuously evolve to reflect the realities of the world
we operate in. We are working faster, we are investing to
remove content that distracts from healthy conversation before
it is reported, including terrorists and violent extremist
content.
Tackling terrorism, violent extremism, and preventing
violent attacks requires a whole of society response including
from social media companies. Let me be clear, Twitter is
incentivized to keep terrorists and violent content off our
service both from a business standpoint and then the current
legal frameworks. Such content does not serve our business
interests, it breaks our rules, but is fundamentally contrary
to our values. Communities in America and around the world have
been impacted by instance of mass violence, terrorism, and
violent extremism with tragic frequency in recent years. These
events demand a robust public policy response from every
quarter.
We acknowledge that technology companies have a role to
play. However, it is important to recognize content removal
alone cannot solve these issues. I would like to outline four
of Twitter's key policies in this area. Firstly, Twitter takes
a zero tolerance approach to terrorists content on our service.
Individuals may not promote terrorism, engage in terrorist
recruitment, or terrorist acts. Since 2015, we have suspended
more than 1.5 million accounts for violations of our rules
related to terrorism and continue to see more than 90 percent
of these accounts suspended through our own proactive measures.
In the majority of cases, we take action at the account
creation stage before account has even tweeted, and the
remaining 10 percent is identified through a combination of
user reports and partnerships. Second, we prohibit the use of
Twitter by violent extremist groups. These are defined in our
rules as groups that whether by statements on or off the
platform use or promote violence against civilians to further
their cause whatever their ideology. Since the introduction of
this policy in 2017, we have taken action on more than 186
groups globally and suspended more than 2,000 unique accounts.
Third, Twitter does not allow hateful conduct on our service.
An individual on Twitter is not permitted to threaten or
promote violence or directly attack people based on their
protected characteristics. Where any of these rules are broken,
we will take action to remove the content and will permanently
remove those promoting terrorism or violent extremism from
Twitter. Fortunately, our rules prohibit the selling, buying,
or facilitating transactions in weapons, including firearms,
ammunition, and explosives, and instructions on making weapons.
So are explosive devices or 3D printed weapons.
We will take appropriate action on any account found to be
engaged in this activity, including a permanent suspension
where appropriate. Additionally, we prohibit the promotion of
weapons and weapon accessories globally through our paid
advertising policies. Collaboration with our industry peers and
civil society is critically important to addressing the common
threats from terrorism globally.
In June 2017, we launched the Global Internet Forum to
Counter Terrorism, GCT, a partnership with Twitter, YouTube,
Facebook, and Microsoft. This facilitates, among other things,
information-sharing, technical cooperation, research
collaboration, including with academic institutions. Twitter
and technology companies have a role to play in addressing mass
violence, ensuring our platforms cannot be exploited by those
promoting violence. This cannot be the only public policy
response and removing content alone will not stop those who are
determined to cause harm.
Quite often, when we remove content from our platforms, it
moves those views, these ideologies into the darker corners of
the Internet where they cannot be challenged and held to
account. As our pair companies are improving their efforts,
this content continues to migrate to less governed platforms
and services.
We are committed to learning and improving, but every part
of the online ecosystem has a part to play. Addressing mass
violence requires a whole of society response. We welcome the
opportunity to continue to work with industry peers, Government
institutions, legislators, law enforcement, academics, and
civil society to find the right solutions. Thank you for your
time today.
[The prepared statement of Mr. Pickles follows:]
Prepared Statment of Nick Pickles, Director, Public Policy Strategy,
Twitter, Inc.
Chairman Wicker, Ranking Member Cantwell, and Members of the
Committee:
At Twitter, our mission is to serve the public conversation.
Twitter is a place where people from around the world come together in
an open and free exchange of ideas. We have made the health of our
service the top priority. Conversely, abuse, malicious automation,
hateful conduct, violent extremist and terrorist content terrorism, and
manipulation will detract from the health of our platform.
Tackling terrorism, violent extremism, and preventing violent
attacks require a whole of society response, including from social
media. It has long been a priority of Twitter to remove this content
from the service. Let me be clear: Twitter has no incentive to keep
terrorist and violent extremist content available on our platform. Such
content does not serve our business interests, breaks our rules, and is
fundamentally contrary to our values.
Communities in America and around the world have been impacted by
incidents of mass violence, terrorism, and violent extremism with
tragic frequency in recent years. These events demand a robust public
policy response from every quarter. We acknowledge that the technology
companies play a critical role, however, it is important to recognize
content removal online cannot alone solve these issues.
We welcome the opportunity to continue to work with you on the
Committee, our industry peers, government, academics, and civil society
to find the right solutions. Partnership is essential.
My statement today will provide information and deeper context on:
(I) Twitter's work to protect the health of the public conversation,
including combating terrorism, violent extremist groups, and hateful
conduct; (II) our policies relating to weapons and weapon accessories;
and (III) our partnerships and societal engagement.
I. TWITTER'S POLICIES ON TERRORIST CONTENT, VIOLENT
EXTREMIST GROUPS, AND HATEFUL CONDUCT
All individuals accessing or using Twitter's services must adhere
to the policies set forth in the Twitter Rules. Accounts under
investigation or that have been detected as sharing content in
violation with the Twitter Rules may be required to remove content, or
in serious cases, will see their account permanently suspended. Our
policies and enforcement options evolve continuously to address
emerging behaviors online.
A. Policy on Terrorism
Individuals on Twitter are prohibited from making specific threats
of violence or wish for the serious physical harm, death, or disease of
an individual or group of people. This includes, but is not limited to,
threatening or promoting terrorism.
We suspended more than 1.5 million accounts for violations related
to the promotion of terrorism between August 1, 2015, and December 31,
2018. In 2018, a total of 371,669 accounts were suspended for
violations related to promotion of terrorism. More than 90 percent of
these accounts are suspended through our proactive measures.
We have a zero-tolerance policy and take swift action on ban
evaders and other forms of behavior used by terrorist entities and
their affiliates. In the majority of cases, we take action at the
account creation stage--before the account even Tweets.
Government and law enforcement reports constituted less than 0.1
percent of all suspensions in the last reporting period. Continuing the
trend we have seen for some time, the number of reports we received
from governments of terrorist content from the second half of last year
decreased by 77 percent compared to the previous reporting period
covering January through June 2018.
We are reassured by the progress we have made, including
recognition by independent experts. For example, Dublin City University
Professor Maura Conway found in a detailed study that ``ISIS's
previously strong and vibrant Twitter community is now . . . virtually
non-existent.''
B. Policy on Violent Extremist Groups
In December 2017, we broadened our rules to encompass accounts
affiliated with violent extremist groups. Our prohibition on the use of
Twitter's services by violent extremist groups--i.e., identified groups
subscribing to the use of violence as a means to advance their cause--
applies irrespective of the cause of the group.
Our policy states:
Violent extremist groups are those that meet all of the below
criteria:
identify through their stated purpose, publications, or
actions as an extremist group;
have engaged in, or currently engage in, violence and/or the
promotion of violence as a means to further their cause; and
target civilians in their acts and/or promotion of violence.
An individual on Twitter may not affiliate with such an
organization--whether by their own statements or activity both on and
off the service--and we will permanently suspend those who do so.
We know that the challenges we face are not static, nor are bad
actors homogenous from one country to the next in how they behave. Our
approach combines flexibility with a clear, consistent policy
philosophy, enabling us to move quickly while establishing clear norms
of unacceptable behavior.
Since the introduction of our policy on violent extremist groups,
we have taken action on 186 groups under this policy and permanently
suspended 2,217 unique accounts. Ninety-three of these groups advocate
violence against civilians alongside some form of extremist white
supremacist ideology.
C. Policy on Hateful Conduct
People on Twitter are not permitted to promote violence against or
directly attack or threaten other people on the basis of race,
ethnicity, national origin, sexual orientation, gender, gender
identity, religious affiliation, age, disability, or serious disease.
We also do not allow accounts whose primary purpose is inciting harm
toward others on the basis of these categories.
We do not allow individuals to use hateful images or symbols in
their profile image or profile header. Individuals on the platform are
not allowed to use the username, display name, or profile bio to engage
in abusive behavior, such as targeted harassment or expressing hate
toward a person, group, or protected category.
Under this policy, we take action against behavior that targets
individuals or an entire protected category with hateful conduct.
Targeting can happen in a number of ways, for example, mentions,
including a photo of an individual, or referring to someone by their
full name.
When determining the penalty for violating this policy, we consider
a number of factors including, but not limited to the severity of the
violation and an individual's previous record of rule violations. For
example, we may ask someone to remove the violating content and serve a
period of time in read-only mode before they can Tweet again.
Subsequent violations will lead to longer read-only periods and may
eventually result in permanent account suspension. If an account is
engaging primarily in abusive behavior, or is deemed to have shared a
violent threat, we will permanently suspend the account upon initial
review.
D. Investing in Tech: Behavior vs. Content
Twitter's philosophy is to take a behavior-led approach, utilizing
a combination of machine learning and human review to prioritize
reports and improve the health of the public conversation. That is to
say, we increasingly look at how accounts behave before we look at the
content they are posting. This is how we seek to scale our efforts
globally and leverage technology even where the language used is highly
context specific. Twitter employs extensive content detection
technology to identify potentially abusive content on the service,
along with allowing users to report content to us either as an
individual or a bystander.
For abuse, this strategy has allowed us to take three times the
amount of enforcement of action on abuse within 24 hours than this time
last year. We now proactively surface over 50 percent of abusive
content we remove using our technology compared to 20 percent a year
ago. This reduces the burden on individuals to report content to us.
Since we started using machine learning three years ago to reduce the
visibility on abusive content:
80 percent of all replies that are removed were already less
visible;
Abuse reports have been reduced by 7.6 percent;
The most visible replies receive 45 percent less abuse
reports;
100,000 accounts were suspended for creating new accounts
after a suspension during January through March 2019 -- a 45
percent increase from the same time last year;
60 percent faster response to appeals requests with our new
in-app appeal process;
3 times more abusive accounts suspended within 24 hours
after a report compared to the same time last year; and
2.5 times more private information removed with a new,
easier reporting process.
II. TWITTER POLICIES REGARDING WEAPONS AND WEAPON
ACCESSORIES
Although Twitter's service does not have an e-commerce function,
our Rules prohibit the selling, buying, or facilitating transactions in
weapons, including firearms, ammunition, and explosives, and
instructions on making weapons, such as bombs or 3D printed weapons. We
will take appropriate action on any account found to be engaged in this
activity, including permanent suspension of accounts where appropriate.
As stated publicly in our advertising policies, Twitter does not
allow the use of our promoted products for the purpose of promoting
weapons and weapon accessories globally. We explicitly ban advertising
of guns, including airsoft guns, air guns, blow guns, paintball guns,
antique guns, replica guns, and imitation guns. Twitter also prohibits
the use of our promoted products for gun parts and accessories,
including gun mounts, grips, magazines, and ammunition. We also do not
allow the advertising of the rental of guns (other than from shooting
ranges), stun guns, taser guns, mace, pepper spray or other similar
self defense weapons. Additionally, we do not permit the advertising of
a variety of weapons including swords, machetes, and other edged/bladed
weapons; explosives, bombs and bomb making supplies and/or equipment;
fireworks, flamethrowers and other pyrotechnic devices; and knives,
including butterfly knives, fighting knives, switchblades, disguised
knives, and throwing stars.
We do allow advertising related to the discussion of public policy
issues pertaining to firearms. Twitter requires extensive information
disclosures of any account involved in political issue advertising and
provides specific information to the public via our Ads Transparency
Center. Such advertisements are distinctly labeled as political issue
promoted tweets. Organizations on both sides of the debate have
utilized Twitter's promoted products and continue to do so, within the
boundaries of our advertising policies.
III. PARTNERSHIPS AND SOCIETAL ENGAGEMENT
We work closely with the Federal Bureau of Investigation, along
with law enforcement and numerous public safety authorities around the
world. As our partnerships deepen, we are able to better respond to the
changing threats we all face, sharing valuable information and promptly
responding to valid legal requests for information.
A. Cooperation with Law Enforcement
We have well-established relationships with law enforcement
agencies, and we look forward to continued cooperation with them on
these issues, as often they have access to information critical to our
joint efforts to stop bad faith actors. The threat we face requires
extensive partnership and collaboration with our government partners
and industry peers. We each possess information the other does not
have, and our combined information is more powerful in combating these
threats together. We have continuous coverage to address reports from
law enforcement around the world and have a portal to swiftly handle
law enforcement requests rendered by appropriate legal process.
Twitter informs individuals using the platform that we may
preserve, use, or disclose an individual's personal data if we believe
that it is reasonably necessary to comply with a law, regulation, legal
process, or governmental request; to protect the safety of any person;
to protect the safety or integrity of our platform, including to help
prevent spam, abuse, or malicious actors on our services, or to explain
why we have removed content or accounts from our services; to address
fraud, security, or technical issues; or to protect our rights or
property or the rights or property of those who use our services.
Twitter retains different types of information for different time
periods, and in accordance with our Terms of Service and Privacy
Policy. Given Twitter's real-time nature, some information (e.g.,
Internet Protocol logs) may only be stored for a very brief period of
time.
Some information we store is automatically collected, while other
information is provided at the user's discretion. Though we do store
this information, we cannot guarantee its accuracy. For example, the
user may have created a fake or anonymous profile. Twitter doesn't
require real name use, e-mail verification, or identity authentication.
Once an account has been deactivated, there is a very brief period
in which we may be able to access account information, including
Tweets. Content removed by account holders (e.g., Tweets) is generally
not available.
Twitter accepts requests from law enforcement to preserve records,
which constitute potentially relevant evidence in legal proceedings. We
will preserve, but not disclose, a temporary snapshot of the relevant
account records for 90 days pending service of valid legal process.
Twitter may honor requests for extensions of preservation requests,
but encourage law enforcement agencies to seek records through the
appropriate channels in a timely manner, as we cannot always guarantee
that requested information will be available.
Our biannual Twitter Transparency Report highlights trends in
enforcement of our Rules, legal requests, intellectual property-related
requests, and e-mail privacy best practices. The report also provides
insight into whether or not we take action on these requests. The
Transparency Report includes information requests from governments
worldwide and non-government legal requests we have received for
account information. In 2018, we received 4,323 requests from United
States authorities, relating to 13,086 accounts,
B. Industry Collaboration
Collaboration with our industry peers and civil society is also
critically important to addressing common threats from terrorism
globally. In June 2017, we launched the Global Internet Forum to
Counter Terrorism (the ``GIFCT''), a partnership among Twitter,
YouTube, Facebook, and Microsoft.
The GIFCT facilitates, among other things, information sharing;
technical cooperation; and research collaboration, including with
academic institutions. In September 2017, the members of the GIFCT
announced a significant financial commitment to support research on
terrorist abuse of the Internet and how governments, tech companies,
and civil society can respond effectively. Our goal is to establish a
network of experts that can develop platform-agnostic research
questions and analysis that consider a range of geopolitical contexts.
Technological collaboration is a key part of GIFCT's work. In the
first two years of GIFCT, two projects have provided technical
resources to support the work of members and smaller companies to
remove terrorist content.
First, the shared industry database of ``hashes''--unique digital
``fingerprints''--for violent terrorist propaganda now spans more than
100,000 hashes. The database allows a company that discovers terrorist
content on one of its sites to create a digital fingerprint and share
it with the other companies in the forum, who can then use those hashes
to identify such content on their services or platforms, review against
their respective policies and individual rules, and remove matching
content as appropriate or block extremist content before it is posted.
Second, a year ago, Twitter began working with a small group of
companies to test a new collaborative system. Because Twitter does not
allow files other than photos or short videos to be uploaded, one of
the behaviors we saw from those seeking to promote terrorism was to
post links to other services where people could access files, longer
videos, PDFs, and other materials. Our pilot system allows us to alert
other companies when we removed an account or Tweet that linked to
material that promoted terrorism hosted on their service. This
information sharing ensures the hosting companies can monitor and track
similar behavior, taking enforcement action pursuant with their
individual policies. This is not a high-tech approach, but it is simple
and effective, recognizing the resource constraints of smaller
companies.
Based on positive feedback, the partnership has now expanded to 12
companies and we have shared more than 14,000 unique URLs with these
services. Every time a piece of content is removed at source, it means
any link to that source--wherever it is posted--will no longer be
operational.
We are eager to partner with additional companies to expand this
project, and we look forward to building on our existing partnerships
in the future.
Finally, GIFCT has established a real-time crisis response process
that allows us to respond to a violent act quickly to ensure that we
share valuable information to limit the spread of terrorist and violent
extremist content.
C. The Christchurch Call to Action
In the months since a terrorist attack in Christchurch, New
Zealand, New Zealand Prime Minister Jacinda Ardern has led the
international policy debate, and that work has culminated in the
Christchurch Call. Twitter's Chief Executive Officer Jack Dorsey
attended the launch of the Christchurch Call in Paris, meeting with the
Prime Minister to express our support and partnership with the New
Zealand Government.
Because terrorism cannot be solved by the tech industry alone, the
Christchurch Call is a landmark moment and an opportunity to convene
governments, industry, and civil society to unite behind our mutual
commitment to a safe, secure open, global Internet. It is also a moment
to recognize that however or wherever evil manifests itself, it affects
us all.
In fulfilling our commitments in the Call, we will take a wide
range of actions. We continue to invest in technology to prioritize
signals, including user reports, to ensure we can respond as quickly as
possible to a potential incident, building on the work we have done to
harness proprietary technology to detect and disrupt bad actors
proactively.
As part of our commitment to educate users about our rules and to
further prohibit the promotion of terrorism or violent extremist
groups, we have updated our rules and associated materials to be
clearer on where these policies apply. This is accompanied by further
data being provided in our transparency report, allowing public
consideration of the actions we are taking under our rules, as well as
how much content is detected by our proactive efforts.
Twitter will take concrete steps to reduce the risk of
livestreaming being abused by terrorists, while recognizing that during
a crisis these tools are also used by news organizations, citizens and
governments. We are investing in technology and tools to ensure we can
act even faster to remove video content and stop it spreading.
Finally we are committed to continuing our partnership with
industry peers, expanding on our URL sharing efforts along with wider
mentoring efforts, strengthening our new crisis protocol arrangements,
and supporting the expansion of GIFCT membership.
D. Partnerships with Civil Society
In tandem with removing content, our wider efforts on countering
violent extremism going back to 2015 have focused on bolstering the
voices of non-governmental organizations and credible outside groups.
These organizations and groups can use our uniquely open service to
spread positive and affirmative campaigns that seek to offer an
alternative to narratives of hate. Ideologies can only be successfully
countered by those who have the credibility to take on the core
messages being propagated, and if these core messages go unchallenged
the removal of content will always be an incomplete response. These
groups do critical work and policy makers should continue to find ways
to broaden support for these efforts.
We have partnered with organizations delivering counter and
alternative narrative initiatives across the globe and we encourage the
Committee to consider the role of government in supporting the work of
credible messengers in this space at home and abroad. Twitter has also
delivered capacity building workshops to a range of organizations who
seek to provide positive, alternative messages and work with
communities and individuals at risk.
E. A Whole of Society Response
The challenges we face as a society are complex, varied, and
constantly evolving. These challenges are reflected and often magnified
by technology. The push and pull factors influencing individuals vary
widely, there is no common catalyst to action and there is no one
solution to prevent an individual turning to violence. This is a long-
term problem requiring a long-term response, not just the removal of
content.
We are committed to playing our part. We will continue to seek to
proactively remove terrorist and violent extremist content, work with
industry peers to respond quickly in a crisis and to support smaller
companies in tackling these challenges.
While we strictly enforce our policies, removing all discussion of
particular viewpoints, no matter how uncomfortable society may find
them, does not eliminate the ideology underpinning them. There is a
risk such an approach moves these views into darker corners of the
Internet where they cannot be challenged and held to account. As our
peer companies improve in their efforts, this content continues to
migrate to less-governed platforms and services often not at the
forefront of public discussions. We are committed to learning and
improving, but every part of the online ecosystem has a part to play.
Furthermore, not every issue will be one where the underlying
factors can be addressed by public policy interventions led by
technology companies.
* * *
We stand ready to assist the Committee in its important work
regarding the issue of the tools that Internet companies can employ to
stop the spread of mass violence on our services.
The Chairman. Thank you very much. Mr. Slater.
STATEMENT OF DEREK SLATER, GLOBAL DIRECTOR, INFORMATION POLICY,
GOOGLE LLC
Mr. Slater. Chairman Wicker, Ranking Member Cantwell,
distinguished members of the Committee, thank you for the
opportunity to appear before you today. My name is Derek
Slater. I am the Global Director of Information Policy at
Google. In that capacity, I lead a team that advises the
company on public policy frameworks for dealing with online
content, including hate speech, extremism, and terrorism.
Before I begin, I would like to take a moment on behalf of
everyone at Google to express our horror in learning of the
tragic attacks in Texas, Ohio and elsewhere, and share our
sincere condolences to the affected families, friends, and
communities.
All Google services were not involved in these recent
incidents. We have engaged with the White House, Congress, and
governments around the globe on steps we are taking to ensure
that our platforms are not used to support hate speech or
incite violence. In my testimony today, I will focus on three
key areas where we are making progress to help protect people.
First, how we work with governments and law enforcement,
second, how our efforts to prohibit the promotion of products
that causes damage, harm, or injury, and third, the enforcement
of our policies around terrorism and hate speech.
First, Google engages in ongoing dialogue with law
enforcement agencies to understand the threat landscape and
respond to threats that affect the safety of our users and the
broader public. For example, when we have a good faith belief
that there is a threat to life or serious bodily harm made on
our platform in the United States, the Google cybercrime
investigation group will report it to the Northern California
Regional Intelligence Center. In turn, that Intelligence Center
quickly gets the report into the hands of officers to respond.
The cybercrime investigation group is on call 24/7 to make
these reports. We are also deeply committed to working with
Government, the tech industry, and experts from civil society
and academia. Since 2017, we have done this in particular
through the Global Internet Forum to Counter Terrorism of which
YouTube is a founded company, and Google was its first chair.
Recently, GIFCT introduced joint content incident protocols for
responding to emerging or active events. The GIFCT also
released its first-ever Transparency Report and a new counter
speech campaign toolkit.
Second, we take the threat posed by gun violence in the
United States very seriously and our advertising policies have
long prohibited the promotion of weapons, ammunition, and
similar products that cause damage, harm, or injury. Similarly,
we also prohibit the promotion of instructions for making guns,
explosives, or other harmful products, and we employ a number
of proactive and reactive measures to ensure that our policies
are appropriately enforced. We know that we must be vigilant on
these issues and are constantly improving our enforcement
procedures, including implementing enhancements to our
automated systems and updating our incident management and
manual review procedures.
Third, on YouTube, we have rigorous policies and programs
that defend against the use of our platform to spread hate or
incite violence. Over the past two years, we have invested
heavily in machines and people to quickly identify and remove
content that violates our policies. This includes machine
learning technology to effectively enforce our policies at
scale, hiring over 10,000 people across Google tasked with
detecting or viewing and removing content.
An intel desk of experts that proactively looks for new
trends and improves escalation pathway for expert NGOs and
governments to notify us about content in bulk through our
trusted flagger program, and finally going beyond removals by
actively creating programs to promote beneficial counter
speech, such as the creators for change program and alphabets
jigsaw groups use for a redirect method. This broad, cross
sectional work has led to tangible results. Over 87 percent of
the 9 million videos we removed in the second quarter of 2019
were first flagged by our automated systems.
More than 80 percent of those auto flagged videos were
removed before they received a single view, and overall, videos
that violate our policies generate a fraction of a percent of
the views on YouTube. Our efforts do not end there. As we are
constantly evolving to new challenges and looking for ways to
improve our policies. For example, YouTube recently further
updated its hate speech policy. The updated policy specifically
prohibits videos alleging that a group is superior in order to
justify discrimination, segregation, or exclusion based on
qualities like age, gender, race, caste, religion, sexual
orientation, or veteran status.
It can take months for us to ramp up enforcement of our new
policies. We have already seen five times spike and removals
and channel terminations on hate speech. In conclusion, we take
the safety of our users very seriously and value our close and
collaborative relationships with law enforcement and government
agencies.
We understand these are difficult issues of great interest
to Congress and want to be responsible actors who are part of
the solution. As these issues evolve, Google will continue to
invest in the people and technology to meet the challenge. We
look forward to continued collaboration with the Committee as
it examines these issues. Thank you for your time, and I look
forward to taking your questions.
[The prepared statement of Mr. Slater follows:]
Prepared Statement of Derek Slater, Director, Information Policy,
Google LLC
Chairman Wicker, Ranking Member Cantwell, and distinguished members
of the Committee: Thank you for the opportunity to appear before you
today. I appreciate Congress' work in looking closely at how to prevent
tragic episodes of mass violence.
My name is Derek Slater, and I am the Global Director of
Information Policy at Google. In that capacity I lead a team that
advises the company on public policy frameworks for dealing with online
content--including hate speech, extremism, and terrorism. Prior to my
role at Google, I worked on Internet policy at the Electronic Frontier
Foundation and at the Berkman Center for Internet and Society.
Before I begin, I would like to take a moment on behalf of everyone
at Google to express our horror in learning of the tragic attacks in
Texas and Ohio and to share our sincere condolences to the affected
families, friends, and communities. While Google services were not
involved in these recent incidents, we have engaged with the White
House, Congress, and governments around the globe on steps we are
taking to ensure that our platforms are not used to support hate speech
or incite violence.
We believe the free flow of information and ideas has important
social, cultural and economic benefits, though society has always
recognized that free speech must be subject to reasonable limits. This
is true both online and off, and it is why, in addition to respecting
the law, we have additional policies, procedures, and community
guidelines that govern what activity is permissible on our platforms.
In my testimony today, I will focus on three key areas where we are
making progress to help protect people: (i) how we work with
governments and law enforcement; (ii) our efforts to prohibit the
promotion of products that cause damage, harm, or injury; and (iii) the
enforcement of our policies around terrorism and hate speech.
Working with Government and Law Enforcement
Google appreciates that law enforcement agencies face significant
challenges in protecting the public against crime and terrorism. Google
engages in ongoing dialogue with law enforcement agencies to understand
the threat landscape and respond to threats that affect the safety of
our users and the broader public. When we become aware of statements on
our platform that constitute a threat to life or that reflect that
someone's life may be in danger, we report this activity to law
enforcement agencies.
For example, when we have a good faith belief that there is a
threat to life or serious bodily harm made on our platform in the
United States, the Google CyberCrime Investigation Group (CCIG) will
report it to the Northern California Regional Intelligence Center
(NCRIC). In turn, NCRIC quickly gets the report into the hands of
officers to respond. CCIG is on call 24/7 to make these reports.
Under U.S. law, the Stored Communications Act allows Google and
other service providers to voluntarily disclose user data to
governmental entities in emergency circumstances where the provider has
a good faith belief that disclosing the information will prevent loss
of life or serious physical injury to a person. Our team is staffed on
a 24/7/365 basis to respond to these emergency disclosure requests
(EDRs). We have seen significant growth in the volume of EDRs that we
receive from U.S. governmental entities, as illustrated in our
transparency report covering government requests for user data. In
fact, the number of EDRs submitted from agencies in the U.S. almost
doubled from 2017 to 2018. We have grown our teams to accommodate this
growing volume and to ensure we can quickly respond to emergency
situations that implicate public safety.
We are also deeply committed to working with government, the tech
industry, and experts from civil society and academia to protect our
services from being exploited by bad actors. The recent tragic events
in Christchurch presented unique challenges, and we had to take
unprecedented steps to address the sheer volume of new videos related
to the events. In the months since, Google and YouTube signed the
Christchurch Call to Action, a series of commitments to quickly and
responsibly address terrorist content online. This is an extension of
our ongoing commitment to working with our colleagues in the industry
to address the challenges of terrorism online. Since 2017, we've done
this through the Global Internet Forum to Counter Terrorism (GIFCT), of
which Google is a founding company and was its first chair. Recently,
GIFCT introduced joint content incident protocols for responding to
emerging or active events. The GIFCT also released its first-ever
Transparency Report and a new counterspeech campaign toolkit that will
help activists and civil society organizations challenge the voices of
extremism online.
Prohibiting the Promotion of Products That May Cause Damage, Harm, or
Injury
We take the threat posed by gun violence in the United States very
seriously and our advertising policies have long prohibited the
promotion of weapons, ammunition, explosive materials, fireworks, and
similar products that cause damage, harm, or injury. Similarly, we also
prohibit the promotion of instructions for making guns, explosives, or
other harmful products.
On platforms like Google Ads and Google Shopping Ads, we employ a
number of proactive and reactive measures to ensure that our policies
are appropriately enforced. For example, we run automated and manual
checks to detect content that violates our policies. If an advertiser
or merchant violates our policies, we will take appropriate action up
to and including suspension of their account. Users can also provide
direct feedback on ads that potentially violate Google policies via an
external form using the `Report a violation' link or via the feedback
link on Google.com and other Google properties to report any products
that may violate our policies. This feedback is reviewed by our teams
and appropriate action is taken.
We know that we must be vigilant on these issues and are constantly
improving our enforcement procedures, including implementing
enhancements to our automated systems and updating our incident
management and manual review procedures.
Policies and Enforcement on YouTube for Terrorism and Hate Speech
We have robust policies and programs to defend our platforms to
spread hate or incite violence. This includes prohibitions on:
terrorist recruitment, violent extremism, incitement to violence,
glorification of violence, and instructional videos related to acts of
violence. We apply these policies to violent extremism of all kinds,
whether inciting violence on the basis of race or religion or as part
of an organized terrorist group.
In order to improve the effectiveness of our policy enforcement, we
have invested heavily in both technology and people to quickly identify
and remove content that violates our policies against incitement to
violence and hate speech:
(1) YouTube's enforcement system starts from the point at which a
user uploads a video. If our technology detects that the video
is similar to videos that we know already violate our policies,
it is sent for humans to review. If they determine that it
violates our policies, they remove it and the system makes a
``digital fingerprint'' or hash of the video so it can't be
uploaded again.
(2) Machine learning technology also helps us more effectively
identify this content and enforce our policies at scale.
However, because hate and violent extremism content is
constantly evolving and can sometimes be context-dependent, we
also rely on experts to help us identify policy-violating
videos. Some of these experts sit at our intel desk, which
proactively looks for new trends in content that might violate
our policies. We also developed an improved escalation pathway
for expert NGOs and governments to notify us of bad content in
bulk through our Trusted Flagger program. We reserve the final
decision on whether to remove videos they flag, but we benefit
immensely from their expertise.
(3) This broad cross-sectional work has led to tangible results.
Over 87 percent of the 9 million videos we removed in the
second quarter of 2019 were first flagged by our automated
systems. More than 80 percent of those auto-flagged videos were
removed before they received a single view. And overall, videos
that violate our policies generate a fraction of a percent of
the views on YouTube.
Our efforts do not end there, as we are constantly evolving to new
challenges and looking for ways to improve our policies. For example,
YouTube recently updated its Hate Speech policy to specifically
prohibit videos alleging that a group is superior in order to justify
discrimination, segregation or exclusion based on qualities like age,
gender, race, caste, religion, sexual orientation or veteran status.
This would include, for example, videos that promote or glorify Nazi
ideology, because it is inherently discriminatory. YouTube also updated
its policies to prohibit content denying that well-documented violent
events, like the Holocaust or the shooting at Sandy Hook Elementary,
took place.
The updated Hate Speech policy was launched in early June, and as
our teams review and remove more content in line with the new policy,
our machine learning algorithms will improve in tandem to help us
identify and remove such content. Though it can take months for us to
ramp up enforcement of a new policy, the profound impact of our Hate
Speech policy update is already evident in the data released in this
quarter's Community Guidelines Enforcement Report: the number of
individual video removals for hate speech saw a 5x spike to over
100,000, the number of channel terminations for hate speech also saw a
5x spike to 17,000, and the total comment removals nearly doubled in Q2
to over 500 million due in part to a large increase in hate speech
removals.
Finally, we go beyond removing policy-violating content by actively
creating programs to promote beneficial counterspeech. These programs
present narratives and elevate credible voices speaking out against
hate, violence, and terrorism. For example, our Creators for Change
program supports creators who are tackling tough issues, including
extremism and hate by building empathy and acting as positive role
models. We launched our most recent Creators for Change global campaign
videos in November 2018. As of June 2019 they already had 59 million
views; the creators involved have over 60 million subscribers and more
than 8.5 billion lifetime views of their channels; and through `Local
Chapters' of Creators for Change, creators tackle challenges specific
to different markets.
Alphabet's Jigsaw group, an incubator to tackle some of the
toughest global security challenges, has deployed the Redirect Method,
which uses targeting tools and curated YouTube playlists to disrupt
online radicalization. The method is open to anyone to use, and NGOs
have sponsored campaigns against a wide-spectrum of ideologically-
motivated terrorists and violent extremists.
Conclusion
We take the safety of our users very seriously and value our close
and collaborative relationships with law enforcement and government
agencies. We have invested substantial resources to tackle the problem
of hate speech. At present, we spend hundreds of millions of dollars
annually and have more than 10,000 people working across Google to
address content that might violate our policies, which include our
policies against promoting violence and terrorism.
We understand these are difficult issues of great interest to
Congress and want to be responsible actors who are a part of the
solution. As these issues evolve, Google will continue to invest in the
people and technology to meet the challenge. We look forward to
continued collaboration with the Committee as it examines these issues.
Thank you for your time. I look forward to taking your questions.
The Chairman. Thank you very much. Mr. Selim, your group
prefers to be known as ADL these days, is that correct?
Mr. Selim. Correct. The Anti-Defamation League goes by ADL
for short.
Mr. Chairman. Great. Well, we appreciate you being with us
today and we are happy to receive your testimony.
STATEMENT OF GEORGE SELIM, SENIOR VICE PRESIDENT,
NATIONAL PROGRAMS, ADL (ANTI-DEFAMATION LEAGUE)
Mr. Selim. Thank you, Mr. Chairman, Ranking Member
Cantwell, thank you for the opportunity to be here with the
distinguished members of this Committee this morning. My name
is George Selim and I serve as the Senior Vice President for
Programs at the ADL or the Anti-Defamation League, and for
decades the ADL has fought against bigotry and anti-Semitism by
exposing extremist groups and individuals who spread hate to
incite violence.
Today, the ADL is the foremost non-governmental authority
on domestic terrorism, extremism, hate groups, and hate crimes.
I have personally served in several roles and the Government's
National Security apparatus at the Department of Justice, the
Department of Homeland Security, at the White House on the
National Security Council, and now outside Government on the
frontlines of combating anti-Semitism and all forms of bigotry
at the ADL. In my testimony, I would like to share with you
some key data, findings, analysis, and urge this Committee to
take action to counter a severe national security threat, the
threat of online white supremacist extremism that threatens our
communities.
The alleged El Paso shooter posted a manifesto to 8chan
prior to the attack. He expressed support for the accused
shooter in Christchurch, New Zealand who also posted on 8chan.
Before the massacre in Poway, California, the alleged shooter
posted a link to his manifesto on 8chan, citing the terrorists
in New Zealand and in the Pittsburgh Tree of Life attack, three
killing sprees, three white supremacist manifestos--one
targeted Muslims, another targeted Jews, and a third targeted
Latino and other immigrant communities. One thing these three
killers had in common was 8chan, an online platform that has
become the go-to for many bigots and extremists.
Unfettered access to online platforms, both fringe and
mainstream, has significantly driven the scale, speed, and
effectiveness of these forms of extremist attacks. Our ADL
research shows that domestic extremist violence is trending up,
and that anti-Semitic hate is trending up. The FBI and DOJ data
shows similar trends. The online environment today amplifies
hateful voices worldwide and facilitates the coordination,
recruitment, and propaganda that fuels the extremism that
terrorizes our communities, all of our communities.
Whether through Government, the private sector, or civil
society, immediate action is paramount to prevent the next
tragedy that could take innocent lives. ADL has worked with the
platforms represented on this table to try to address that hate
and its rampant nature online. We have been part of the
conversations to improve the terms of service, content
moderation programs, and better support for those individuals
experiencing hate and harassment on those platforms.
We appreciate this work greatly but much more needs to be
done. ADL has called on these companies at this hearing as well
as many others to be far more transparent about the prevalence
and nature of hate on their platforms. We need meaningful
transparency to give actionable information to policymakers and
stakeholders, but the growth of hate and extremist violence
will not be solved by addressing these issues online alone. We
urge this Committee to take immediate action.
First, our Nation's leaders must clearly and forcefully
call out bigotry in all its forms at every opportunity. Our
Nation's law enforcement leadership must make enforcing hate
crimes laws a top priority. Our communities need this
Congress's immediate action on a range of ways, notably to
codify Federal offices to address domestic terrorism and
extremism and create transparent and comprehensive reporting
such as that required in the Domestic Terrorism Prevention Act
and similar measures in the Domestic Terrorism Data Act. Our
Federal legal system currently lacks the means to prosecute a
white supremacist terrorist as a terrorist. Congress should
explore whether it is possible to craft a rights protecting
domestic terrorism statute.
Any statute that Congress should consider would need to
include specific, careful Congressional and civil liberties
oversight to ensure the spirit of such protections is
faithfully executed. In addition, the State Department should
examine whether certain foreign, white supremacist groups meet
the criteria for designation an FTO, foreign terrorist
organizations. For technology and social media companies, we
look forward to companies expanding their terms of service and
exploring accountability and governance challenges, aspiring to
greater transparency in how you address these issues and
partnering with civil society groups to help in all of these
efforts.
ADL stands ready both with both the Government and the
private sector to better address all forms and threats online.
This is an all-hands-on-deck moment to protect all of our
communities. I look forward to your questions. Mr. Chairman,
Ranking Member, and other distinguished members of this
Committee. Thank you.
[The prepared statement of Mr. Selim follows:]
Prepared Statement of George Selim, Senior Vice President,
National Programs, ADL (Anti-Defamation League)
Introduction
Since 1913, the mission of ADL (Anti-Defamation League) has been to
``stop the defamation of the Jewish people and to secure justice and
fair treatment to all.'' For decades, ADL has fought against bigotry
and anti-Semitism by exposing extremist groups and individuals who
spread hate and incite violence. Today, ADL is the foremost non-
governmental authority on domestic terrorism, extremism, hate groups,
and hate crimes. ADL plays a leading role in exposing extremist
movements and activities, while helping communities and government
agencies alike in combating them. ADL's team of experts--analysts,
investigators, researchers, and linguists--use cutting-edge
technologies and investigative techniques to track and disrupt
extremists and extremist movements worldwide. ADL provides law
enforcement officials and the public with extensive resources,
including analytic reports on extremist trends and databases of Hate
Symbols and Terror Symbols that can help alert online platforms of
problematic content.
White Supremacy and Mass Shootings \1\
---------------------------------------------------------------------------
\1\ Datasets for this section are available on ADL's HEAT Map: ADL,
ADL H.E.A.T Map, updated June 19, 2019, https://www.adl.org/education-
and-resources/resource-knowledge-base/adl-heat-map.
---------------------------------------------------------------------------
When white supremacist Robert Bowers entered the Tree of Life
Synagogue in Pittsburgh in October 2018 to launch a killing spree
against Jews attending services, taking 11 lives and wounding seven
more, his senseless and hate-fueled violence directly impacted not just
the victims' families, friends and neighbors, but all residents of
Pittsburgh--and communities nationwide and around the world. The
deadliest attack against American Jews, unfortunately, was only one of
many in the past year tied to a white supremacist ideology that has
found fertile ground online with consequences affecting not only
Americans but people around the world. Extremist-related killings are
comparatively few when compared to the total number of homicides in the
U.S. each year. Nevertheless, such killings, especially when they are
committed as hate crimes or terrorist attacks, can send shock waves
through entire communities--and beyond. A list of selected white
supremacist shooting sprees is included at the end of this document.
Recent analysis by ADL's Center on Extremism shows that domestic
extremists took the lives of at least 50 people in 2018, a sharp
increase from the 37 people killed by extremists in 2017. In fact, 2018
is the fourth-deadliest year since 1970, behind only 1995 (which saw
184 deaths, most attributed to the Oklahoma City bombing), 2016 (72
deaths) and 2015 (70 deaths).
2018's high death toll is due in large part to the number of
shooting sprees by extremists. In 2017, only one extremist-related
shooting spree occurred; in 2018, there were five shooting sprees
collectively responsible for 38 deaths and 33 wounded. There were fewer
lethal incidents in 2018 than in 2017 (17 compared to 21), but the
events were significantly deadlier--and the 2018 shooting sprees were
responsible for most of the deaths.
These attacks are in large part intensified by the use of guns. In
both high-and low-casualty attacks, domestic extremists used guns in 42
of the 50 murders they committed in 2018, far outpacing edged weapons
or physical assaults. Over the past ten years, firearms were used in 73
percent of domestic extremist-related killings in the United States.
Guns are the weapon of choice among America's extremist murderers,
regardless of their ideology.
White supremacists were responsible for the great majority of
extremist-related killings in 2018, which is the case almost every
year. Right-wing extremists were responsible for 49 (or 98 percent) of
the 50 domestic extremist-related killings in 2018, with white
supremacists alone accounting for 39 (or 78 percent) of those murders.
Hate Crimes in America
While most anti-Semitic incidents are not directly perpetrated by
extremists or white supremacists, there are important connections
between the trends. We found in our annual Audit of Anti-Semitic
Incidents that in 2018, 249 acts of anti-Semitism (13 percent of the
total incidents) were attributable to known extremist groups or
individuals inspired by extremist ideology, making it the highest level
of anti-Semitic incidents with known connections to extremists or
extremist groups since 2004.\2\ Of those, 139 incidents were part of
fliering campaigns by white supremacist groups. Another 80 were anti-
Semitic robocalls allegedly perpetrated by anti-Semitic podcaster Scott
Rhodes in support of the candidacy of Patrick Little, an unabashed
white supremacist who ran an unsuccessful campaign for U.S. Senate in
California.
---------------------------------------------------------------------------
\2\ ADL, 2018 Audit of Anti-Semitic Incidents, https://www.adl.org/
audit2018, April 2019.
---------------------------------------------------------------------------
The Audit also noted spikes at several points during the year. The
final three months of the year were unusually active, with 255
incidents in October, 300 in November and 194 in December. The high
number in October included 45 propaganda distributions by white
supremacists. The incidents in November and December immediately
followed the Pittsburgh massacre, which likely drew more attention to
anti-Semitic activities. Incidents first spiked in May, when 209 anti-
Semitic acts were reported, including 80 anti-Semitic robocalls sent by
white supremacists, which targeted Jewish individuals and institutions
with harassing messages.
Hate crimes are only an element of the anti-Semitic incidents that
we track. The most recent data about hate crimes made available by the
FBI is for 2017.\3\ The FBI has been tracking and documenting hate
crimes reported from federal, state, and local law enforcement
officials since 1991 under the Hate Crimes Statistics Act of 1990
(HCSA). Though clearly incomplete, the Bureau's annual HCSA reports
provide the best single national snapshot of bias-motivated criminal
activity in the United States. The Act has also proven to be a powerful
mechanism to confront violent bigotry, increasing public awareness of
the problem and sparking improvements in the local response of the
criminal justice system to hate violence--since in order to effectively
report hate crimes, police officials must be trained to identify and
respond to them.
---------------------------------------------------------------------------
\3\ FBI, 2017 Hate Crime Statistics, 2017 https://ucr.fbi.gov/hate-
crime/2017, November 2018.
---------------------------------------------------------------------------
The FBI documented 7,175 hate crimes reported by 16,149 law
enforcement agencies across the country--the highest level of
participation since the enactment of the HCSA, and a 6 percent increase
over 2016 participation of 15,254. Of the 7,175 total incidents:
Religion-based crimes increased 23 percent, from 1,273 in
2016 to 1,564 in 2017--the second highest number of religion-
based crimes ever [only 2001, after 9/11, recorded more--
1,828].
Crimes directed against Jews increased 37 percent--from 684
in 2016 to 938 in 2017. Crimes against Jews and Jewish
institutions were slightly more than 13 percent of all reported
hate crimes--and 60 percent of the total number of reported
religion-based crimes. Every year since 1991, crimes against
Jews and Jewish institutions have been between 50 and 80
percent of all religion-based hate crimes.
Race-based crimes were the most numerous (as they have been
every year since 1991), totaling 4,131 crimes, almost 58
percent of the total. Crimes against African-Americans, as
always, were the plurality of these crimes--2,013, about 28
percent of all reported hate crimes.
Reported crimes against Muslims decreased 11 percent, from
307 in 2016 to 273 in 2017. However, the 273 anti-Muslim hate
crimes recorded was the highest reported number of crimes
against Muslims ever--behind 2016's 307 and 481 in 2001, after
the 9/11 terrorist attacks.
Crimes directed against LGBTQ people increased from 1,076 in
2016 to 1,130 in 2017. Crimes directed against individuals on
the basis of their gender identity decreased slightly, from 124
in 2016 to 119 in 2017, slightly less than two percent of all
hate crimes.
Importantly, only 2,040 of the 16,149 reporting agencies--less than
13 percent--reported one or more hate crimes to the FBI. That means
that about 87 percent of all participating police agencies
affirmatively reported zero (0) hate crimes to the FBI (including at
least 92 cities over 100,000). And more than 1,000 law enforcement
agencies did not report any data to the FBI (including 9 cities over
100,000).
Moreover, we need to remember that these are only reported crimes.
Many communities and individuals do not feel comfortable going to law
enforcement for a variety of reasons, so there is likely an undercount
of hate crimes resulting from unwillingness to report.
The Role of Online Platforms in White Supremacist Violence
The real-world violence of extremists does not emerge from a
vacuum. In many cases the hatred that motivates extremist violence, and
especially these documented white supremacist murders, is nurtured in
online forums such as Gab, 4chan, 8chan, and other platforms.\4\
---------------------------------------------------------------------------
\4\ Anti-Defamation League, '' Hatechan: The Hate and Violence-
Filled Legacy of 8chan,'' ADL Blog, August 7, 2019, https://
www.adl.org/blog/hatechan-the-hate-and-violence-filled-legacy-of-8chan;
ADL, Gab and 8chan: Home to Terrorist Plots Hiding in Plain Sight,
https://www.adl.org/resources/reports/gab-and-8chan-home-to-terrorist-
plots-hiding-in-plain-sight.
---------------------------------------------------------------------------
Extremist groups are empowered by access to the online world; the
Internet amplifies the hateful voices of the few to reach millions
around the world. The online environment also offers community: while
most extremists are unaffiliated with organized groups, online forums
allow isolated extremists to become more active and involved in virtual
campaigns of ideological recruitment and radicalization. As Internet
proficiency and the use of social media are nearly universal, the
efforts of terrorist and extremist movements to exploit these
technologies and platforms to increase the accessibility of materials
that justify and instigate violence are increasing exponentially. Both
terrorist and extremist movements, here at home and abroad, use online
and mobile platforms to spread their messages and to actively recruit
adherents who live in the communities they target.
Individuals can easily find sanction, support, and reinforcement
online for their extreme beliefs or actions, and in some cases neatly
packaged alongside bomb-making instructions. This enables adherents
like violent white supremacist mass shooters such as Bowers to self-
radicalize without face-to-face contact with an established terrorist
group or cell.
Perhaps the most important contributor to the subculture of white
supremacists are the so-called ``imageboards,'' a type of online
discussion forum originally created to share images. One of the most
prominent is 4chan, a 15-year-old imageboard whose influence extends
far beyond the alt right, as a key source of Internet memes. Its/pol
subforum is a disturbing site, an anarchic collection of posts that
range from relatively innocuous to highly offensive, with most users
posting content anonymously.
Due in part to its extremely lax content moderation policies, 4chan
has become home to many racists and openly and vocal white
supremacists. Some of its imitators, such as 8chan, lean even more
towards racism and white supremacy. Parts of Reddit, a popular website
that contains a massive collection of subject-oriented discussion
threads, also share the ``chan'' subculture.
ADL has assessed that individuals do not primarily utilize 8chan
for sharing hateful images and messages, but they also use it to turn
real-world killings into entertainment, canonizing the perpetrators of
previous massacres and keeping track of their respective body counts,
like scores in a video game.
The current ADL assessment is that at its core, 8chan is a haven
for both violent daydreamers and real-life murderers to virtually meet,
network and recruit more followers. This intersection poses
considerable risk both online and in the physical world.
Patrick Crusius, the alleged El Paso shooter charged with killing
22 people and injuring many more, is believed to have posted a four-
page manifesto to 8chan prior to the attack. His justification for the
deadly spree was that he was defending his country from ``cultural and
ethnic replacement brought on by an invasion.'' \5\
---------------------------------------------------------------------------
\5\ Anti-Defamation League, ``Mass Shooting in El Paso: What We
Know,'' ADL Blog, August 4, 2019, https://www.adl.org/blog/mass-
shooting-in-el-paso-what-we-know.
---------------------------------------------------------------------------
One of the most telling elements of Crusius's post is that in it,
he also expressed support for Australian, white supremacist, mass-
murderer Brenton Tarrant, the accused shooter in the March 2019 mosque
attacks in Christchurch, New Zealand that left 51 people dead.\6\
---------------------------------------------------------------------------
\6\ Ibid.
---------------------------------------------------------------------------
Like the El Paso shooter, we assess that Tarrant likely turned to
8chan to post what he referred to as a ``explanation'' for his deadly
rampage, providing links to his own manifesto, which he called ``The
Great Replacement.'' In it, he fixated on the white supremacist theory
that white European society will be overrun by migration from Muslim
and African nations.\7\
---------------------------------------------------------------------------
\7\ Anti-Defamation League, ``White Supremacist Terrorist Attacks
at Mosques in New Zealand,'' March 15, 2019, https://www.adl.org/blog/
white-supremacist-terrorist-attack-at-mosques-in-new-zealand.
---------------------------------------------------------------------------
In his manifesto, Tarrant addressed the 8chan community directly--
as if they were co-conspirators--explicitly directing them to ``do your
part.''
Just one month later, someone did. Before his massacre at the
Chabad Congregation in Poway, California, the shooter posted a link to
his own manifesto on 8chan, offering the same kind of white supremacist
tropes and cited the Christchurch and Pittsburgh shooters for inspiring
his own deadly attacks.
Three white supremacist manifestos, three killing sprees. One
targeted Muslims, another Jews, the third Latinx and immigrants. What
these three men had in common was 8chan, the platform for their final
messages.
While the most extreme forms of online content normally thrive on
platforms like 8chan, Gab, and 4chan, larger social media platforms
like Facebook, Twitter, and YouTube must also remain vigilant.
Extremists leverage larger mainstream platforms to ensure that the
hateful philosophies and messages that begin to germinate on message
boards like Gab and 8chan find a new and much larger audience.
Twitter's 300 million users and Facebook's 2.4 billion dwarf the
hundreds of thousands on 8chan and Gab. Extremists make use of
mainstream platforms in specific and strategic ways to exponentially
increase their audience while avoiding content moderation activity that
Facebook and Twitter use to remove hateful content. These include
creating private pages and events, sharing links that directly lead
users to extreme content on websites like 8chan and using coded
language called ``dogwhistles'' to imply and spread hateful ideology
while attempting to circumvent content moderation systems.
Since the white supremacist rally in Charlottesville in 2017 and
subsequent attacks and murders by extremists to date, there have been
many well-publicized efforts by the technology and social media
companies that run mainstream social platforms and services to stem the
tide of hate and extremism online. After Charlottesville, tech
companies ranging from large social platforms like Facebook to payment
processors like Paypal to cybersecurity services like Cloudflare took
action to expel white supremacists from their services. Even so, these
same companies and others in this market sector have been forced to
repeatedly respond to violent white supremacist activity on their
platforms in the past 12 months. The Christchurch video was streamed on
Facebook live, leading Facebook to change its livestreaming policy.\8\
Paypal provided payment services to the fringe platform Gab, where the
Pittsburgh shooter was believed to be radicalized, but cut off its
services after the massacre.\9\ Cloudflare provided cybersecurity
services to 8chan, and publicly cut it off after the site was
implicated in the shooting in El Paso (among others).\10\ Although it
appears that these companies and others took significant action to
address white supremacy and hate in 2017 and claim to have continued to
do so, ADL assesses that the above-mentioned platforms are still being
abused, including today, by people espousing this hateful and violent
ideology even two years later.
---------------------------------------------------------------------------
\8\ ``Christchurch Attacks: Facebook Curbs Live Feature,'' BBC
News, May 15, 2019, https://www.bbc.com/news/technology-48276802.
\9\ Adam Smith, ``GoDaddy and PayPal Ban Gab After Pittsburgh
Shooting,'' PCMag, October 28, 2018, https://www.pcmag.com/news/364650/
godaddy-and-paypal-ban-gab-after-pittsburgh-shooting.
\10\ Matthew Prince, ``Terminating Service for 8chan,'' Cloudflare,
August 5, 2019, https://blog.cloudflare.com/terminating-service-for-
8chan/.
---------------------------------------------------------------------------
Scoping the Problem
One of the key drivers of these complicated and at times deadly
issues is the size and scale of these platforms. For example, on
Twitter approximately 6,000 tweets are posted every second and
approximately 500 million tweets are posted every day. If the company's
policies and systems operated at 99 percent effectiveness in detecting
and responding to violent hate and extremist rhetoric, that would still
leave five million tweets unaddressed every day. Imagine that each of
those tweets, on the low end, reached just 60 people: those tweets
would reach the number of people equal roughly to the population of the
United States (330 million people) every day.
The policies and systems of these companies are very likely not
operating with a high degree of accuracy, leaving possibly millions of
users exposed and impacted by hateful and extreme content every day. As
an example, YouTube in June 2019 announced a policy change focusing on
prohibiting white nationalist and other extremist content from existing
on its platform.\11\ In August 2019, an ADL investigation found a
number of prominent white nationalists and other forms of hateful
extremists still active and easily found on the platform, despite the
policy change.\12\ Similarly, after Facebook very publicly banned Alex
Jones from its platforms in May 2019, Jones was quickly able to shift
his operations to another account on the platform.\13\ These instances
raise alarming questions about the degree to which social media
platforms, through their own internal policies and systems, are able to
meaningfully detect, assess, and act on hateful content at the global
scale their platforms operate.
---------------------------------------------------------------------------
\11\ Casey Newton, ``YouTube Just Banned White Supremacist Content,
and Thousands of Channels are About to be Removes,'' The Verge, June 5,
2019, https://www.theverge.com/2019/6/5/18652576/youtube-supremacist-
content-ban-borderline-extremist-terms-of-service.
\12\ Anti-Defamation League, ``Despite YouTube Policy Update, Anti-
Semitic, White Supremacist Channels Remain,'' ADL Blog, August 15,
2019, https://www.adl.org/blog/despite-youtube-policy-update-anti-
semitic-white-supremacist-channels-remain.
\13\ Craig Timberg, ``Alez Jones Banned from Facebook? His videos
are still there--and so are his followers,'' The Washington Post,
November 5, 2018, https://beta.washingtonpost.com/technology/2018/11/
05/alex-jones-banned-facebook-his-videos-are-still-there-so-are-his-
followers/.
---------------------------------------------------------------------------
The U.S. Congress and American public admittedly have limited
knowledge of just how well platforms are dealing with the problem of
white supremacist extremism. To evaluate their efforts, civil society
organizations like ADL can conduct limited external research similar to
the manner mentioned above, in which we use the platform information
that is publicly available to objectively assess the stated actions and
policy implications of a given platform. Or we can look to the
platforms' own limited efforts at transparency about their policies and
practices. The mainstream social media platforms have several
potentially relevant metrics related to the issue of extremism,
especially white supremacist extremism, that they share in their
regular transparency reports. These differ slightly as described by
each platform. The metrics are self-reported by the companies, and
there is no way to fully understand the classification of content
categories outside of the brief descriptions given by the platforms as
part of this reporting.
For example, the platforms provide information related to
terrorism. Facebook reported 6.4 million pieces of content related to
terrorist propaganda removed from January to March 2019. This may seem
meaningful, but it is not a particularly insightful datapoint.
Typically, the social media platform companies are only looking at
international terrorism from designated groups such as Al Qaeda and
ISIS and are not including white supremacist violence and related
activity as part of this terrorism classification.
White supremacist content could fall under the category of hate
speech or violent content on a platform. Twitter reported 250,806
accounts actioned for hateful conduct and 56,577 accounts actioned for
violent threats from July to December 2018. Yet a wide variety of other
types of content not associated with extremism or white supremacy might
also fall in this category, making it difficult to glean meaningful
analysis about white supremacist content from these metrics.
Additionally, when Facebook claims in its transparency report that
it took action on four million pieces of hate speech from January to
March 2019, it is difficult to understand what this means in context as
we do not know how that compares to the level of hate speech reported
to them, which communities are impacted by those pieces of content, or
whether any of that content is connected with extremist activity on
other parts of their platform.
In order to truly assess the problem of hate and extremism on
social media platforms, technology companies must provide meaningful
transparency with metrics that are agreed upon and verified by trusted
third parties, like ADL, and that give actionable information to users,
civil society groups, governments, and other stakeholders. Meaningful
transparency will allow stakeholders to answer questions such as: ``How
significant is the problem of white supremacy on this platform?'' ``Is
this platform safe for people who belong to my community?'' ``Have the
actions taken by this company to improve the problem of hate and
extremism on their platform had the desired impact?'' Until tech
platforms take the collective actions to come to the table with
external parties and meaningfully address these kinds of questions
through their transparency efforts, our ability to understand the
extent of the problem of hate and extremism online, or how to
meaningfully and systematically address it, will be extremely limited.
What We Know About Online Hate and Harassment
One way in which ADL has tried to address this gap in knowledge is
by conducting a national representative survey on the hate and
harassment experienced by Americans online. Our survey found that over
half of respondents (53 percent) experienced some type of online
harassment; 37 percent of American adults reported experiencing severe
harassment (including physical threats, sexual harassment, stalking and
sustained harassment), up from 18 percent in 2017.\14\
---------------------------------------------------------------------------
\14\ ADL, Online Hate and Harassment: The American Experience,
2019, https://www.adl.org/onlineharassment.
---------------------------------------------------------------------------
We also found that identity-based harassment was most common
against LGBTQ+ individuals, with 63 percent of LGBTQ+ respondents
experiencing harassment because of their sexual orientation. Religious-
based harassment was very common against Muslims (35 percent) and, to a
lesser extent, Jewish (16 percent) respondents.
Harassment was also common among other minority groups, with race-
based harassment affecting 30 percent of Hispanics or Latinos, 27
percent of African-Americans, and 20 percent of Asian-Americans.
Finally, women also experienced harassment disproportionately, with
gender identity-based harassment affecting 24 percent of female-
identified respondents, compared to 15 percent of male-identified.\15\
---------------------------------------------------------------------------
\15\ Ibid.
---------------------------------------------------------------------------
Hate and harassment are also endemic to online games. Fifty-three
percent of the total population of the United States and 64 percent of
the online population of the United States plays video games. Following
our wider online survey, we surveyed Americans who play online games
and found that 74 percent of respondents experienced some form of
harassment while playing games online. Sixty-five percent of players
experienced some form of severe harassment, including physical threats,
stalking, and sustained harassment.\16\
---------------------------------------------------------------------------
\16\ ADL, Free to Play? Hate, Harassment, and Positive Social
Experiences in Online Games, July 2019, https://www.adl.org/free-to-
play, page 18.
---------------------------------------------------------------------------
We are also seeing an increase in extremist and white supremacist
content within online games and gaming forums. Scholars have observed
white supremacist recruiters actively prey on disaffected youth within
the gaming community, and use these channels to plant seeds of hate by
invoking sentiments of ``us versus them.'' Our survey found that nearly
a quarter of players (23 percent) are exposed to discussions about
white supremacist ideology and almost one in ten (9 percent) are
exposed to discussions about Holocaust denial in online multiplayer
games. These are alarming insights into an industry that has managed to
avoid the intense media scrutiny that more traditional social media
platforms have experienced.\17\
---------------------------------------------------------------------------
\17\ Ibid, page 7.
---------------------------------------------------------------------------
Online hate and harassment, whether carried out by extremists or
simply by those who feel freer to harm others by the distance and
anonymity of being online have real-life, sometimes devastating
consequences. Our online game survey found that 23 percent of harassed
players become less social and 15 percent felt isolated as a result of
in-game harassment. One in ten players had depressive or suicidal
thoughts as a result of harassment in online multiplayer games, and
nearly one in ten took steps to reduce the threat to their physical
safety (8 percent).\18\ Alarmingly, nearly a third of online
multiplayer gamers (29 percent) had been doxed--had their personal
information shared with the goal of harassment.\19\
---------------------------------------------------------------------------
\18\ Ibid, page 27.
\19\ Ibid, page 18.
---------------------------------------------------------------------------
Our wider survey found that among those who had been targeted, or
feared being targeted, approximately 38 percent stopped, reduced or
changed their activities online, such as posting less often, avoiding
certain sites, changing privacy setting, deleting apps, or increasing
filtering of content or users. Some 15 percent took steps to reduce
risk to their physical safety, such as moving locations, changing their
commute, taking a self-defense class, avoiding being alone, or avoiding
certain locations.\20\
---------------------------------------------------------------------------
\20\ ADL, Online Hate and Harassment: The American Experience.
---------------------------------------------------------------------------
Our survey also found societal consequences among respondents. More
than half (59 percent) said that online hate and harassment were making
hate crimes more common, and half said that they are increasing the use
of derogatory language. More than one-third (39 percent) thought that
online hate and harassment are making young Americans lose faith in the
country, and 30 percent believed that they are making it harder to
stand up to hate. Some felt less comfortable in their more immediate
environments: approximately 22 percent of Americans report that online
hate and harassment makes them feel less safe in their community while
18 percent feel that it makes family members trust each other less.\21\
---------------------------------------------------------------------------
\21\ Ibid.
---------------------------------------------------------------------------
Critically, those surveyed wanted to see private technology
companies take action to counter or mitigate online hate and
harassment. Eighty-four percent said that platforms should do more,
including making it easier for users to filter (81 percent) and report
(76 percent) hateful and harassing content. In addition, Americans want
companies to label comments and posts that appear to come from
automated ``bots'' rather than people. Finally, a large percentage of
respondents were in favor of platforms removing problematic users as
well as having outside experts independently assess the amount of hate
on a platform.\22\
---------------------------------------------------------------------------
\22\ Ibid.
---------------------------------------------------------------------------
Over 80 percent of those surveyed wanted government to act by
strengthening laws and improving training and resources for police on
cyberhate. Strong support exists for these changes regardless of
whether an individual has previously experienced online hate and
harassment and regardless of political belief. Although respondents
identifying as liberal reported even greater agreement with the
actions, those identifying as conservatives overwhelmingly supported
all the actions as well.\23\
---------------------------------------------------------------------------
\23\ Ibid.
---------------------------------------------------------------------------
Moving Forward: Policy Recommendations to Counter the Threat
1. Bully Pulpit The President, cabinet officials, and Members of
Congress must call out bigotry at every opportunity. The right
to free speech is a core value, but the promotion of hate
should be vehemently rejected. Simply put, you cannot say it
enough: America is no place for hate.
2. Enforcement of Existing Laws The Administration must send loud,
clear, and consistent messages that violent bigotry is
unacceptable and ensure that the FBI and the Justice
Department's Civil Rights Division will enforce relevant
Federal laws and vigorously investigate and prosecute hate
crimes.
3. Improve Federal Hate Crime Training and Data Collection The
Department of Justice should incentivize and encourage state
and local law enforcement agencies to more comprehensively
collect and report hate crimes data to the FBI, with special
attention devoted to large underreporting law enforcement
agencies that either have not participated in the FBI Hate
Crime Statistics Act program at all or have affirmatively and
not credibly reported zero hate crimes. More comprehensive,
complete hate crime reporting can deter hate violence and
advance police-community relations. In addition, the
administration, DHS and DOJ should take steps to ensure that it
is efficient and safe for all victims of hate crimes to contact
the police. If marginalized or targeted community members--
including immigrants, people with disabilities, LGBTQ community
members, Muslims, Arabs, Middle Easterners, South Asians and
people with limited language proficiency--cannot report, or do
not feel safe reporting hate crimes, law enforcement cannot
effectively address these crimes, thereby jeopardizing the
safety of all.
4. Legislation to Address White Supremacy and Domestic Terrorism
Congress must act to counter the threat of domestic terrorism
and prevent more attacks. No legislative action is perfect, but
inaction should not be an option. Congress should enact the
following measures:
Domestic Terrorism Prevention Act (DTPA) (S. 894/HR
1931) This legislation would enhance the Federal
government's efforts to prevent domestic terrorism by
authorizing into law the offices addressing domestic
terrorism, and would require Federal law enforcement
agencies to regularly assess those threats. The bill would
also provide training and resources to assist non-federal
law enforcement in addressing these threats, requiring DOJ,
DHS, and the FBI to provide training and resources to
assist state, local, and tribal law enforcement in
understanding, detecting, deterring, and investigating acts
of domestic terrorism.
Domestic Terrorism Documentation and Analysis of
Threats in America (DATA) Act (HR 3106) Data on extremism
and domestic terrorism is being collected by the FBI, but
not enough, and the reporting is insufficient and flawed.
Data drives policy; we cannot address what we are not
measuring. The DATA Act focuses on increasing the
coordination, accountability, and transparency of the
Federal government in collecting and recording data on
domestic terrorism.
The Khalid Jabara and Heather Heyer National
Opposition to Hate, Assault, and Threats to Equality Act of
2019 (NO HATE Act of 2019 S. 2043/H.R. 3545) This
legislation would authorize incentive grants to spark
improved local and state hate crime training, prevention,
best practices, and data collection initiatives--including
grants for state hate crime reporting hotlines to direct
individuals to local law enforcement and support services.
Disarm Hate Act (S. 1462/H.R. 2708) This legislation
would close the loophole that currently permits the sale of
firearms to individuals who have been convicted of
threatening a person based on their race, religion, gender,
sexual orientation, or disability. The measure would
prohibit individuals convicted of a misdemeanor hate crime
from obtaining a firearm.
In addition, more consideration is needed for two additional
initiatives that could help address white supremacy and domestic
terrorism in the United States.
Congress should examine whether a rights-protecting domestic
terrorism criminal charge is needed--and could be appropriately
crafted. Our Federal legal system currently lacks the means to
prosecute a white supremacist terrorist as a terrorist.
Perpetrators can be prosecuted for weapons charges, acts of
violence (including murder), racketeering, hate crimes, or
other criminal violations. But we cannot legally prosecute them
for what they are: terrorists. Many experts have argued that,
without being so empowered, there is a danger that would-be
domestic terrorists are more likely to be charged with lesser
crimes and subsequently receive lesser sentences. Congress
should begin immediate hearings and consultations with legal
and policy experts, marginalized communities, and law
enforcement professionals on whether it is possible to craft a
rights-protecting domestic terrorism statute. Any statute
Congress would seriously consider should include specific,
careful Congressional and civil liberties oversight to ensure
the spirit of such protections are faithfully executed.
The State Department should examine whether certain white
supremacist groups operating abroad meet the specific criteria
to be subject to sanctions under its Designated Foreign
Terrorist Organization (FTO) authority. The criteria, set out
in 8 U.S.C. Sec. 1189(a)[1] are: (1) the
organization must be foreign; (2) the organization must engage
in terrorist activity or retain the capability and intent to
engage in terrorist activity or terrorism; and (3) the
terrorist activity or terrorism of the organization must
threaten the security of U.S. nationals or the national
security of the U.S.
---------------------------------------------------------------------------
\[1]\ ``8 U.S. Code Sec. 1189.Designation of foreign
terrorist organizations,'' Cornell Law School Legal Information
Institute, accessed September 16, 2019; (https://www.law.cornell.edu/
uscode/text/8/1189)
None of the current 68 organizations on the FTO list is a
white supremacist organization.[2] And while the
possibility of designating white supremacist organizations
under the State Department's FTO authority holds promise, there
are some important considerations that must be taken into
account.
---------------------------------------------------------------------------
\[2]\ State Department, ``Foreign Terrorist Organizations,''
accessed September 16, 2019; (https://www.state.gov/foreign-terrorist-
organizations/)
First, while several countries have added white supremacist
groups to their own designated terrorist lists in recent days--
including Canada [3] and England [4]--
white supremacist groups do not operate exactly like other
FTOs, such as ISIS and al-Qaeda. For example, individual white
supremacists that carry out attacks--wherever they are--very
rarely receive specific operational instructions from organized
white supremacist groups abroad to carry out these attacks.
---------------------------------------------------------------------------
\[3]\ Harmeet Kaur, ``For the first time, Canada adds white
supremacists and neo-Nazi groups to its terror organization list,''
CNN, June 28, 2018, (https://www.cnn.com/2019/06/27/americas/canada-
neo-nazi-terror-organization-list-trnd/index.html)
\[4]\ Emma Lake, ``Terror Crackdown: Which terror groups are banned
under UK law and when was National Action added to the list?'' The Sun
(UK), October 26, 2017 (https://www.thesun.co.uk/news/4569388/banned-
terror-groups-uk-national-action)
These groups generally do not have training camps in Europe
or elsewhere where individuals travel to learn tactics and then
return home to carry out an attack. Instead, individuals in the
United States are typically motivated to act based on their own
white supremacist ideology, which primarily stems from domestic
sources of inspiration but which can sometimes also stem from
inspirational sources abroad--including the violent actions of
white supremacists--whether that foreign source is associated
with an organization or not. Second, in the United States,
unlike in Canada and England, the First Amendment provides
unique, broad protection for even the most vile hate speech and
propaganda. While clearly criminal conduct would not be
protected under the First Amendment, a great deal of non-
criminal association, speech, and hateful propaganda would be
protected speech. The First Amendment's assembly and speech
protections would not permit designation of white supremacist
organizations operating here, but designating foreign white
supremacist groups could make knowingly providing material
support or resources to them a crime--extending authority for
law enforcement officials to investigate whether such a crime
is being planned or is occurring.[5]
---------------------------------------------------------------------------
\[5]\ Mary B. McCord, ``White Nationalist Killers Are Terrorists.
We Should Fight Them Like Terrorists,'' Washington Post, Aug. 8, 2019,
(https://www.washingtonpost.com/outlook/white-nationalist-killers-are-
terrorists-we-should-fight-them-like-terrorists/2019/08/08/3f8b761a-
b964-11e9-bad6-609f75bfd97f_story.html)
---------------------------------------------------------------------------
5. Address Online Hate and Harassment
Strengthen laws against perpetrators of online hate Hate and
harassment translate from real-world to online spaces,
including in social media and games, but our laws have not kept
up. Many forms of severe online misconduct are not consistently
covered by cybercrime, harassment, stalking and hate crime law.
Congress has an opportunity to lead the fight against cyberhate
by increasing protections for targets as well as penalties for
perpetrators of online misconduct. Some actions Congress can
take include revising Federal law to allow for penalty
enhancements based on cyber-related conduct; updating Federal
stalking and harassment statutes' intent requirement to account
for online behavior; and legislating specifically on
cybercrimes such as doxing, swatting, non-consensual
pornography, and deepfakes.
Urge social media platforms to institute robust governance
Government officials have an important role to play in
encouraging social media platforms to institute robust and
verifiable industry-wide self-governance. This could take many
forms, including Congressional oversight or passing laws that
require certain levels of transparency and auditing. The
Internet plays a vital role in allowing for innovation and
democratizing trends, and that should be preserved. At the same
time the ability to use it for hateful and severely harmful
conduct needs to be effectively addressed.
Improve training of law enforcement Law enforcement is a key
responder to online hate, especially in cases when users feel
they are in imminent danger. Increasing resources and training
for these departments is critical to ensure they can
effectively investigate and prosecute cyber cases and that
targets know they will be supported if they contact law
enforcement.
6. Platform Responsibility to Address Online Hate and Harassment
Terms of Service Every social media and online game platform
must have clear terms of service that address hateful content
and harassing behavior, and clearly define consequences for
violations. These policies should state that the platform will
not tolerate hateful content or behavior based on protected
characteristics. They should prohibit abusive tactics such as
harassment, doxing and swatting. Platforms should also note
what the process of appeal is for users who feel their content
was flagged as hateful or abusive in error.
Responsibility and Accountability Social media and online
game platforms should assume greater responsibility to enforce
their policies and to do so accurately at scale. They should
improve the complaint and flagging process so that it provides
a more consistent and speedy resolution for targets. They
should lessen the burden of the complaint process for users,
and instead proactively, swiftly, and continuously addressing
hateful content using a mix of artificial intelligence and
humans who are fluent in the relevant language and
knowledgeable in the social and cultural context of the
relevant community.
Additionally, given the prevalence of online hate and harassment,
platforms should offer far more services and tools for
individuals facing or fearing online attack. They should
provide greater filtering options that allow individuals to
decide for themselves how much they want to see likely hateful
comments. They should consider the experience of individuals
who are being harassed in a coordinated way, and be able to
provide aid to these individuals in meaningful ways. They
should allow users to speak to a person as part of the
complaint process in certain, clearly defined cases. They
should provide user-friendly tools to help targets preserve
evidence and report problems to law enforcement and companies.
Governance and Transparency Perhaps most importantly, social
media and online game platforms should adopt robust governance.
This should include regularly scheduled external, independent
audits so that the public knows the extent of hate and
harassment on a given platform. Audits should also allow the
public to verify that the company followed through on its
stated actions and assess the effectiveness of company efforts
over time. Companies should provide information from the audit
and elsewhere through more robust transparency reports.
Finally, companies should create independent groups of experts
from relevant stakeholders, including civil society, academia
and journalism, to help provide guidance and oversight of
platform policies.
Beyond their own community guidelines, transparency efforts and
content moderation policies, features available on social media and
online game platforms need to be designed with anti-hate principles in
mind. Companies need to conduct a thoughtful design process that puts
their users first, and incorporates risk and radicalization factors
before, and not after, tragedy strikes. Today, the most popular method
of developing technology tools is through a Software Prototyping
approach: an industry-wide standard that prompts companies to quickly
release a product or feature and iterate on it over time. This approach
completely devalues the impact of unintended design consequences. For
example, the Christchurch shooter used Facebook's livestreaming feature
to share his attack with the world. The feature could have been
designed to limit or lock audiences for new or first-time streamers or
prevent easy recording of the video.
These kinds of attacks, designed to leverage social media to
attract maximum attention and encourage the next attack, force us to
reassess the threat of hateful echo chambers like 8chan as well as the
exploitable features in mainstream platforms like Facebook--and how
they help drive extremist violence.
Conclusion
ADL data clearly and decisively illustrates that hate is rising
across America. Hate has found fertile ground on online platforms,
which disrupt societal norms, lowering the barrier of entry to peddlers
of hate by making it anonymous and virtual. The Internet also gives
extremists a platform and amplifies their reach, giving them easy
access to each other and to those who might be radicalized.
All technology and social media companies have a responsibility to
address this hate, through the tools they use, the guidelines they set,
the transparency they offer, their engagement with civil society and
the way they design their platforms.
But we cannot solve the scourge of hate in America simply by fixing
online platforms. First, everyone who has a bully pulpit must speak out
against such hate. We must also look at our education systems, at our
law enforcement capacity and training and at our laws. And we must hold
perpetrators accountable for the harm that they cause online and off.
______
Addendum: Ideological Extremist shooting sprees, 2009-2019
The following is a sampling of white supremacist shooting sprees
which took place between 2009 and 2019 compiled by ADL's Center on
Extremism. More information and statistics about extremist violence of
all ideological backgrounds in the U.S. is available at https://
www.adl.org/education-and-resources/resource-knowledge-base
/adl-heat-map
El Paso, Texas, August 2019. White supremacist Patrick Crusius was
arrested following one of the deadliest white supremacist attacks in
modern U.S. history, a shooting spree at an El Paso Wal-Mart targeting
people of perceived Mexican origin or ancestry that left 22 dead and 24
injured.
Gilroy, California, July 2019. Santino Legan opened fire at the
Gilroy Garlic Festival killing 3 and injuring 15 before being fatally
wounded by police. In an Instagram post, which appears to have been
made by Legan, he asked why towns were overcrowded and open space paved
over to make room for ``hoards [sic] of mestizos and Silicon Valley
white tweets.'' Legan also urged people to read the book Might is
Right, by Ragnar Redbeard. Might is Right, or The Survival of the
Fittest is a book argues in favor of self-interest and the primacy of
the individual. It also attacks Christianity and Judaism, as religions
that weaken people; non-Anglo-Saxons, as lesser races; women, as
greatly inferior beings compared to men; urban-dwellers, as weak
creatures; and the American concept of government based on the notion
that all people are created equal.
Poway, California, April 2019. White supremacist John T. Earnest
allegedly opened fire at a synagogue in Poway, California, killing one
person and injuring three before fleeing. He was reportedly emulating
white supremacist Brenton Tarrant's killing spree in New Zealand in
March 2019. Shortly after Tarrant's spree, Earnest allegedly set fire
to a mosque in Escondido, California, leaving behind graffiti that
referenced Brenton Tarrant's attack. People inside the mosque were able
to put out the fire. Earnest's connection to the Escondido mosque
attack was not known before the Poway attack.
Pittsburgh, Pennsylvania, October 2018. White supremacist Robert
Bowers murdered 11 people and injured seven more, including four police
officers, during services at the Tree of Life Synagogue. Bowers was a
virulent anti-Semite who, among other things, blamed Jews for
orchestrating the immigration of non-whites into the United States.
Parkland, Florida, February 2018. Nikolas Cruz launched a deadly
shooting spree at his former high school, Marjory Stoneman Douglas High
School, killing 17 people and wounding 17 more. According to CNN, Cruz,
19, belonged to a racist Instagram group and hated blacks and Jews,
even claiming Jews wanted to destroy the world. Cruz also allegedly
referred to women who engaged in interracial relationships as
``traitors.'' A South Florida Sun-Sentinel article reported that Cruz
had racist and Nazi symbols on his backpack and that he had etched
swastikas onto ammunition magazines left behind at the school after the
shooting. However, little evidence has so far emerged to suggest that
the MSDHS shooting spree itself was conducted as a white supremacist
attack.
Reston, Virginia, December 2017. Accused white supremacist teen
Nicholas Giampa allegedly shot and killed his girlfriend's parents
after they became upset by his rumored neo-Nazi views. Giampa, was, at
the very least, influenced by Atomwaffen and praised Mason's book,
Siege, a book based on a collection of newsletters written by neo-Nazi
James Mason in the 1980s. Giampa retweeted material from the ``Siege
Culture'' website and at least one Atomwaffen photo. He also admired
someone named ``Ryan Atomwaffen'' for his white supremacist book
collection.
Aztec, New Mexico, December 2017. White supremacist David Atchison
disguised himself as a student in order to conduct a school shooting at
a local high school, where he killed two students before killing
himself.
Mesa, Arizona, March 2015. White supremacist Ryan Elliott Giroux
killed one and injured five others during a shooting spree in Mesa. The
shootings began at a hotel where two people were shot, one fatally.
Giroux then went to a nearby restaurant where he shot a woman and stole
a car. Other shootings occurred as he tried to evade apprehension.
Charleston, South Carolina, June 2015. White supremacist Dylann
Storm Roof conducted a deadly shooting spree at the AME Emanuel Church
in Charleston, killing nine people. Roof deliberately targeted the
church because its parishioners were African-American; he hoped to
incite a ``race war'' that he thought whites would win. Roof had
written a racist and anti-Semitic manifesto prior to carrying out the
attack. Both Federal and state authorities charged Roof in connection
with the massacre; in January 2017, Roof was convicted of the Federal
charges against him and sentenced to death.
Lafayette, Louisiana, July 2015. White supremacist John Russell
Houser killed himself after conducting a vicious shooting spree at a
movie theater in Lafayette, Louisiana, that left two people dead and
nine others injured. Houser, obsessed at the perceived moral decay of
the United States, may have chosen the movie theater as his target
because it was showing the Amy Schumer movie Trainwreck.
Minneapolis, Minnesota, November 2015. Police arrested Allen
``Lance'' Scarsella in November 2015 after Scarsella and others
travelled to a Black Lives Matter protest in north Minneapolis, where
Scarsella opened fire on protesters there, shooting five people, though
none fatally. During his trial in early 2017, prosecutors showed jurors
text messages in which Scarsella had described his intent to kill black
people. Scarsella was convicted of 12 counts of first-degree assault
and one count of riot.
Austin, Texas, November 2014. Larry Steve McQuilliams of Austin,
Texas, a suspected adherent of the racist and anti-Semitic religious
sect known as Christian Identity, launched a shooting attack in
downtown Austin, Texas, firing over 100 rounds of ammunition at targets
including the Austin Police Department, a Federal court house and the
Mexican consulate. According to police reports, McQuilliams had
improvised explosive devices, a map of 34 other targets, including
churches, and a copy of the Christian Identity-related book Vigilantes
of Christendom: The Story of the Phineas Priesthood in his rental van.
McQuilliams died at the scene after an Austin police officer shot him
at long range.
Overland Park, Kansas, April 2014. Long-time Missouri white
supremacist Frazier Glenn Miller launched an attack on Jewish
institutions in the greater Kansas City area, opening fire at two
institutions in a shooting spree that took the lives of three people,
including one child, before police were able to take him into custody.
Miller told police and the media that he launched the attacks ``for the
specific purpose of killing Jews.'' Prosecutors have indicted Miller on
capital murder charges.
Oak Creek, Wisconsin, August 2012. Racist skinhead Wade Michael
Page opened fire at a Sikh temple in Oak Creek, Wisconsin, killing six
people and wounding four others, including a police officer responding
to the shootings. Page killed himself at the scene after being shot by
police. Page was a member of the Hammerskins, a racist skinhead group.
He also played in the white power bands End Apathy and Definite Hate.
Washington, Oregon, and California, September 2011. White
supremacists David Pedersen and Holly Grigsby engaged in a multi-state
killing spree that resulted in four murders in three states. The couple
murdered Pedersen's father and stepmother in Washington, a white man in
Oregon as part of a carjacking, and an African-American male in
California as part of another carjacking. In court, Pederson said he
targeted the Oregon man because he believed he was Jewish and the
Californian man because he was black. After their arrest, the couple
admitted they had been headed to Sacramento to find a prominent Jewish
person to kill.
Washington, D.C., June 2009. White supremacist James von Brunn
attacked the United States Holocaust Memorial Museum in Washington,
D.C., entering the facility and opening fire on security guards inside,
shooting and killing one of them. Two other security guards returned
fire, wounding von Brunn and preventing further deaths. Von Brunn was
arrested and charged with murder. He died of natural causes while
awaiting trial.
Boston, Massachusetts, January 2009. White supremacist Keith Luke
embarked upon a spree of murderous violence against ethnic and
religious minorities in the Boston area in early 2009. He raped and
shot an African immigrant, and shot and killed her sister, who had
tried to help her. Shortly thereafter, he shot and killed a homeless
African immigrant. Although he planned to go to a synagogue that
evening to kill as many Jews as possible, then commit suicide, police
intercepted him before he could do so. Luke fired at police during a
chase before he crashed his vehicle. Police subsequently arrested him
without incident. Luke was convicted of murder in 2013 and killed
himself in prison the following year.
The Chairman. Thank you, Mr. Selim. To Ms. Bickert, Mr.
Pickles and Mr. Slater, on your platforms, how do you define
violent content? How do you define extreme content Ms. Bickert?
Ms. Bickert. Thank you, Mr. Chairman. We will remove any
content that celebrates a violent act, and this is a serious
physical injury or death of another person. We also will remove
any organization that has proclaimed a violent mission or is
engaged in acts of violence. We also don't allow anybody who
has engaged in organized hate to have a presence on the site,
and we remove hate speech. And hate speech we define as an
attack on a person based on his or her characteristics, like
race, religion, sexual orientation, gender. We list them out in
our policies.
The Chairman. Harder to define extreme than violent, is
that correct?
Ms. Bickert. Yes, and we see different people use that word
in different ways. Senator, so what we do is any organization
that has proclaimed violent mission or engaged in documented
acts of violence, we remove them. It doesn't matter what the
reason is for the violence, we just do not allow the violence
period.
The Chairman. Mr. Pickles, what is your platform's
definition of extreme?
Mr. Pickles. So, similar to Facebook. Agree that the word
extremism itself is very subjective. And in some context can be
a positive thing. People who are extremely active on this issue
and itself is not a bad thing. And so we have a three stage
test that defines violent extremist groups, and that test is
that we identify through their stated purpose publications or
actions as extremists, then engage in violence, so they
actually may currently be involved in violence presently, or
they promote violence as a means to further their cause, and
they target civilians.
So we have got that three-stage test of both the ideology
and the violence, because we believe that that framing allows
us to protect speech and to protect debates but also remove
violent extremist from our platform. We then have a broader
framework that prohibits, for example, threats of violence,
call for harm, and wish of harm against people that is much
broader. And again not dependent on ideology.
The Chairman. Mr. Slater, can you add any nuances to----
Mr. Slater. Thank you, Chairman. Broadly similar in that we
ban designated foreign terrorist organizations from using our
platform as well as incitement of violence, glorification of
violence, encouragement to violence, and of course hate speech.
So broadly similar lines.
The Chairman. Now, Mr. Selim has suggested that your three
platforms need to be more transparent. What do you say to that,
Mr. Slater?
Mr. Slater. Thank you Chairman. And I think transparency is
the bedrock of the work we do, particularly around online
content and to try and help people understand both what the
rules are and how we are enforcing them. It is something we
need to continue to get better on. Look forward to working with
this Committee, and Mr. Selim and others on that. We have in
the last year on YouTube provided our YouTube community
guidelines enforcement report, where you can go and see how
many videos we have removed in a quarter, for what reasons,
which were flagged by machines versus users, and we break that
down by violent extremism, hate speech, child safety, and other
issues. So I think this is a really key issue and we look
forward to continuing to improve.
The Chairman. Mr. Selim before I ask Ms. Bickert and Mr.
Pickles to respond, perhaps you could help them understand how
you frankly don't believe they are quite transparent enough at
this point.
Mr. Selim. Mr. Chairman, thank you for your question. To be
clear, the point I am making on transparency is to make sure
that there are more clearly delineated categories between the
point that Mr. Slater was making in terms of what the machines
or the algorithms use to remove certain types of content or
stop it from going up in the first place and what users on any
of these platforms go on to say, like we think this is a
violation of the terms of service.
There are degrees of inconsistencies across these platforms
that are at the table as well as others. And so to get a
holistic picture of what a certain issue may be while
individuals may flag versus what some algorithms pull down,
there are different consistencies in that. And so when we are
asking for transparency, we are really looking for a much more
balanced approach in that across all the platforms.
The Chairman. So, Mr. Pickles, is he touching on something
that has a point?
Mr. Pickles. Yes. Absolutely. I think the balance between
particularly for companies who are investing in technology
understanding what came down because a person saw it and
reported it versus did the content come down because technology
found it is very important. We have now published a breakdown
of six policy areas and the number of user reports we receive.
It is about 11 million reports every year, but 40 percent of
the content that we remove, we removed because technology found
it, not because of user reports.
The Chairman. 40 percent?
Mr. Pickles. Yes. So telling that story in a meaningful way
is absolutely a challenge and one that we are certainly
investing in.
The Chairman. What is that percentage in Facebook, Ms.
Bickert?
Ms. Bickert. Mr. Chairman, when it comes to violent content
and terror content, more than 99 percent of what we remove is
flagged by our technical tools, and we have had a productive--
--
The Chairman. The artificial intelligence?
Ms. Bickert. Some of it is artificial intelligence, some of
it is image matching. So known videos where we use a software
to reduce that to basically a digital fingerprint, and we are
able to stop uploads of that video again. And we have worked
with the ADL for years on this, and I think transparency is
key. I think we would all agree. We, for the past year and a
half, have published not only our detailed implementation
guidelines for exactly how we define hate speech and violence,
but also reports on exactly how much we are removing in each
category, and how much of that like, Mr. Pickles said, how much
of it is actually flagged by our technical tools before we get
user reports.
The Chairman. Thank you very much. Senator Cantwell.
Senator Cantwell. Thank you, Mr. Chairman. Mr. Selim, I
think you mentioned 8chan, but what do you think we need to do
to monitor incitement on 8chan and other dark websites?
Mr. Selim. So I think you can really approach this issue
from two categories. There are a number of increased measures,
some of which I noted in my written statement submitted to this
Committee, that these companies as well as others can take to
create a greater degree of transparency and standards so that
we can have a really accurate measure of the types of hatred
and bigotry that exists in the online environment writ at
large. As a result of that increased or better data, we can
make better policies that apply to content moderation, terms of
service, et cetera. So I think really having the good data is a
framework for better policies and better applications and
content moderation programs.
Senator Cantwell. So you are saying there is more that they
can do? Social media companies, there is more that they can do?
Mr. Selim. Yes, ma'am. There is much more that they can do.
Senator Cantwell. I look in your statement, you include
auditing and you know third-party evaluation for that
transparency as well as you know responsibility, but as I
mentioned in my opening statement, basically then drive all of
this to a dark web that we have less access to. I am going to
get to them and ask them a question, but what more do you think
we should be doing together to address the hate that is taking
place on these darker websites too?
Mr. Selim. So a number of measures. I mean, the first is
having our public policy be very starting from place where we
are victim focused. We know that whether it is Pittsburgh,
Poway, El Paso, or any of the number cities that other
panelists and members of this committee have mentioned in their
statements, we need to start to make measures that combat
extremism or domestic terrorism be from preventing other such
horrific tragedies. And in order to do that we really need to
start from a place that prevents and has a better accounting of
hate crimes, bias-motivated crimes, hate related incidents,
etc.
And when we start from that place, I think we can make
better policy and better programs at the Federal Government,
and State and local, and also in the private industry levels as
well.
Senator Cantwell. Well, one of the reasons I am definitely
going to be, you know, calling on the Department of Justice to
ask what more we can do in this coordination is several years
ago Interpol, Microsoft, the others worked on trying to address
on an international basis child pornography to better skill law
enforcement at policing crime scenes online. And I would assume
that the representatives today would be supportive, maybe
helpful, maybe even financially helpful in trying to address
these crimes as they exist today as hate crimes on the dark
side of the web. Is that--do I have any responses from our tech
companies here?
Ms. Bickert. Thank you, Senator Cantwell. This is something
that across the industry we have been working on for the past
few years in a manner very similar to how the industry came
together against child exploitation online. We launched the
global GIFCT, the Global Internet Forum to Counter Terrorism,
which both of my colleagues referred to as a way of getting
industry to create sort of a no-go zone for this terrorist and
violent content.
As part of that, we trained hundreds of smaller companies
on best practices and we make technology available to them. The
reality is for the bigger companies, we often are able to build
technical tools that will stop videos at the time of upload. It
is much harder for smaller companies, which is why we provide
technology to them. We now have 14 companies that are involved
in a hash sharing consortium so that we can help even these
small companies stop terrorist content at the time of upload.
Senator Cantwell. Well, I appreciate, and I agree with Mr.
Selim. There is more that you can do on your own sites. But
setting that aside for a minute, what do you think we should do
about 8chan and the dark websites? What are what do you all
think we should do?
Ms. Bickert. I can tell you what we do on Facebook,
Senator, which is we ban any link that connects to 8chan pol
where these manifestos have appeared. So those manifestos with
the El Paso shooting, with Poway, were not available through
Facebook.
Senator Cantwell. I am saying what more do you think in
Government and law enforcement working together, besides what
you do to address this, anybody else? Mr. Pickles?
Mr. Slater. Well, I think, to follow up on Mr. Selim's
point, I think certainly if this criminal activity is happening
on these platforms then a law enforcement response is primary.
As I say, add the tools we have in our toolbox related to
content and if people are promoting violence against
individuals, that is criminal offenses, a law enforcement
intervention at that point is something I think should be
looked at. And I think if we can strengthen this industry, our
cooperation with law enforcement, we can make sure that the
information sharing is a strong as it needs to be to support
those interventions.
Senator Cantwell. So, you think we need more law
enforcement resources addressing this issue?
Mr. Slater. I think it is a question of both resources and
I think again to follow Mr. Selim's point, there was a paper
from George Washington University last week looking at the
statutory framework around some of these spaces and if there
are opportunities to strengthen them? And in many of the areas
Mr. Selim mentioned, and again, I think that is a worthwhile
public health policy conversation to have.
Senator Cantwell. I definitely believe you need more law
enforcement resources on this issue, and I look at what
progress we made with Interpol and the tech industry fighting
on other issues. I think this is something, and I hear that
from Mr. Selim, more resources. So, thank you all very much.
Mr. Chairman.
The Chairman. Thank you, Senator Cantwell.
Senator Fischer.
STATEMENT OF HON. DEB FISCHER,
U.S. SENATOR FROM NEBRASKA
Senator Fischer. Thank you, Mr. Chairman. In June, Senator
Thune held a subcommittee hearing on persuasive design, and as
we discussed, Facebook, Twitter, and YouTube are engineered to
track, capture, and keep our attention, whether it is through
predictions of the next video to keep us watching or what
content to push, to the top of our news feeds. I think we have
to realize that when social media platforms fail to block
extremist content online, this content doesn't just slip
through the cracks, it is amplified, and it is amplified to a
wider audience.
And we saw those effects during the Christchurch shooting.
The New Zealand's terrorists Facebook live broadcast was up for
an hour, that was confirmed by The Wall Street Journal, before
it was removed, and it gained thousands of views during that
timeframe. Ms. Bickert, how do you concentrate on the increased
risk from how your algorithms boost content while gaps still
exists in getting dangerous content off the platform? You
touched on that a little bit in your response to Senator
Wicker, but how are you targeting solutions to address that
specific tension that we see?
Ms. Bickert. Senator, thank you for the question. It is a
real area of focus, and there are three things that we are
doing. Probably the most significant is technological
improvements, which I will come back to in a second. Second is
making sure that we are staffed to very quickly review reports
that come in. So the Christchurch video, once that was reported
to us by law enforcement, we were able to remove it within
minutes. That response time is critical to stopping the
virality you mentioned.
And finally, partnerships. We have hundreds of safety and
civil society organizations that we partner with. So if they
are seeing something, they can flag it for us through a special
channel. Now, going back to the technology briefly, with the
horrific Christchurch video, one of the challenges for us was
that our artificial intelligence tools did not spot violence in
the video. What we are doing going forward is working with law
enforcement agencies, including in the U.S. and the UK, to try
to gather videos that could be helpful training data for our
technical tools, and that is just one of the many efforts.
We have to try to improve these machine learning
technologies so that we can stop the next viral video at the
time of upload or the time of creation.
Senator Fischer. When you talk about working with law
enforcement, you said law enforcement contacted you, is that
reciprocal? Do you see something show up and then you in turn
try to get it to law enforcement as soon as possible so that
individuals can be identified? What is the working relationship
there?
Ms. Bickert. Absolutely. Senator. We have a team that is
our law enforcement outreach team. Anytime that we identify a
credible threat of imminent harm, we will reach out proactively
to law enforcement agencies. And we do that regularly. Also
when there is some sort of mass violence incident, we reach out
to them, even if we have no indication that our service is
involved at all, we want to make sure the lines of
communication are open. They know how to submit emergency
process to us. We respond around the clock in a very timely
fashion because we know that every minute is critical in this
type of situation. I am a former prosecutor myself and so these
things are very personal to me.
Senator Fischer. I know that the platforms that are
represented here today, you have increased your efforts to take
down this harmful content, but as we know there are still
shortfalls that exist in order to get that response made in not
just a timely manner but one that is really going to truly have
an effect. Mr. Slater, when it comes to liability, do media
platforms--you guys need more skin in the game so that you can
ensure better accountability and be able to incentivize some
kind of timely solution?
Mr. Pickles. Thank you, Senator, for the question. I think
if you look at the practices that we are all investing in,
certainly looking from our perspective, and the way we are
getting better over time. The current legal framework strikes a
reasonable balance.
In particular, it both provides protection from liability
that would go too far that would be overbroad but also acts as
a sword not just a shield, empowering us and giving us the
legal certainty that we need to invest in these technologies,
the people to monitor or detect, review, and remove this sort
of violative content. That way the legal framework continues to
work well.
Senator Fischer. Mr. Selim, can you comment on this as
well? Do you think there is enough legal motivation for social
media platforms to prioritize some kind of solutions out there?
I mean, that is what this hearing is about to find the
solutions so that we can curb that online hate that I think
continues to grow.
Mr. Selim. When thinking through the issues of content
moderation, the authorities that exist within the current legal
frameworks that reside within the companies represented at this
table is sufficient for them to take actions on issues of
content moderation, transparency reporting, etc. So there
certainly is a degree of legal authorities that affords these
companies as well as others the opportunity to take any number
of measures.
Senator Fischer. Ms. Bickert, in your testimony you say
that Facebook live will ban a user for 30 days for first-time
violation of its platform policies. Is that enough? Can users
be banned permanently? Would that be something to look at?
Ms. Bickert. Senator, thank you for the question. One
serious violation will lead to a temporary removal of the
ability to use live. However, if we see repeated serious
violations, we simply take that person's accounts away, and
that is something that we do across the board not just with
hate and inciting content, but other content as well.
Senator Fischer. Thank you.
The Chairman. Thank you so much, Senator Fisher.
Senator Blumenthal.
STATEMENT OF HON. RICHARD BLUMENTHAL,
U.S. SENATOR FROM CONNECTICUT
Senator Blumenthal. Thank you, Mr. Chairman. Thank you all
for being here today and thank you for outlining the increased
attention and intensity of effort that you are providing to
this very profoundly significant area. I welcome that you are
doing more and trying to do it better, but I would suggest that
even more needs to be done and it needs to be better, and you
have the resources and technological capabilities to do more
and better.
And just to take the question that Senator Fischer asked of
you, Mr. Selim, about incentives. Your answer was that they
have authority to provide them with opportunities. The question
is, really don't they need more incentives to do more and do it
better, to prevent this kind of mass violence that may be
spurred by hate speech appearing on the site or in fact may
actually be a signal of violence to come?
And I just want to highlight that 80 percent of all
perpetrators of mass violence provide clear signals and signs
that they are about to kill people. That is the reason that
Senator Graham and I have a bipartisan measure to provide
incentives to more states to adopt extreme risk protection
order laws that will, in fact, give law enforcement the
information they need to take guns away from people who are
dangerous to themselves or others.
And that information is so critically important to prevent
mass violence, but also suicides, domestic violence, and the
keys and information and signals often appear on the internet.
In fact just this past December in Monroe, Washington a clearly
troubled young man made a series of anti-Semitic rants and
violent posts online. He bragged about planning to ``shoot up
an expletive school'' in a video while armed with an AR-15
style weapon, and on Facebook posted that he was ``shooting for
30 Jews.''
Fortunately, the ADL saw that post, it went to the FBI, and
the ADL's vigilance prevented another Parkland or Tree of Life
attack. Fred Gutenberg of Coral Springs, Florida met with me
yesterday, told me about a similar incident involving a young
man in Coral Springs who said he was about to shoot up the high
school there, and law enforcement was able to foresaw it using
an extreme risk protection order statute.
So my question is to Facebook, Twitter, and Google, what
more can you do to make sure that these kinds of signs and
signals involving references to guns, it may not be hate
speech, but it is references to possible violence with guns or
use of guns, to make that available to law enforcement? Ms.
Bickert, and Mr. Pickles, and Mr. Slater.
Ms. Bickert. Thank you, Senator Blumenthal. One of the
biggest things we can do is engage with law enforcement to find
out what is working in our relationship and what isn't, and
that is the dialogue that over the past years has led to us
establishing a portal through which they can electronically
submit request for content with legal process and we can
respond very quickly----
Senator Blumenthal. But what are you doing proactively? And
I apologize for interrupting, but my time is limited.
Proactively, what are you doing with the technology you have to
identify the signs and signals that somebody is about to use a
gun in a dangerous way? That someone is dangerous to himself or
others and is about to use a gun?
Ms. Bickert. Senator, we are now using technology to try to
identify any of those early signs, including gun violence, but
also suicide or self-injury.
Senator Blumenthal. Do you report it to law enforcement?
Ms. Bickert. We do. In 2018, we referred a number of many
cases of suicide or self-injury, but we detected them using
artificial intelligence to law enforcement so that they were
able to then intervene, and in many cases, save lives.
Mr. Pickles. We have is a very similar approach where we
have a credible threat that something, someone is at risk to
others or themselves. We work with the FBI to ensure they have
the information they need.
Senator Blumenthal. Mr. Slater?
Mr. Slater. Thank you, Senator. Similarly, when we have a
good faith belief of a credible threat, then we will
proactively refer to the Northern California Regional
Intelligence Center who will then fan that out to the right
authorities.
Senator Blumenthal. Because my time has expired, I am going
to ask each of you if you would please give me more details in
writing as a follow up for how you--what identification signs
you use, what kind of technology, and how you think it can be
improved assuming that the Congress approves, as I hope it
will, the emergency risk protection order statute to provide
incentives to more than just the 18 states that have them now,
but others to do the same. Thank you.
The Chairman. Thank you so much, Senator Blumenthal.
Senator Thune.
STATEMENT OF HON. JOHN THUNE,
U.S. SENATOR FROM SOUTH DAKOTA
Senator Thune. Thank you, Mr. Chairman. And thank all of
you for being here today. Your participation in this hearing is
appreciated as this Committee continues its oversight of the
difficult tasks each of your companies face preserving an
openness on your platforms while seeking to responsibly manage
and thwart the actions of those who use your services to spread
extremist and violent content. Last Congress, we held a hearing
looking at terrorist recruitment propaganda online.
We discussed the cross sharing of information between
Facebook, Microsoft, Twitter, and YouTube which allowed each of
those companies to identify potential extremism faster and more
efficiently. So I would just direct this question and ask that
how effective is that shared database of hashes been?
Ms. Bickert. Senator, thank you for the question. Through
the shared data base, we now have more than 200,000 distinct
hashes of terror propaganda, and that has allowed--I can speak
for Facebook only, but that has allowed us at Facebook to
remove a lot more than we otherwise would have been able to do.
Mr. Pickles. I would just add, since that hearing actually,
I think the reassuring thing is that we don't just share hashes
now. We have grown that partnership, so we share URLs. So if we
see a link to a piece of content like a manifesto, we are able
to share that across industry. And furthermore, I think an area
that after Christchurch we recognize we need to improve, we now
have real-time communications in a crisis.
So industry can talk to each other in real time,
operationally to say, even you know, not content related but
situational awareness, that partnership between industry now
also involves law enforcement. That wasn't there when I think
we had that hearing last, and so I think it not just about the
hash program but broadening our new programs that are
developing that work further.
Mr. Slater. Yes, I think broadly, I would say look at how
we have been improving over time. Surely systems are not
perfect. We are always going to have to evolve to deal with bad
actors, but I think on the whole, we are doing a better job in
part because of this technology sharing, this information
sharing, in removing the sort of content before it has wide
exposure of any sort or is viewed widely.
Mr. Selim. Senator, I would only add that the threat
environment that we are in today as a country has changed and
evolved in the past 24 to 36 months. And likewise, the tactics
and techniques that these platforms as well as others use to
evolve, the evolving nature of the terrorist landscape online,
whether it be foreign or domestic, needs to keep pace with the
threat environment that we are in today.
Senator Thune. And so just as a follow-up, are there
similar partnerships among your companies as well as the
smaller platforms to specifically identify mass violence?
Ms. Bickert. Senator, one of the things that we have done
over time is expand the mandate of the Global Internet Forum to
Counter Terrorism. So we relatively recently expanded to
include mass violent incidents, and we are now sharing both
through our crisis incident protocol and our hash sharing, we
are sharing a broader variety of violent incidents.
Senator Thune. Mr. Slater, YouTube's, I should say,
automated recommendation systems comes under criticism for
potentially steering users toward increasingly violent content,
and earlier this year, I led a subcommittee hearing examining
the use of persuasive technologies on Internet platforms,
algorithm transparency, and algorithmic content selection. I
asked the witness that Google provided at that time for that
hearing several specific questions for the record about YouTube
that were not thoroughly answered, and I would just say that
providing complete answers to questions members submit for the
record is essential as we look to work together as partners to
combat many of the issues discussed here today.
So I would like your commitment to provide thorough
responses to any questions you might get for the record. Do I
have that?
Mr. Slater. Surely, Senator, to the best of our ability.
Senator Thune. OK. In addition, I would like to just
explore the nexus between persuasive technologies and today's
topic, specifically what percentage of YouTube video views are
the result of YouTube automatically suggesting or playing
another video after the user finishes watching the video?
Mr. Slater. So I do not have a specific statistic there,
but I can say the purpose of our watch next, our recommendation
system, is to show people videos they may like that are similar
to what they have watched before. At the same time we do
recognize this concern about recommendations for borderline
content that is content that maybe is not removed but brushes
right up against those lines. And we have introduced changes
this year to reduce recommendations for those sort of
borderline videos.
Senator Thune. Could you get the number? And I assume you
have that somewhere. That has got to be available and furnished
for the record, but so the question again is to ask you
specifically what is YouTube doing to address the risk that
some of these features which as, you note, are pointing a user
in the direction of increasingly violent content?
Mr. Slater. Yes, and that change we made in January to
reduce recommendations has been key. And it is still in its
early days, but it is working. We have reduced the views from
those recommendations for that borderline content by 50 percent
just since January. As those systems get better, we hope that
that will improve and happy to discuss it further.
Senator Thune. Thank you. Thank you, Mr. Chairman.
The Chairman. Thank you, Senator Thune. Now based on
presence at the gavel, we next have Senator Blackburn followed
by Senator Scott.
Senator Blackburn.
STATEMENT OF HON. MARSHA BLACKBURN,
U.S. SENATOR FROM TENNESSEE
Senator Blackburn. Thank you, Mr. Chairman, and I want to
thank each of you for being here this morning and for talking
with us. This Committee has looked at this issue on the
algorithms and their utilization for some time and we are going
to continue to do this. Looking at content and the extremists
content that is online is certainly important. We know there
are a host of solutions that are out there, and we need to come
to an agreement and an understanding of how you are going to
use these technologies to really protect our citizens.
And social media companies are in a sense open public
forums, and they should be where people can interact with one
another. And part of your responsibility in this vein is to
have an objective cop on the beat and be able to see what is
happening because you are looking at it in real time. But what
has unfortunately happened many times is you don't get an
objective view, you don't get a consistent view, you get a
subjective view. And this is problematic, and it leads to
confusion by the public that is using the virtual space for
entertainment, for their transactional life, for obtaining
their news.
So indeed as we look at this issue, we are looking for you
to approach it in a consistent and objective manner. And we
welcome the opportunity to visit with you today. Ms. Bickert, I
have got a couple of things that I wanted to talk with you
about. We have all heard about these third-party facilities
where contractors are working long hours and they are looking
at grotesque and violent images, and they are doing this day in
and day out. So talk a little bit about how you transition from
that to using modern technologies.
What Facebook is going to do in order to capture this, to
extract it and to minimize harm. You have talked about you have
got 30,000 employees that are working on safety and security,
and then there are third-party entities that are working on
this. So let's talk about that impact on the individuals and
then talk about the use of technologies to speed up this
process and to make it more consistent and accurate.
Ms. Bickert. Thank you for the question, Senator. Making
sure that we are enforcing our policies is a priority for us,
making sure that our content reviewers are healthy and safe in
their jobs is paramount. And so one of the things that we do is
we make sure that we are using technology to make their jobs
easier and to limit the amount of content, types of content
that they have to see. I will give you a couple examples with
child exploitation videos, with graphic violence, with terror
propaganda. We are now able to use technology to review a lot
of that content so that people don't have to. And in situations
where----
Senator Blackburn. Let me ask you this, I am sorry to
interrupt, but we need to move forward, your 30,000 reviewers,
are they all located in Palo Alto or are they scattered around
the country, or around the globe?
Ms. Bickert. No Senator, the more than 50--we have 30,000
people working in safety and security. Some of them are
engineers or lawyers. The content reviewers, we have more than
15,000. They are based around the world.
Senator Blackburn. OK. Yes, great.
Ms. Bickert. And for any of them, not only are we using
technology, and there are ways that we are using even where we
cannot make a decision on the content using technology alone,
there are things we can do like removing the volume or
separating a video into still frames, that can make the
experience better for the reviewer.
Senator Blackburn. OK. Now, let me ask you about this. Mark
Zuckerberg in a Washington Post op-ed had called for us to
regulate, to define ``lawful but awful'' speech. So tell me how
you think you could define, or we could define lawful but awful
speech but not overreach or infringe on somebody's First
Amendment, free speech rights?
Ms. Bickert. Senator, one of the things that we are looking
to with our dialogue with Government is clarity on the actions
that Government wants us to take. So we have our set of
policies that lays out very clearly how we define things, but
we don't do that in a vacuum. We do that with a lot of input
from civil society organizations and academics around the world
but we also like to hear the views from governments so we can
make sure we are mindful of all of the different safety----
Senator Blackburn. No, ours are constitutionally based. I
am out of time. Mr. Pickles, I am going to submit a question to
you for the record. Mr. Selim, I have got one that I am going
to send to you. Mr. Slater, I always have questions for Google,
so you can depend on me to get one to you and we do hope that
you all are addressing your prioritization issues also. With
that, Mr. Chairman, I yield back.
The Chairman. Thank you very much.
Senator Scott.
STATEMENT OF HON. RICK SCOTT,
U.S. SENATOR FROM FLORIDA
Senator Scott. Thank you for being here today. I am glad we
are having a meaningful conversation about what is happening in
our Nation. It is time we face the fact that our culture has
produced an underclass of predominantly white young men who
place no value on human life. These individuals live
purposeless lives of anonymity and digital dependency, and
increasingly act on their most evil desires, sometimes with
racial hatred. As you all know, while I was Governor, we had
the horrible shooting at the school in Parkland.
Within three weeks we passed historic legislation,
including the risk protection orders that Senator Blumenthal
was talking about. We did it by sitting down with law
enforcement, mental health counselors, and educators to come up
with the right solution. Now with regard to the shooting at
Parkland, the killer, Nicholas Cruz, had a long, long history
of violent behavior. In September 2017, the FBI learned that
someone with the username Nicholas Cruz had posted a comment on
a YouTube video that said, ``I am going to be a professional
school shooter.''
And Nicholas Cruz made other threatening comments on
various platforms. The individual whose video Nicholas Cruz
posted this comment on reported it to the FBI. Unfortunately,
the FBI closed the investigation after 16 days without ever
contacting Nicholas Cruz. The FBI claimed they were unable to
identify the person who made the comment. Unfortunately, we now
have 17 innocent lives that were lost because of Nicholas Cruz.
My question is to Mr. Slater: How was it a platform like
YouTube which is owned by Google not able to track down the IP
address and identity of the person who made that comment? When
did YouTube remove the comment? Did YouTube report this comment
to law enforcement? If so, who and when? If you did report this
comment to law enforcement, did you follow-up? What was the
process, and was there any follow up to see if there was any
corrective action?
Mr. Slater. Senator, thank you for the question. First, it
was a horrendous event. And you know, we strive to be vigilant,
to invest heavily, to proactively report where we see an
imminent threat. I don't have the details on the specific facts
you are describing. I will be happy to get back to you, but let
me say this going forward, looking ahead, Parkland was a moment
that did spur us to proactively reach out to law enforcement to
start talking about, how can we do this better?
And that is part of how we then reached out and started
working more closely with the Northern California Regional
Intelligence Center to make sure that when we did have these
good faith beliefs, we could go to a one-stop shop who could
get it to the right law enforcement, locally rather than us
trying to call the right people. And this is something we are
just this month in fact, or in the last month, there was an
incident where PBS was streaming the NewsHour on YouTube,
somebody put a threat in the live chat.
We refer that to the Regional Intelligence Center, and they
refer it to the Orlando Police who then took the person into
custody appropriately. And this was reported in the news. So
that is not to say things are perfect. We always have to strive
to get better and I look forward to working with you and law
enforcement on that. But I do think that we continue to improve
over time.
Senator Scott. So with regard to Nicholas Cruz, you will
give me the information of, you know, who did you contact, when
did you contact, when was it taken down? So to this day I
cannot get an answer on what anybody did with regard to this
shooter. What YouTube did, what the FBI did, nobody wants to
talk about it, which is fascinating to me. So if you give me
that information.
And then second, are you comfortable that if another
Nicholas Cruz put something up, you have the process now that
you will contact somebody and there will be a follow-up
process?
Mr. Slater. Senator, I think our processes are getting
better all the time. They are robust. I think this is an area
where it is an evolving challenge, both because technology
evolves, because people's tactics evolve. They might use code
words, and so on, but I would be happy to follow up with the
team and get more information on how those practices operate
and how we continue to work together.
Senator Scott. Thank you. Mr. Pickles, how can Nicolas
Maduro, who is committing genocide against his citizens, who is
withholding clean water, food, and medicine still have a
Twitter account with 3.7 million followers?
Mr. Pickles. Well, you rightly highlight that the behavior
that is being taken there is abhorrent and the question for us,
as a public company that provides a public space for dialogue
is, is someone breaking our rules on our service? We recognize
that there are situations where there are geopolitical
circumstances where there are world leaders who have Twitter
accounts in countries where Twitter has blocked, where there is
no free speech, and so we do take a view that and we hope that
the dialogue that that person being on the platform starts,
helps contribute to solving the challenges that you have
outlined.
Senator Scott. But he has been doing it for a long time and
it is not getting better in Venezuela, it is getting worse.
Mr. Pickles. And I think this is a good illustration of how
the role technology companies along with other parts of public
policy responses. And If we remove that person's account, it
would not change the facts on the ground. And so we need to
bear in mind how did the other levers come into play.
Senator Scott. I completely disagree. Maduro sits there and
talks about things and continues to act like he is a world
leader, and he is a pariah. And it sure seems to me that what
you are doing is allowing him to continue to do that.
Mr. Pickles. Well, as I said, his current account has not
broken all the rules. Were he to break all rules, he would be
treated the same as every other user, and we would take action
when necessary.
Senator Cantwell. Mr. Chairman, I know that we have votes
already starting and you are trying to get other people. I
would be happy to work with the Senator from Florida on this
issue. I do think that we are not doing enough, and I think
this specific case I mentioned in my opening statement about
the Rohingya and what happened on Facebook is another example,
so happy to work with you on this issue.
The Chairman. Well, yes, and thank you, Senator Cantwell,
and thank you Senator Scott for raising this. I am told there
is a vote on, and I am shocked to hear that they are going to
leave it open till 11:30 a.m., which is generally what happens.
Senator Duckworth.
STATEMENT OF HON. TAMMY DUCKWORTH,
U.S. SENATOR FROM ILLINOIS
Senator Duckworth. Thank you, Mr. Chairman. While I do
appreciate this Committee's consideration of issues at the
intersection of extremism and social media, many I think would
agree that today's hearing is another data point on a long
history of congressional hand-wringing on gun violence.
According to the gun violence archive, since 2019 began,
260 days ago, we have witnessed 318 mass shootings in the U.S.,
more than one per day. Mass shootings are those in which at
least four people are shot, excluding the shooter. After 20
children, 6 adults, and a shooter lost their lives at Sandy
Hook Elementary School in 2012, many elected officials
including myself declared an end to Congressional inaction. No
more we said, but since that day, our Nation has endured 2,226
mass shootings. Think about that number for a minute. But here
we are not focused on ways to stop gun violence, but rather the
scourge of social media.
I am not going to say that there is no connection but every
other country on the planet has social media, video games,
online harassment, hate groups, crime, and mental health
issues, but they do not have mass shootings like we do. Nothing
highlights the absurdity of Congress's inability to solve the
gun violence crisis than seeing 318 mass shootings in 260 days,
and then holding our hearings on extremism and social media.
Ms. Bickert and Mr. Pickles, this is a chart from the Digital
Marketing Institute that according to their website highlights
the average number of hours that social media users spend on
platforms like Facebook and Twitter.
As you will see, the United States and our users are
relatively middle of the pack when it comes to time spent
online. My question to you both is this, do you agree that
Americans use of social media is not especially unique on a per
capita basis? In other words, are you aware of specific trends
on your platforms to explain the amount of gun violence in the
United States?
The Chairman. Senator Duckworth, and this will not come out
of your time, do sort of explain to us, because some of us
cannot see the detail.
Senator Duckworth. Sure, this is how much time average
number of hours that social media users spend using social
media each day via any device.
The Chairman. And the arrow points to the United States?
Senator Duckworth. To the United States. The highest is the
Philippines. The lowest is Japan. The U.S. is right in the
middle. So American users and I have got a four and a half year
old and I have an 18 month old and when I get home says iPhone,
iPhone and she is on it. She knows how to select YouTube kids
on my phone, and she knows how to go right to what she wants to
watch. OK, so I am just as concerned that the United States in
terms of social media usage, which you both agree, is somewhere
in the middle of the pack compared to the rest of the world.
Ms. Bickert. Yes, Senator, according to the study which I
am not more familiar with, yes.
Senator Duckworth. In other words, are you aware, are
either of you aware of specific trends on your platforms to
explain the amount of gun violence in the United States?
Mr. Pickles. No, I think your study reflects our view,
about 80 percent of our users are outside the United States.
And so I think you are right. The image speaks for itself.
Senator Duckworth. Thank you. Mr. Selim, you brought up the
role that video games can play on online hate and harassment. I
agree with you that any dissemination of hate must be addressed
regardless of the platform used. But if a meaningful connection
between video games and gun violence exists, you think that the
widespread use of video games in Japan and South Korea would
reflect that connection, correct? If you look at this chart, I
think there is something to be said for the availability of
guns in the U.S.
If you look into the amount of time that the folks in Japan
and South Korea spend on video games is far greater than
anywhere else. We are third, and yet if you look at the number
of incidents of gun violence and gun deaths per every 100,000
people in 2017, here is the U.S., but we are not the biggest
users of video games. Would this be accurate?
Mr. Selim. Senator, thank you for your question. I have not
read this specific study, but I do have one data point, if I
may share with you for just a moment, according to an ADL
report looking at extremists related murders and homicides over
the past decade, our research shows that 73 percent of
extremist related murders and homicides were in fact committed
with firearms. So to the extent that you are making the point
that extremists with weapons results in violence and homicide,
we have the data that backs that point up.
Senator Duckworth. Thank you. As we are reminded daily, the
world is full of individuals who use social media platforms to
disparage others, cast false equivalencies, and question facts.
Some will use the unanimity of online platforms to spread hate
but our use of social media, video games, and other variables
does little to explain the 2,226 mass shooting since Sandy
Hook. The Internet has emboldened and empowered hate by
allowing individuals to develop online communities and share
their warped ideas, but it is our weak gun laws here in the
U.S. that allows that hate to become lethal. There is a clear
and undeniable connection between the number of guns in the
United States and the number of gun deaths in our community.
Look at this platform. This is the number of guns per 100
people and this is the number of gun related deaths per 100,000
people. We are up here. Here is the rest of the world. Some of
whom use more social media than we do. Some of whom actually
engage in more video games than we do. We are saturated in
weaponry that was designed for war but is made available to
nearly anyone who attends a local gun show.
A Dayton shooter has hundred round drum. I didn't have a
hundred round drum when I served in Iraq. We did not send
Marines into Fallujah with hundred round drums, but yet you can
buy them at gun shows. Look, 90 percent of Americans that agree
that Congress should expand background checks and red flag
laws. 60 percent of Americans agree that banning high-capacity
ammunition clips is what we need to do.
This is not controversial. It is well past time that Leader
McConnell brings HRA to the House and passed bipartisan
background checks back to the Senate floor for a vote. I hope
Leader McConnell will also allow votes on to keep American Safe
Act, the Extreme Risk Protection Act, the Disarm Hate Act, and
the Domestic Terrorism Prevention Act. Each of these bills will
keep our children and our neighbors safer. I hope my Republican
colleagues will join in these bipartisan efforts. Thank you and
I yield back.
The Chairman. Senator Duckworth, let's do this so we can
have a complete record. If you would reduce those three posters
to a size that we can copy, and they will be admitted in the
record at this point in the hearing without objection.
[The information referred to follows:]
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
Senator Duckworth. Thank you very much, Mr. Chairman, that
is generous of you.
The Chairman. Senator Young.
STATEMENT OF HON. TODD YOUNG,
U.S. SENATOR FROM INDIANA
Senator Young. Thank you, Mr. Chairman. I want to thank all
of our panelists for being here today. I really do appreciate
your testimony and your answering our questions. Look, we all
need to collaborate in curbing online extremism, which I
understand to be one of multiple causes that we could cite as
we all think about the issue of mass casualty events and
extremist events, or generally. The Nation is wrestling with
mass violence extremism and issues of responsibility, digital
responsibility, for some of these events.
In fact, in my home state of Indiana, Hoosiers and Crown
Point, Indiana recently experienced firsthand how a person can
become radicalized over the internet, something I know that
many of your companies have studied and are working on. In
2016, a Crown Point man was arrested and convicted for planning
a terrorist attack after becoming radicalized by ISIS over the
internet. Thankfully the FBI in the Indianapolis Joint
Terrorism Task Force intervened before any violent attack
occurred. However, that is not always the case as we know, and
we have seen this across the country.
And that is why it is critically important that we have
this hearing, that we continue to work together
collaboratively, knowing that your products and platforms
provide incredible value to consumers and they obviously were
not intended for this purpose. So it is our responsibility in
Congress, it is definitely your responsibility as business
people, to make sure that we monitor how the great value that
you provide can be used in an illicit, improper, dangerous, and
nefarious manner.
In one minute or less because I have three minutes less
left, I would request that the representatives from Google and
Facebook and Twitter tell us why Americans should be confident
that each of your companies are taking this issue seriously,
and why Americans should be optimistic about your efforts going
forward?
The Chairman. One minute each?
Senator Young. Yes, indeed.
Mr. Slater. Thank you, Senator. I would start by pointing
to YouTube community guidelines enforcement report, which
details every quarter videos we have removed, the reasons why,
and indeed how much is being flagged first by machines in
dealing with this issue, removing violative content, as a
combination of technology and people. Technology can get better
and better at identifying patterns.
People can help deal with the right nuances and we have
seen over time that the technology is getting better and better
at taking down the content faster and before people have viewed
it. As I've said at the outset, of the 9 million videos that we
removed in the second quarter of this year, 87 percent of those
were first flagged by our machines, and 80 percent of those
were removed before a single view. When we talk about violent
extremism which it is generally better in terms of removable
before wide viewing.
So, you know, we are already seeing advancements in machine
learning not just in this area but across the industry broadly,
and the thing about machine learning is as it is fed more data,
as it learns from mistakes, as we say, you got to learn here.
Those systems will get better, and so why one should be
optimistic if those systems ideally will continue to get
better. Will they be perfect? No, bad actors will continue to
evolve, but I do think there is room for optimism, and I think
there is reason for optimism based on the collaboration between
all of us today.
Senator Young. Thank you. Facebook.
Ms. Bickert. Thank you, Senator. The first thing I will say
is Facebook will not work as a service if it is not a safe
place, and this is something that we are keenly aware of every
day. If we want people to come together to build this
community, they have to know they are safe. And so the
incentives are there for us to make sure we are doing our part.
One of the things that we have on our team of more than 350
people who are primarily dedicated in their jobs to countering
terrorism and hate is expertise. So I lead this team, my
background is with more than a decade as a Federal criminal
prosecutor and safety and security are personal to me. But the
people that I have hired onto this team have backgrounds in law
enforcement, in academia, studying terrorism and
radicalization. This is something that people come to work on
at Facebook because this is what they care about. They are not
assigned to work on it while they are at Facebook. This is
bringing in expertise, and I want to make that very clear.
And then finally, similar to my colleagues here, we have
taken steps to make what we are doing very transparent. The
reports we published in the past year and a half show a steady
increase in our ability to detect terror, violence, and hate
much earlier when it is uploaded to the site and before anybody
reports it to us. Now more than 99 percent of the violent
videos and the terrorist propaganda that we removed from the
site we are finding ourselves before anybody reports it to us.
Senator Young. Thank you. Twitter.
Mr. Pickles. Thank you, Senator. I think people can be
optimistic. A few years ago, at the peak of Islamic caliphate
so-called, people challenged our industry to do more be better.
I now look at a time where 90 percent of the terrorist content
Twitter removes is detected through technology. I look at
independent academics like Professor Morecambe who talked about
the IS community being decimated on Twitter. I look at the
collaboration that we have between our companies, which didn't
exist when I first joined Twitter five and a half years ago.
All of those areas have been driven by better technology,
faster response, and a much more aggressive posture toward bad
actors.
Twitter is now showing benefit in other areas, but I think
we can also take confidence that no one is going to tell this
committee our work is done. And every one of us will leave here
today knowing we have more to do and we can never sleep. These
actors are adversarial, and we have to keep it active.
Senator Young. Thank you so much. I could spend five days,
five weeks, maybe five months, or five years in this. I only
had five minutes. I am already one minute over, Mr. Chairman.
The Chairman. Thank you. Senator Rosen, you are next. I am
going to go vote and I can assure you I will not let them close
that vote until you have asked your questions and get over
there.
Senator Rosen.
STATEMENT OF HON. JACKY ROSEN,
U.S. SENATOR FROM NEVADA
Senator Rosen. I appreciate that, Senator. Thank you for
holding this important hearing. I want to thank all the
witnesses for being here to talk about this very real and
difficult issue. The rise of extreme on extremism online is a
serious threat and the Internet is unfortunately proven of
valuable tool to extremists who are connecting with one another
through various forms to spread hate and dangerous ideologies.
While we are here to focus today on the proliferation of
extremism online, which of course is incredibly important, we
must not lose sight of the fact that violent individuals who
find communities online to fuel their hatred have also acted in
the name of hate.
We cannot ignore the fact that the absence of sensible
common-sense gun safety measures like background checks are
allowing individuals to access dangerous weapons far too
easily. And so we know the majority of Americans want us to
support that, but I represent the great State of Nevada, and as
we approach unfortunately the 2-year anniversary of the one
October shooting in Las Vegas, the deadliest mass shooting in
modern American history, we know that coordination with and
between law enforcement is more important than ever. The
Southern Nevada Counterterrorism Center also known as our
Fusion Center is an example of a dynamic partnership between 27
different law enforcement agencies to rapidly and accurately
respond to terrorists and other threats.
With Las Vegas hosting nearly 50 million tourists and
visitors each year, the Fusion Center is responsible for
preventing countless crimes and even acts of terrorism. So to
all of you, can you please discuss with us your coordination
efforts with law enforcement when violent or threatening
content is identified on your platforms, and what do you need
from us as a legislative body to promote and enable,
facilitate, whatever word you want to use, to facilitate this
partnership to keep our communities safe from another shooting
like the one in October? Please.
Ms. Bickert. Thank you, Senator. The attack was incredibly
tragic, and our hearts are with those who have suffered and did
suffer in that attack. Our relationship with law enforcement
first is an ongoing effort. We have a team that does trainings
to make sure that law enforcement understand how they can best
work with us. And that is something that we do proactively, we
reach out and offer those.
Anytime there is a mass violence incident, we reach out to
law enforcement immediately even if we are not aware of any
connection between our service and the incident. We want to
make sure that they know where we are and how to reach us. We
also have an online portal through which they can submit legal
process, including emergency requests, and we have a team that
office is staffed 24 hours a day so that we can respond
quickly.
And finally, we proactively refer imminent threat of
serious physical harm to law enforcement whenever we find them.
Senator Rosen. Thank you.
Mr. Pickles. Thank you, Senator, and I just wanted to echo
firstly Monika's sympathies for your constituents who were
victims of that horrible tragedy. The lessons I think we have
learned since that attack have continued to inform our
thinking, and for example, not waiting for the ideological
intent of the shooter to be known before acting. I think one of
the challenges we have is in the traditional terrorist space,
we might look for an organization affiliation before we would
say, this is a terrorist attack.
We don't wait for that anymore. We act first to stop people
using our services. As Monika said, we do cooperate with law
enforcement and provide credible threats. I think one of the
questions and I along with colleagues from other companies
actually met with a number of agencies yesterday to discuss how
we can further deepen our collaboration, and one of the
questions we had there is a huge amount of information within
the law enforcement community, within the DHS umbrella, that is
classified that might help us understand the threats, the
trends, the situational awareness.
So understanding how more information can be shared with
our industry to inform us about the threats----
Senator Rosen. Can you provide us in writing some of the
tools that you think you might need to help you better
cooperate to protect our communities?
Mr. Pickles. Absolutely, and that was the subject of the
meeting yesterday and we had a very productive conversation.
Senator Rosen. Thank you.
Mr. Slater. Senator, broadly similar here, both in horror
and sympathy. Tragedies like that one and in the ways that we
proactively cooperate with law enforcement refer credible
threats as well as receive valid request emergency disclosure
request and respond to them expeditiously.
Senator Rosen. Thank you. I see my time is up. I am going
to submit a question for the record about combating violent
anti-Semitism online. I know other people are waiting. We have
votes. I appreciate your time and your commitment to solving,
working on this issue.
STATEMENT OF HON. MIKE LEE,
U.S. SENATOR FROM UTAH
Senator Lee [presiding]. Thank you, Senator Rosen. Your
questions will be submitted for the record. I want to start
with a simple yes or no question. I don't mean this to be a
trick yes or no question answer. It is either yes or no, or yes
or no with a brief one sentence caveat if you need to. I would
like to hear from each of the three of you, from Ms. Bickert,
and then Mr. Pickles, and then Mr. Slater: Do you provide a
platform that you regard and present to the public as neutral
in the political sense?
Ms. Bickert. Yes, Senator, our rules are politically
neutral, and we apply them neutrally.
Senator Lee. So you aspire to political neutrality as to
left versus right?
Ms. Bickert. We want to be a service for political ideas
across the spectrum.
Senator Lee. Mr. Pickles?
Mr. Pickles. We enforce our rules impartially and our rules
are crafted without ideology included.
Senator Lee. Mr. Slater?
Mr. Slater. Similarly, we craft our services without regard
to political ideology though as we have discussed today, we are
not neutral against terrorism or violent extremism.
Senator Lee. Yes, and I appreciate you pointing that out
that is of course not what I am talking about. And that leads
into the next question I wanted to raise with each of you. I
think it is important the work each of you are doing in this
area is important. It is important for anyone occupying this
space to be conscious of those things. You do a service to
those who access your services by removing things like
pornography, terrorism advocacy, and things like that. There is
a lot of debate that surrounds this issue and surrounds some of
the legal framework surrounding it.
As you know Section 230 of the Communications Decency Act
has received a lot of criticism. It protects a website from
being held liable as a publisher of information by another
information content provider. And significantly, Section 230 is
a good Samaritan provision. It gives you the promise that you
won't be held liable for taking down this type of objectionable
content that we are talking about, whether it is something that
is constitutionally protected or not. And so for each of the
same witnesses, again, I would ask you, each of you represents
a private company and each of you are accountable to your
consumers within your company. This means that in some sense,
that you have incentives to provide a safe and enjoyable
experience on your respective platform. So I have got a
question about Section 230.
Does Section 230, particularly the Good Samaritan
provisions, help you in your efforts to swiftly take down
things like pornography and terrorist content off your
platforms? And would it be more difficult without the legal
certainty that Section 230 provides?
Ms. Bickert. Absolutely, Senator. Section 230 is critical
to our efforts in safety and security.
Senator Lee. Mr. Pickles?
Mr. Pickles. Absolutely. I would go further and say that
Section 230 has been critical to the leadership of American
industry in the information technology sector.
Senator Lee. Mr. Slater?
Mr. Slater. Absolutely. Yes.
Senator Lee. On a related point, imagine a world where this
is suddenly taken away, where those provisions no longer exist.
Large companies like yours might be able to--I strongly suspect
still would be able to and still probably would filter out this
content between the artificial intelligence capabilities at
your disposal and the human resources that you have.
I suspect you could and probably would still do your best
to perform the same function. What about a startup, what about
a company trying to enter into the space that each of your
companies entered into when they were created not very many
years ago? What would happen to them? Ms. Bickert?
Ms. Bickert. Senator, thank you for that question. This
reminds me of industry conversations involving smaller
companies back before we formed the Global Internet Forum to
Counter Terrorism in June 2017. We were having closed door
sessions with companies, large and small, to talk about the
best ways to combat the threat of terrorism online, and the
smaller companies were very concerned about liability. Section
230 is very important for them to be able to begin to
proactively act and assess content.
Mr. Pickles. I would say it is a fundamental part of
maintaining a competitive online ecosystem. And without it, the
ecosystem is less competitive.
Senator Lee. Mr. Slater?
Mr. Slater. Yes, and I just add, the U.S. has Section 230
and that is part of the reason why we have been a leader in
economic growth and innovation and technological development.
Other countries that don't have something like it suffer, and
study after study has shown that. And we will be happy to
discuss that more.
Senator Lee. If it were to be taken away--so all three of
your companies, in particular Mr. Slater, not exactly known for
being a small business or a business with a modest economic
impact, but you can identify, I assume, with this concern I am
expressing if we were to take that away Google might be able to
keep up with what it needs to do, but wouldn't it be harder for
someone to start say a new search engine company, a new tech
platform of one sort or another, as somebody starting out in
the same position where your company was a couple of decades
ago. Wouldn't that be exponentially more difficult?
Mr. Slater. I think it would create problems for innovators
of all stripes, but certainly small, medium sized businesses
would have a lot of trouble potentially getting their arms
around that sort of significant change to the fundamental legal
framework of the internet.
Senator Lee. Thank you. My time has expired.
Senator Baldwin.
STATEMENT OF HON. TAMMY BALDWIN,
U.S. SENATOR FROM WISCONSIN
Senator Baldwin. Thank you. I wanted to begin by thanking
our full committee Chairman Wicker for holding this hearing. I
think it is a vital conversation for us to be having. We need
to be taking a hard look at how we address the rising tide of
online extremism and its real world consequences in our
country. I do have some questions for you on this important
topic, but first I wanted to echo some of what my colleagues
have already said, which is there is much more that the Senate
must do to address gun violence, whether or not it is connected
to hatred espoused on the internet.
So more than 200 days ago, the House of Representatives
passed a bipartisan universal background check bill and this
common-sense gun safety measure has an extraordinary level of
public support. It deserves a vote on the Senate floor, and I
feel like we can't simply have hearings, but we have to act to
reduce gun violence. Mr. Selim, ADL Center on Extremism has
closely studied hate crimes and extremist violence in this
country. Is it fair to say that there has been an alarming
increase in bias-motivated crimes including extremist killings
in the last several years?
Mr. Selim. Yes, Senator, that is accurate.
Senator Baldwin. In the case of extremist killings, what
role do you feel that access to firearms has played in that
increase?
Mr. Selim. Senator, thank you for that question. As I
briefly alluded to earlier just to expand on what I was
mentioning, according to our recent ADL report, extremists of
all ideological spectrums that committed murders or homicides
in the United States, 73 percent of those acts were committed
with firearms.
Senator Baldwin. Thank you. What impact do you believe this
increase in hate crimes, including extremist killings, have on
the minority communities whose members have been the targets of
these attacks, and let me just add to that question. One of the
unique aspects of a hate crime is that it now not only
victimizes the targeted victim, but it strikes fear among those
who share the same characteristics with the victim or victims.
Mr. Selim. Senator, thank you for making this point. In the
past 24 months, we saw a calendar year 2017 with a 57 percent
increase of anti-Semitic incidents across the country. The FBI
and DOJ's own hate crime data showed a 17 percent increase in
hate crimes and bias-motivated crimes in calendar 2017. We
continue to see these troubling statistics year after year and
so it is imperative, and part of my testimony today, both the
submitted written and my oral testimony, speaks to the need for
greater enhancement and enforcement of hate crime laws and
protections for victims.
Senator Baldwin. I am an original co-sponsor of Senator Bob
Casey's legislation to disarm hate crime, hate act, which would
bar those convicted of misdemeanor hate crimes from obtaining
firearms. Do you agree that this measure could help keep guns
out of the hands of individuals who might engage in extremist
violence?
Mr. Selim. Yes, Senator. Thank you for your leadership and
all members who have supported this legislation. ADL supports
this legislation.
Senator Baldwin. Thank you. I appreciate the efforts that
our witnesses from the social media companies have described
regarding their company's efforts to combat online extremism,
including to provide some transparency to their users and the
general public. It is of course critically important to
understand how you are addressing problems within your existing
services and platforms. I would actually like to learn more
from you about how you are thinking about this issue as you
develop and introduce new products.
In other words, I think a lot of us feel that the approach
of rapidly introducing a new product and then assessing the
consequences later is a problem. So I would like to ask you how
do you plan to build combating extremism into the next
generation of ways in which individuals engage online, and why
don't we start with you Ms. Bickert?
Ms. Bickert. Thank you for the question, Senator. Safety by
design is an important part to building new products at our
company. One of the things we have built in the past maybe 5
years is a new products policy team that is under me. Their
responsibility is to make sure they are aware of new products
and features that are being built and explaining to these
engineers who are thinking of all the wonderful ways that the
service could be used, all of the abuse scenarios that we could
also envision and making sure that we have reporting mechanisms
or other safety features in place.
Mr. Pickles. I think as I said earlier, we are in a very
adversarial space. We know that bad actors will change the
behavior. And so every time we have a feature, a policy
decision, one of the key processes in that part of the
discussion is how can this be used against us? How can this be
gamed? How will people change their behavior to try and
circumvent the policy? And you are absolutely right. We need to
take that learning and share it with smaller companies. Certain
the work that FCT has done, working with more than I think 200
small companies around the world to share that knowledge with
them, to help them understand the challenges, is also
invaluable.
Mr. Slater. Similarly, our trust and safety teams are at
the table with product managers and engineers from the
conception of an idea all the way through the development and
possible release. So from ground up, it is safety by design.
Senator Baldwin. Thank you.
STATEMENT OF HON. DAN SULLIVAN,
U.S. SENATOR FROM ALASKA
Senator Sullivan [presiding]. So I want to thank the
witnesses, and I am going to be taking over as the Chair, and I
will call on myself as the next witness. I want to actually ask
all of you, you know, your companies, your technology, you are
famous for its algorithms, which seemed to have the ability to
pinpoint on what people want. You know, you can put an e-mail
out or even some people think, talk about say your interest in
yellow sweaters, and next thing you know, you have ads popping
up on your Facebook or other accounts that talk about yellow
sweaters. Who knows how that happens but to a lot of us it has.
It is pretty impressive.
But here is my question. If your algorithm technology is so
good at, kind of, pinpointing things like that, what people are
interested in, particularly as it relates to ads, what are the
challenges with regard to directing that kind of technology to
help us and help you find what is being talked about here on
both sides of the aisle which is the people who are committing
this kind of violence are typically disaffected young males,
and aren't there signs, aren't there things that you can do
with the technology that you do so well in other spaces to at
least provide more warning signs of this kind of violence from
these kind of individuals who in some ways already have a
profile online? Throw that out to any of you. And are you
working on that?
Ms. Bickert. Thank you for the question, Senator.
Technology plays a huge role in what we are doing to enforce
our safety policies at Facebook. In the area of terrorism to
extremism, and violence, it is not just the matching software
that we have to stop things like organized terror propaganda
videos.
We are now using artificial intelligence machine learning
to get better at identifying new content that we have not seen
before that might be promoting violence or trying to incite
violence or engage in other harmful behavior. Anytime that we
find a credible threat of imminent physical harm, we
proactively send that out to law enforcement. And these systems
are getting better every day.
Senator Sullivan. And are you using algorithms and the
advanced technologies that you use in other spaces to help
identify those threats?
Ms. Bickert. There are certainly cross learnings across the
company. There are different products that work in different
ways, but----
Senator Sullivan. But is it a priority of yours, the way it
would be for selling yellow sweaters?
Ms. Bickert. Oh, absolutely. And this is something that we
do----
Senator Sullivan. Can I ask that of all the companies here?
Mr. Pickles. Absolutely, investing in technology to find
content that is terrorist content, violent extremist content,
is absolutely a priority.
Mr. Slater. It is a top priority. Yes.
Mr. Selim. Senator, I would only add to this part of the
conversation as someone who studied the research in the data
around these issues for nearly two decades, the threat
environment that we are in today has changed significantly.
White supremacist terrorists in the United States do not have
training camps in the same way that foreign terrorist groups do
like Al-Qaeda or Isis. Their training camp where they connect,
learn, and coordinate with one another is in the online space.
So it is imperative that the question you are asking about
the machine learning, the technology, the artificial
intelligence continue to advance to disrupt that environment
and make it an inhospitable place for individuals that want to
promote violent content of any ideological spectrum to be
disrupted.
Senator Sullivan. Let me ask another question. This is kind
of a bigger kind of policy question, but you all of your
companies kind of have this tension between you want eyeballs,
on right, you want more clicks, you want more time on, and
yet--with Facebook or Google or Twitter, and yet there I think
there is increasing studies that are showing for example the
amount of young men and women, young girls, who feel kind of a
sense of loneliness from their time online.
You know, there is indications that among teenagers, the
suicide rates are increasing particularly for young girls. One
of the things that I worry about, you know, we are all dealing
with this opioid epidemic right now and we are looking back
going, my God, how did we how did we do that? How did we get to
this position in the 90s and the policies, and other things
that you know, 72,000 Americans died of overdoses last year.
And so we are, kind of, looking backward saying, how did
this happen? Do you, in your kind of c-suites of policymaking,
do you ever wonder why we can be looking back in 20 years
going, how in the hell did we addict a bunch of young Americans
to look at their damn iPhones 8 hours a day and 20 years from
now we are going to be seeing the social and physical and
psychological ramifications where we all might be kicking
ourselves in the head saying, why did we allow that to happen?
Do you guys ever think about that? Because I think about
that and it worries me, but you have tension because you want--
don't you want more Facetime, don't you want young teenagers
spending 7 hours a day staring at their iPhones because that
helps your revenues? Do you worry that 15, 20 years from now,
we are going to be in the same spot that we are with opioids
and saying, what did we do to our kids? What did we do to our
citizens? Do any of you guys worry about that? Your power, your
negative implications of what is happening in society right
now.
Ms. Bickert. Senator, thank you for the question. As a
mother, I take these questions about wellness very seriously
and our company does as well. And this is something that we
look at and we talk to youth wellness groups to make sure that
we are crafting products and policies that are in the best
long-term interests of the people who want to come and connect
through Facebook.
I also want to say that we have seen social media be a
tremendous place for support for those who are thinking of
harming themselves or struggling with eating disorders or
opioid addiction or getting exposed to hateful content. And so
we are also exploring and developing ways of linking people up
with helpful resources. We already do that now for opioid
addiction, for thoughts of self-harm, for people who are asking
or searching for hateful content. We now provide them with help
resources. We do think that this can be a really positive thing
for overall wellness.
Mr. Pickles. I just thought we have similar programs in
place for both opioids searches and also for people who are
using terms referencing self-harm or suicide where we will
provide, intervene and provide them with a source of support.
And that is something we have rolled out around the world. I
think the other thing is we certainly recognize that things
like digital literacy are issues that we as industry and
certainly we as Twitter need to invest in to make sure that as
people using our services, they also have the skills and the
awareness to use them discerningly.
And then finally, our CEO is committed to the company, to
looking at the health of the conversation, and not just using
the kind of metrics that you have referenced but looking at
much more broader metrics that measure the health of the
conversation rather than just revenue.
Senator Sullivan. Thank you, Mr. Chairman.
The Chairman [presiding]. Thank you, Senator Sullivan.
Senator Cruz.
STATEMENT OF HON. TED CRUZ,
U.S. SENATOR FROM TEXAS
Senator Cruz. Thank you, Mr. Chairman, and I will say thank
you to my friend from Alaska for sharing apparently this deep
void and longing in your heart. And I just want to reassure you
for Christmas, you will be getting that yellow sweater.
[Laughter.]
Senator Cruz. Mr. Slater, I want to start with you. I want
to talk a little bit about Project Dragonfly. In August 2018 it
was reported that Google was developing a censored search
engine under the alias of Project Dragonfly. In response to
those concerns, Alphabet shareholders requested that the
company publish a human rights impact assessment by October 30
of this year examining the actual and potential impacts of
censored Google search in China.
However, during Alphabet shareholder meeting on June 19,
the proposal for the assessment was rejected. In fact
Alphabet's Board of Directors explicitly encouraged
shareholders to vote against the proposal and Alphabet
commented that ``Google has been open about its desire to
increase its ability to serve users in China and other
countries.
We have considered a variety of options for how to offer
services in China in a way that is consistent with our mission
and have gradually expanded our offerings to consumers in
China.'' So I want to start with just some clarity. Mr. Slater,
has Google ceased any and all development and work on Project
Dragonfly?
Mr. Slater. Senator, to my knowledge, yes.
Senator Cruz. And has Google committed to foregoing future
projects that may be named differently, but would be focused on
developing a censored search engine in China?
Mr. Slater. Senator, we have nothing to announce at this
time. And I think whatever we would do, we would look very
carefully at things like human rights. In fact, we work with
the Global Network Initiative on an ongoing basis to evaluate
how our principles, our practices, our products comport with
human rights in the law.
Senator Cruz. So, roughly contemporaneously, Google decided
that it didn't want to work with the U.S. Department of
Defense. How does Google justify having been willing to work
with the Chinese government on complex projects including
artificial intelligence under Project Maven and at the same
time not being willing to help the Department of Defense
develop ways to minimize civilian casualties through better AI?
How do you how do you reconcile those two approaches?
Mr. Slater. Senator, as we have talked about today, we do
partner with law enforcement and we do partner with the
military in certain ways offering some of our services. Also as
a business, we draw responsible lines about where we want to be
in business, including limitations on and getting in the field
of building weapons and so on, and you know, we will continue
to evaluate that over time.
Senator Cruz. Let me shift to a different topic which is
this panelists talked about combating extremism and the efforts
of social media to do that. Many Americans, including myself,
have a long-standing concern that when big tech says it is
combating extremism that that is often a shield for advancing
political censorship. Mr. Pickles, I want to talk about
recently Twitter extended its pattern of censorship to the
level that it took down the Twitter account of the Senate
Majority Leader Mitch McConnell.
That I found a pretty remarkable thing for Twitter to do,
and it did so because that account, as I understand it, had
sent out a video of angry protesters outside of Senator
McConnell's house, including an organizer of Black Lives Matter
in Louisville, who is heard in the video saying that the Senate
Majority Leader ``should have broken his little raggedy
wrinkled ass neck,'' and someone else who had a voodoo doll of
the Majority Leader and another angry protester said, ``just
stab the mf's heart,'' although that person did not abbreviate
mf. Senate Majority Leader sent out those threats of violence
and found rather remarkably his own Twitter account taken down.
How does Twitter explain that?
Mr. Pickles. Well, thank you Senator for the opportunity to
discuss this. Something we have been asked around the world is
the climate in many political jurisdictions of safety of people
who hold public office. And so when we saw a video posted by
numerous users that clearly identified someone's home and
clearly contained as you so referenced some quite severe
threats out of an abundance of caution, we did remove that
video. We didn't remove the accounts. We moved that single
tweet that contained the video from everybody who had posted it
because the essence of a video with someone's personal home
where the Senate Majority Leader may have been residing at the
time with several violent references, we felt was something out
of an abundance of caution we should remove. We then discussed
this further with the Leader's office. We understood their
intent was to call attention to those very threats of violence.
And so we did permit the video to be put on Twitter with a
warning message saying this is sensitive media, but it is that
balance that we are striking between--I have been in many
different situations where I have been asked the exact opposite
which is similar content should be removed because it contains
a clear violent threat, and that balance is something that we
strive to get right every day. But our first thought in that
instance was the safety of Leader McConnell and his family.
Senator Cruz. You would agree there is a difference between
someone posting video where they are threatening someone else
and the target of that threat posting the video. Do you agree
that those are qualitatively different?
Mr. Pickles. I think that is holy fair, but I think in the
situation where you have the person's home visible in the
video, there is still a risk there and we are motivated by
preventing that offline harm that could have occurred because
the home was visible. It was a hardcore and we appreciate the
Leader's discussion and discussing with his campaign team and
his Senate office, and we appreciate their insight. But this
was something that our motivation was to prevent harm, not the
kind of potentially ideological issues you may allude to.
Senator Cruz. Thank you.
The Chairman. But Mr. Pickles, have you rethought your
policy since the instance that Senator Cruz asked about? And I
would call your attention to Ms. Bickert's testimony, written
testimonial, on page 2 which says, and I quote, ``we do not
allow propaganda or symbols that represent any of these
organizations or individuals to be shared on a platform unless
they are being used to condemn or inform.'' Is that language
instructive to your platform and don't you think that clearly
it was readily evident from the beginning that Senator
McConnell and his campaign had posted that video to condemn and
inform?
Mr. Pickles. I think this is an absolutely relevant issue.
We as a company have taken a more aggressive posture after the
Christchurch attack. We did see people posting both excerpts of
the manifesto and content of the video to condemn it, and we
decided even in those circumstances we would remove it. And for
other attacks more recently in the United States where images
have been posted to other manifestos with large chunks of the
manifestos even where they are condemning it, we have taken the
decision to remove that material.
So this is something that is constantly under tension and I
think the case you illustrate highlights for us, the complexity
in getting this right. But again if we are going to err on the
side of caution, fewer violent threats and fewer people's homes
being visible on our platform is notably a good thing. We have
to work harder at taking into account the kind of context you
outline, but this is something where this is the first time--I
have been with the company five and a half years, I have never
been asked why didn't we leave something up that contains a
violent threat, and so I think that in itself is illustrative
of the complexity of the situation.
The Chairman. Well in terms of the context in this instance
it was the owner of the home who chose to inform the world
about what was being said against him, and it was the
individual himself who posted this. And it seems to be a clear
cut case in that instance that differentiates it from the
condemnation of the larger incident of the Christchurch
violence. I would just suggest that it shouldn't have taken
very long for Twitter to understand that. Senator Sullivan, you
are recognized.
Senator Sullivan. Thank you, Mr. Chairman. I just have a
couple of follow-up questions. Mr. Slater, one of Senator
Cruz's questions. You know, I think it is--whether a company
wants to work with the Pentagon I think is something that
leadership of the company, individual companies have to make
that decision. I think that is certainly something that is
fine.
I think what troubles a number of us is that where there is
a declaration that you are not willing to work with the
Department of Defense on certain issues and yet there is a
willingness to work with one of our country's potential
adversaries, particularly on sensitive technological issues
that are important to the competition between the two nations.
Do you understand why that has caused bipartisan concern
here? And how should we address it? Should Congress take action
on those kinds of situations? Not saying everybody has to work
for the Pentagon, that is your decision. But if you don't want
to work to help with our Nation's defense, but you are working
with the country that poses a very significant threat long-term
to the United States, do you understand why that causes concern
here?
Mr. Slater. Senator, I do appreciate the concern. We are
proudly an American company. We are a business that wants to
draw a responsible lines and we look forward to continue to
engage with you, the Committee, and others to make sure we are
doing that.
Senator Sullivan. Do you think if there are instances of
that, a clear-cut example of, hey, we are not going to do
anything on the Nation's defense with the U.S. Department of
Defense, but we are going to work with the Chinese, something
very clear and obvious. Do you think there is something that we
should do to prevent that or penalize that? We the Congress?
Mr. Slater. I think it is an important question. I think as
a business we try and strike responsible and consistent lines,
but the details would certainly have to matter.
Senator Sullivan. OK. Mr. Pickles let me ask just a one
final question. It is really a follow up to Senator Scott's
earlier question. You said that the Twitter account of Maduro
in Venezuela has not ``broken any of the rules.'' What are
those rules? And at what point would you look to have somebody
who is certainly not treating his citizens well? And Senator
Scott has been a leader on this issue, but, you know, what are
those rules and at what point would you look at what they are
doing in their own citizen as a way to maybe not provide them
the platform that you have?
Mr. Pickles. Thank you. Well firstly the rules apply to any
user on Twitter at the same. I can make a make a full copy
available and it will be, for example, whether it is
encouragement of violence. If the Twitter account was used in
some of the ways that we have seen around the world to
encourage violence against minorities, to organize violence, we
would take action on those accounts breaking those rules.
Senator Sullivan. Would Twitter allow Putin to have an
account or Xi Jinping to have an account?
Mr. Pickles. If they were acting within our rules, the one
thing I would note is, and this is slightly different but
important, some worldly, some governments have sought to
manipulate our platform to spread propaganda information
through breaking our rules. One of those governments is
Venezuela, and we have made a public declaration of every
account that we removed from Twitter for engaging in
information operations covertly that we believe is responsible
for that government.
We made that whole archive available to the public and to
researchers. We have taken this same step with information
operations that have been directed we believe from countries
including China, Iran, and Russia because we believe that it is
not just those single Twitter accounts, that some governments
do also seek to manipulate our platform. And where they do so,
we will take action to remove that manipulation and make it
public so people can learn----
Senator Sullivan. So if the government takes violence
against its own citizens, is that breaking the Twitter rules?
Mr. Pickles. What I think of that act is activities
happening offline, and the key question for us is, what is
happening on Twitter?
Senator Sullivan. Thank you. Thank you, Mr. Chairman.
The Chairman. Thank you, Senator Sullivan. And thank you to
our witnesses. The hearing record will remain open for two
weeks. During this time, Senators are asked to submit any
questions for the record. Upon receipt, the witnesses are
requested to submit their complete written answers to the
Committee as soon as possible but no later than Wednesday,
October 2, 2019 by close of business.
I thank each and every one of you for appearing today. This
hearing is now adjourned.
[Whereupon, at 12:08 p.m., the hearing was adjourned.]
A P P E N D I X
16 September 2019
To: Chairman Roger Wicker,
Ranking Member Maria Cantwell,
U.S. Senate Committee on Financial Services,
Washington, DC.
Fr: Gretchen Peters and Professor Amr al-Azm
The Alliance to Counter Crime Online
Re: Concerns Facebook Platforms Facilitate Terror, Spread Crime
Dear Chairman Wicker and Ranking Member Cantwell,
As the Senate Committee on Commerce, Science and Transportation
prepares to question Facebook's Head of Global Policy Management, we
want to express our grave concern that Facebook's platforms are
infested by criminal syndicates and terror groups. Facebook has been
grossly negligent both in monitoring and removing this toxic content.
Moreover, one of our members has performed research indicating the firm
has knowingly deceived lawmakers, investors and the public about the
extent to which the firm is able to remove extremist content.
We want the committee to understand that the world's largest social
media company does more than just connect people. The public should not
trust Facebook's claim that they have been successful in removing 99
percent of ISIS content because it is only a talking point that they
have never been forced to prove. Our research indicates Facebook and
its family of platforms are also used by terrorist groups as a
megaphone for propaganda, for recruiting new members, and even to
fundraise. Just this week, ACCO is preparing to release a report that
documents extensive fund-raising activities by designated terror groups
such as Lebanese Hezbollah.
ACCO members also include a group of brave Syrian archeologists
investigating the illicit antiquities trade on Facebook. They have
recorded closed groups where almost 2 million regular users log on to
trade tens of thousands of artifacts trafficked from conflict regions
including Syria, Iraq and Yemen--a war crime. Many of the sellers
openly declare they are donating proceeds of these sales to ISIS.
The Facebook family of apps are ground zero for organized crime
syndicates to connect with buyers, market their illegal goods, and move
money, using the same ease of connectivity enjoyed by ordinary users.
Instead of acknowledging his technology is being used for illegal
purposes and fixing the problem, Facebook CEO Mark Zuckerberg clings to
immunities provided by Section 230 of the Communications Decency Act of
1996, which courts have interpreted to mean that tech firms shouldn't
be held liable for content posted by third-parties.
There is a huge problem with this approach. The algorithms Facebook
has touted to connect the world have connected criminals and terrorists
faster than Facebook's own beleaguered moderators can delete them. The
impact of this illegal activity is affecting our communities, our
cultures, and our environment, and it's happening in the same digital
spaces where our children play, our families connect, and our companies
advertise.
In light of all this, Zuckerberg's announcement that he plans to
alter Facebook to focus on groups--and also launch a cryptocurrency--
are downright alarming. Groups are already the epicenter for illicit
activity on Facebook. Do we want Facebook to become an even safer place
for terrorists and criminals?
There's no reason to believe Facebook's proposed changes will make
user data any more secure. After all, Facebook hasn't changed its
fundamental business model. But the changes will make it harder for
authorities and civil society groups to track and counter illegal
activity on the platform.
The firm's continued negligence in the moderation of criminal and
terror content makes clear that the time for self-regulation has
passed.
The challenge is that Federal laws take time, something that human
trafficking victims, drug addicts and endangered species don't have.
But there are other ways U.S. regulators can address crime on social
media. Facebook's IPO may hold the key to effective regulation.
When Facebook went public in 2012, the firm voluntarily entered
into a strict regulatory regime that negates CDA 230 immunities in the
context of Facebook's obligations under securities law. The firm's lack
of internal controls and effective compliance programs implicate
potentially serious securities law violations. Your committee can
influence immediate action by asking the Securities and Exchange
Commission (SEC) to utilize its existing regulatory power.
As a result of Facebook's failure to establish appropriate internal
controls, criminal activity has accelerated on its platform and
continues to grow. Now is not the time to let Facebook launch a
cryptocurrency. It's time to make social media a safer space for all.
Respectfully,
Gretchen Peters, Executive Director
Alliance to Counter Crime Online
Dr. Amr AI-Azm, Co-founder of ACCO
Director of ATHAR Project
______
[GRAPHIC(S) NOT AVAILABLE IN TIFF FORMAT]
______
Response to Written Questions Submitted by Hon. Shelley Moore Capito
to Monika Bickert
Question 1. You mentioned in your testimony the importance of
counter speech to prevent people from becoming radicalized.
Radicalization comes in many forms and threatens the core values our
Nation was created upon. Are there different strategies and best
practices in combating domestic vs foreign extremism?
Answer. Terrorists, terrorist content, and hate speech in all
forms--including white supremacy and domestic terrorist content--have
no place on Facebook. We prohibit content that incites violence, and we
remove terrorists and posts that support terrorism whenever we become
aware of them. We use a variety of tools in this fight against
terrorism and violent extremism, including artificial intelligence,
specialized human review, industry cooperation, and counterspeech
training.
Our definition of terrorism is agnostic to the ideology or
political goals of a group, which means it includes everything from
religious extremists and violent separatists to white supremacists and
militant environmental groups. It is about whether they use violence or
attempt to use violence to pursue those goals. And we recently updated
our definition in consultation with experts in counterterrorism,
international humanitarian law, freedom of speech, human rights, and
law enforcement. The updated definition still focuses on the behavior,
not ideology, of groups. But while our previous definition focused on
acts of violence intended to achieve a political or ideological aim,
our new definition more clearly encompasses attempts at violence,
particularly when directed toward civilians.
In addition to combating foreign terrorism, we are committed to
identifying and rooting out domestic hate organizations. We define hate
organizations as ``any association of three or more people that is
organized under a name, sign, or symbol and that has an ideology,
statements, or physical actions that attack individuals based on
characteristics, including race, religious affiliation, nationality,
ethnicity, gender, sex, sexual orientation, and serious disease or
disability.'' In evaluating groups and individuals for designation as
hateful, we have an extensive process that takes into account a number
of different signals, and we regularly engage with academics and
organizations to refine this process.
While we work 24/7 to identify, review, and remove terrorist and
violent extremist content, our efforts do not stop there. We have also
started connecting people who search for terms associated with white
supremacy and hate-based organizations to resources focused on helping
people leave behind hate groups. For example, people searching for
these terms in the U.S. will be directed to Life After Hate (https://
www.lifeafterhate.org/), an organization founded by former violent
extremists that provides crisis intervention, education, support
groups, and outreach. We have also recently expanded this initiative to
Australia and Indonesia, where we work with organizations with local
expertise on how best to counter hate in their communities. For more
information, see https://newsroom.fb.com/news/2019/03/standing-against-
hate.
People use our platform to speak out against hatred and extremism.
They counter hateful content by responding to it directly, raising
awareness on important issues, and supporting positive and moderate
voices. We believe these efforts to resist and stand up to racism,
violence, extremism, and hate are essential. That is why we work
closely with local communities, experts 17 in civil society and
academia, and policymakers to support counterspeech initiatives across
the globe. More information can be found at https://
counterspeech.fb.com/en.
Question 2. This Committee has held a number of hearings on the
rise and importance of artificial intelligence (AI) in today's digital
economy. AI has been invaluable in collecting and sorting massive
amounts of data. In the case of today's hearing, AI has become critical
in order to identify radicalization and terrorist threats. Each company
has identified key tools each company uses in identifying bad actors on
your platforms, but machine learning being one of the most critical.
What factors are given priority when determining radicalized or
terrorist content?
a. You also mention the importance of human expertise in
determining more nuanced cases. When does human expertise step in after
AI has identified or flags content?
b. After content has been flagged for law enforcement involvement,
what is the process that takes place afterward? Does that content get
sent to the FBI and then disseminated to state law enforcement?
Answer. We use a sophisticated machine learning tool to assess
Facebook posts that may signal support for terrorist organizations. The
tool produces a score indicating how likely it is that the post
violates our policies. In some cases, we will automatically remove
posts when the tool indicates with very high confidence that the post
contains support for terrorism. But in most cases, we still rely on
specialized reviewers to evaluate posts, and we use these scores to
prioritize which posts our reviewers assess first.
We are careful not to reveal too much about our automated
enforcement techniques, including the specific factors our machine
learning prioritizes when evaluating content, because of adversarial
shifts by terrorists. But we are seeing real gains as a result of this
work: we've removed more than 26 million pieces of content related to
global terrorist groups like ISIS and al-Qaeda in the last two years,
99 percent of which we proactively identified and removed before anyone
reported it to us.
We reach out to law enforcement whenever we see a credible threat
of imminent harm. We contact federal, state, or local law enforcement
depending on the specific circumstances of a threat.
We have a long history of working successfully with the Department
of Justice, the FBI, state and local law enforcement, and other
government agencies to address a wide variety of threats to our
platform, including terrorist threats. We have been able to provide
support to authorities around the world that are responding to the
threat of terrorism, including in cases where law enforcement has been
able to disrupt attacks and prevent harm. We have strict processes in
place to handle government requests we receive, and we disclose account
records in accordance with our terms of service and applicable law. We
also have law enforcement response teams available around the clock to
respond to emergency requests.
______
Response to Written Questions Submitted by Hon. Amy Klobuchar to
Monika Bickert
Online transparency and accountability is a top priority for me. In
April, I sent a letter urging the Department of Homeland Security and
Federal Bureau of Investigation to create a joint task force to combat
election interference and the spread of misinformation, and this week I
introduced legislation to create a Center at the Office of the Director
of National Intelligence to coordinate the existing efforts of agencies
and departments in combating foreign influence campaigns.
Question 1. Can you speak to the importance of joint efforts by
agencies, the intelligence community, tech companies, and elections
officials in combating the spread of misinformation?
Answer. We work closely with law enforcement, regulators, election
officials, other technology companies, researchers, academics, and
civil society groups to strengthen our platform against election
interference and the spread of misinformation. This coordination is
incredibly important--we can't do this alone, and we have worked to
strengthen our relationships with government, outside experts, and
other technology companies in order to share information and bolster
our security efforts.
Our partnerships, as well as our own investigations, help us find
and remove bad actors from Facebook. For example, ahead of the U.S.
midterm elections on October 26, 2018, we took down 82 Pages, Groups,
and accounts linked to Iran. And in the 48 hours ahead of the
elections, we also got a tip from the FBI which allowed us to move
quickly to take down a coordinated effort by foreign entities on
Facebook and Instagram. Based on this tip, we quickly identified a set
of accounts that appeared to be engaged in coordinated inauthentic
behavior, which is banned on Facebook because we want people to be able
to trust the connections they make on our services. So we immediately
blocked these accounts and publicly announced what we found and the
action we were taking. We also shared that information with the
government and other companies to help them with their own
investigations.
We're continuing to work closely with the FBI, the Department of
Homeland Security (DHS), and other companies on ways to protect
elections from interference on our platform. In September, security
teams from Facebook and a number of technology companies met at
Facebook with representatives from the FBI, the Office of the Director
of National Intelligence (DNI), and DHS to further strengthen strategic
collaboration regarding the security of the 2020 U.S. state, federal,
and presidential elections.
We're constantly following up on thousands of leads of potential
bad activity globally, including information shared with us by law
enforcement, industry partners, and civil society groups, and insights
from past takedowns. Over the past two years, we've seen that threats
are rarely confined to a single platform or tech company. That's why
we're working closely with our fellow tech companies to deal with the
threats we have all seen during and beyond elections. A number of
takedowns we have conducted and announced were in close collaboration
with other tech platforms, security companies, and law enforcement. We
also partner with the Atlantic Council's Digital Forensic Research Lab,
Graphika, and other researchers and experts who provide additional
analysis of the coordinated inauthentic behavior we identify, remove,
and publicly share, including their behavior off-platform, across
different Internet services.
Question 2. How could these efforts be improved?
Answer. Our partnerships have been immensely helpful, but it can be
challenging to coordinate the operations and timing of these
investigations. Timing is a key to our success, and the more entities
involved, the harder it inevitably is to get everyone synced
seamlessly. That's why it is so important to have open lines of
communication with all of these partners so we can ensure we are all
aligned, and that we take action pursuant to a timeline that best
disrupts the adversary.
The current state of the law also does not make it easy to share
information with other entities, which can hamper our partnerships in
these areas. Clear authorities or liability protections that allow for
sharing between companies and organizations would be helpful to reduce
this friction.
Security is never finished, and it will take our continuous efforts
to stay one step ahead of bad actors seeking to disrupt our elections.
The better we can be at working together, the better we will do by our
community.
In April, reports highlighted that the records of more than 540
million Facebook users were publicly exposed on Amazon's cloud service.
One provision in the privacy legislation that I lead with Senator
Kennedy requires that U.S. consumers are notified of breaches within 72
hours.
Question 3. What are your views on a Federal requirement to ensure
that consumers are informed in a timely manner when their personal
information has been compromised?
Facebook is committed to continuing to comply with breach
notification laws. In some cases, we have gone beyond our legal
obligation to notify consumers about instances where their personal
information has been compromised, even when the law did not require us
to do so.
At present, there are data breach laws in 50 states, with differing
notification thresholds and time frames. We support a Federal data
breach notification law that would create a consistent nationwide
standard for breach notifications. To avoid notification fatigue and
ensure consumer attention, legislation should establish clear rules
that require notification of the breaches most likely to harm people.
Organizations should notify people when there has been a breach
affecting their personal information that could cause them a risk of
significant harm (for example, identity theft, fraud, real-time
location-tracking, or economic loss). Breaches of information that is
encrypted, anonymized, or otherwise de-identified do not pose a risk of
significant harm and would not require notification, unless the
encryption key is also breached or information that would allow the
breached information to be re-identified is also breached. Legislation
should set forth factors that inform whether a data breach presents a
risk of significant harm, such that notification to consumers should be
required.
______
Response to Written Questions Submitted by Hon. Richard Blumenthal to
Monika Bickert
NO HATE Act and Reporting
I have introduced legislation, the Jabara-Heyer NO HATE Act, which
would help states implement and train officers in the National
Incident-Based Reporting Systems. The NO HATE Act would also provide
grants to states to better address hate crimes by training law
enforcement, establish specialized units, create community relations
programs, and run hate crime hotlines.
Question 1. Do you support the Jabara-Heyer NO HATE Act?
Answer. Hate has no place on Facebook, and we have strong
relationships with law enforcement, academics, and experts to help us
fight hate speech and hate-related violence on our platform. For
example, we're partnering with Life After Hate, an organization founded
by former violent extremists, to connect people who search for terms
associated with white supremacy to resources focused on helping people
leave behind hate groups. We also provide training to governments on
how best to flag violating content, and we have portals for law
enforcement to legally request data in ongoing and crisis scenarios.
We support providing resources to programs that deal with these
issues, and we would be happy to discuss the specifics of the proposal
with your office.
I understand that it has taken some time for Google and Facebook to
establish reliable and timely channels to report threats made on your
platform to the proper authorities. Mr. Slater testified that Google
now has a strong relationship with the Northern California Regional
Intelligence Center, who has been effective at quickly getting reports
of threats into the right hands.
Question 2. Would you support adding measures to the Jabara-Heyer
NO HATE Act to expand the NCRIC model of integrated threat reporting
nationwide?
Answer. As discussed above, we would be happy to discuss the
specifics of the proposal with your office.
When it comes to working with law enforcement, we reach out
whenever we see a credible threat of imminent harm. We contact federal,
state, or local law enforcement depending on the specific circumstances
of a threat. We have a long history of working successfully with the
Department of Justice, the FBI, state and local law enforcement, and
other government agencies to address a wide variety of threats on our
platform, including terrorist threats. We have been able to provide
support to authorities around the world that are responding to the
threat of terrorism, including in cases where law enforcement has been
able to disrupt attacks and prevent harm. We also have law enforcement
response teams available around the clock to respond to emergency
requests.
Question 3. What steps would improve communications channels with
law enforcement to make sure the right information gets into the right
hands quickly?
Answer. Please see the above responses. As discussed, we work
closely with law enforcement. Indeed, we have been able to provide
support to authorities around the world that are responding to the
threat of terrorism, including in cases where law enforcement has been
able to disrupt attacks and prevent harm.
Amplification of 8Chan and Other Hate Sites
We have seen over this year that fringe sites are a breeding ground
for racist and violent hate communities. However, extremists then use
mainstream platforms to recruit and amplify their hate and ideologies
to a larger audience. In particular, the site 8chan has had a repeated
role in multiple mass shootings this year. The perpetrators of
Christchurch mosque shootings, Poway synagogue shooting, and El Paso
massacre each posted manifestos to 8chan before their attacks. It is
also sites such as 8chan that facilitate campaigns of harassment and
terrorism that target the victims of mass shootings, such as the Sandy
Hook families. 8chan is currently offline after webhosting providers
finally cut their ties after the El Paso shootings. However, 8chan's
owner has said that he plans to revive the site as soon as this week.
Question 1. Has your company taken any steps to limit the spread of
8chan content, including the communities that hosted the manifestos of
shooters, on your platforms?
Answer. For years, we've worked to block URLs when we identify that
the content at the URL violates our policies when it is shared on
Facebook. For example, we blocked links from 8chan and 4chan when the
content shared violated our policies. And earlier this year, we started
blocking any link that connects to 8chan's/pol/board, where the
Christchurch, El Paso, and Poway attacks were advertised and where a
large amount of other hateful content has appeared.
We also work with others in the industry to limit the spread of
violent extremist content on the Internet. For example, in 2017, we
established the Global Internet Forum to Counter Terrorism (GIFCT) with
others in the industry with the objective of disrupting terrorist abuse
on our platforms. Since then, the consortium has grown and collaborates
closely on critical initiatives focused on tech innovation, knowledge-
sharing, and research. Most recently, we reached our 2019 goal of
collectively contributing more than 200,000 hashes, or unique digital
fingerprints, of known terrorist content into our shared database,
enabling each of us to quickly identify and take action on potential
terrorist content on our respective platforms.
Question 2. Please describe the specific steps you to restrict the
amplification of 8chan and other violent sites on your platforms,
including what sites you have taken action to restrict.
Answer. Please see the response to the previous question.
Testing of Consumer Platforms
Question 1. Please describe the process you use to test and
evaluate new consumer facing products, including algorithms designed to
promote forms of engagement. What methods are employed to assess the
impact of these products on individuals and groups, both for an
immediate and medium term response?
Answer. While the specific processes that we use to build and
evaluate improvements to our services vary by product, we generally
work to ensure that the features we build meet the needs and
preferences of our community. Methods we use to hear from our community
might include inviting people to sit down for one-on-one interviews,
join focus groups, try new products and features, or keep diaries about
their experiences with apps over time. We also invite large groups of
people to take surveys, often via the Facebook app itself--indeed, tens
of thousands of people opt into taking surveys every week. And with
more than 2 billion people using Facebook every month, we have to
carefully consider ways to ensure we are hearing from representative
swaths of the community, all over the world.
Before new products and services are launched, company executives
perform internal reviews, and we often release features slowly so that
we can understand how people are using new features before they are
available to everyone on Facebook.
Changes to our products and services that involve people's personal
information are also reviewed through a cross-functional evaluation
process overseen by the Chief Privacy Officer for Product, which
involves our Chief Privacy Officer for Policy, legal compliance
experts, and participants from other departments across the company.
This process is a collaborative approach to privacy that seeks to
promote strong privacy protections and sound decision-making at every
stage of the product development process. Moreover, the new FTC Consent
Order, which has not yet been finalized, will impose new, rigorous
process and documentation requirements in this area. Our privacy
program is responsible for reviewing product launches, major changes,
and privacy-related bug fixes to products and features to ensure that
privacy policies and procedures are consistently applied and that key
privacy decisions are implemented. This approach has several key
benefits:
First, it is designed to consider privacy early in the product
development process. This allows us to consider the benefits that a
feature is intended to have for people who use our services, how data
will be used to deliver those benefits, and how we can build features
from the ground up that include privacy protections to enable those
benefits while protecting people's information and putting them in
control.
Second, taking a cross-disciplinary approach to privacy encourages
us to think about data protection as more than just a compliance
exercise. Instead, we evaluate how to design privacy into the features
that we build. We consider this from the perspective of things like
designing interfaces that make data use intuitive, taking a consistent
approach to privacy across our services, and building protections in
how our software is engineered. Accordingly, while we scale our privacy
review process depending on the complexity of a particular data use,
reviews typically involve experts who evaluate proposed data practices
from the perspective of multiple disciplines.
Facebook also undergoes ongoing privacy assessments to test the
effectiveness of its privacy controls, which are conducted by an
independent third-party professional pursuant to the procedures and
standards generally accepted in the profession. Facebook's privacy
program and related controls are informed by GAPP principles, which are
considered industry-leading principles for protecting the privacy and
security of personal information. We monitor the privacy program and
update the controls as necessary to reflect evolving risks. And, under
the new FTC Consent Order, we will continue to undergo these
independent reviews on a biennial basis.
Question 2. Do you ever identify unintended consequences of such
proposed products and then revise them or decide not to launch?
Answer. As described in the previous response, we test and evaluate
new products for impacts of various kinds. If we determine that a
negative impact outweighs the product's potential benefit, we do not
launch the product.
Question 3. What testing and measurement methodologies are
routinely used and how are the product evaluation teams selected?
Please submit any criteria you have developed for new or revised data
driven products or applications, including their intended impact,
demographic reach, and revenue potential.
Answer. Please see the response to your Question 1.
______
Response to Written Questions Submitted by Hon. Tom Udall to
Monika Bickert
Question 1. We cannot talk about mass violence without talking
about the social and political climate that is dividing America. Most
recently, content that demonizes and spreads hate against immigrant
communities is proliferating across social media. This content is too
often indistinguishable from social media posts from some elected
representatives.
Facebook announced recently that it would exempt politicians from
certain rules that prohibit hate speech, incite violence, or post fake
news. How did Facebook come to the decision that any content posted by
a political figure should be considered newsworthy, even if it clearly
espouses hate, incites violence, or is designed to spread
misinformation? Why does your commitment to protect users from harmful
content end when the poster is a political figure?
Answer. Our recent work builds on an existing policy. Since 2016,
we have assessed newsworthiness on a case-by-case basis, balancing the
public interest value and the risk associated with the speech.
We've applied this policy to a range of organic content not limited
in scope to politicians. We have, for example, allowed as newsworthy
images that depict war or famine, or that attempt to raise awareness of
issues like indigenous rights. The newsworthiness analysis does not
apply to ads.
Even in the case of politicians' speech, no content is
automatically deemed newsworthy. Newsworthiness requires a balancing
test to make a determination of the public interest value versus the
potential for harm. We take a number of factors into consideration,
including country-specific context such as whether there is an election
underway or the country is at war, as well as the speaker and subject
matter of the speech, such as whether it relates to governance or
politics.
In evaluating the risk of harm, we will consider the severity of
the harm. Content that has the potential to incite violence poses a
safety risk that we will take into account. And there are some types of
violations--for example, the posting of terrorist propaganda or voter
suppression--where the risk of harm will always override any public
interest value.
When it comes to fact checking, we rely on third-party fact-
checkers to help reduce the spread of false news and other types of
viral misinformation, like memes or manipulated photos and videos. We
don't believe, however, that it is an appropriate role for us to
referee political debates and prevent a politician's speech from
reaching its audience and being subject to public debate and scrutiny.
This is some of the most scrutinized speech in our society, and we
believe people should decide what is credible, not tech companies.
That's why politicians are not subject to Facebook's third-party fact-
checking program. We have had this policy on the books for over a year
now, posted publicly on our site under our eligibility guidelines. This
means that we will not send organic content or ads from politicians to
our third-party fact-checking partners for review. However, when a
politician shares previously debunked content, we will demote that
content, display related information from fact-checkers, and reject its
inclusion in advertisements.
Question 2. Knowing that the problem of extremism and mass violence
extends beyond the screen, I would like you to describe your
partnerships with communities and organizations around the country to
fight against extremism and hate. What are you doing to promote their
voices on your platforms? Moreover, what makes them effective?
Answer. We are proud of the work we have done to make Facebook a
hostile place for those committed to acts of violence. We understand,
however, that simply working to keep violence off Facebook is not an
adequate solution to the problem of online extremism and violence,
particularly because bad actors can leverage a variety of platforms and
operate offline as well. We believe our partnerships with other
companies, civil society, researchers, and governments are crucial to
combating this threat. For example, our P2P Global Digital Challenge,
which engages university students around the world in competitions to
create social media campaigns and offline strategies to challenge
hateful and extremist narratives, has launched over 600 counterspeech
campaigns from students in 75 countries, engaged over 6,500 students,
and reached over 200 million people. We're also partnering with Life
After Hate, an organization founded by former violent extremists, to
connect people who search for terms associated with white supremacy to
resources focused on helping people leave behind hate groups.
And we are continuing our work with the Global Internet Forum to
Counter Terrorism (GIFCT), an endeavor that focuses on fighting
terrorism and extremism through knowledge sharing, support for
counterterrorism work, and technical cooperation. In September, GIFCT
released a digital Campaign Toolkit produced by the Institute for
Strategic Dialogue that instructs NGOs running online counterspeech
programs in best practices for utilizing a range of digital platforms.
Just as bad actors utilize a range of platforms to get their message
out, so must counterspeech practitioners. For more information, please
see https://www.campaigntool
kit.org.
GIFCT recently announced that it will become an independent
organization led by an Executive Director and supported by dedicated
technology, counterterrorism, and operations teams. Evolving and
institutionalizing GIFCT's structure from a consortium of member
companies will build on our early achievements and deepen industry
collaboration with experts, partners, and government stakeholders--all
in an effort to thwart increasingly sophisticated efforts by terrorists
and violent extremists to abuse digital platforms.
Question 3. We are entering another election year and we know that
foreign actors have amplified divisive rhetoric on social media and, in
some cases, orchestrated actual protests. What specific actions are you
taking to prepare for 2020 to prevent Russia and other foreign actors
from trying to inflame racial and political tensions through social
media?
We have a responsibility to stop abuse and election interference on
our platform. That's why we've made significant investments since 2016
to better identify new threats, close vulnerabilities, and reduce the
spread of viral misinformation and fake accounts.
Combating Inauthentic Behavior
Over the last three years, we've worked to identify new and
emerging threats and remove coordinated inauthentic behavior across our
apps. In the past year alone, we've taken down over 50 networks
worldwide, many ahead of major democratic elections. As part of our
effort to counter foreign influence campaigns, most recently we removed
three networks of accounts, Pages, and Groups on Facebook and Instagram
for engaging in foreign interference. These manipulation campaigns
originated in Russia and targeted a number of countries in Africa. We
have identified these manipulation campaigns as part of our internal
investigations into suspected Russia-linked inauthentic behavior in the
region.
We took down these networks based on their behavior, not the
content they posted. In each case, the people behind this activity
coordinated with one another and used fake accounts to misrepresent
themselves, and that was the basis for our action. We have shared our
findings with law enforcement and industry partners. More details can
be found at https://newsroom.fb.com/news/2019/10/removing-more-
coordinated-inauthentic-behavior-from-Russia. As we've improved our
ability to disrupt these operations, we've also built a deeper
understanding of different threats and how best to counter them. We
investigate and enforce against any type of inauthentic behavior.
Protecting the Accounts of Candidates, Elected Officials, and Their
Teams
We also recently launched Facebook Protect to further secure the
accounts of elected officials, candidates, their staff, and others who
may be particularly vulnerable to targeting by hackers and foreign
adversaries. As we've seen in past elections, they can be targets of
malicious activity. However, because campaigns are generally run for a
short period of time, we do not always know who these campaign-
affiliated people are, making it harder to help protect them.
Page admins can enroll their organization's Facebook and Instagram
accounts in Facebook Protect and invite members of their organization
to participate in the program as well. Participants will be required to
turn on two-factor authentication, and their accounts will be monitored
for hacking, such as login attempts from unusual locations or
unverified devices. And, if we discover an attack against one account,
we can review and protect other accounts affiliated with that same
organization that are enrolled in our program. You can find more
information about Facebook Protect at https://www.facebook.com/gpa/
facebook-protect.
Making Pages More Transparent
We want to make sure people are using Facebook authentically and
that they understand who is speaking to them. Over the past year, we've
taken steps to ensure Pages are authentic and more transparent by
showing people the Page's primary country location, whether the Page
has merged with other Pages, and information about the organization
that owns the Page. This gives people more context on the Page and
makes it easier to understand who is behind it.
Labeling State-Controlled Media
We want to help people better understand the sources of news
content they see on Facebook so they can make informed decisions about
what they are reading. We will soon begin labeling media outlets that
are wholly or partially under the editorial control of their government
as state-controlled media. This label will be on both their Page and in
our Ad Library. We will hold these Pages to a higher standard of
transparency because they combine the opinion-making influence of a
media organization with the strategic backing of a state.
Making it Easier to Understand Political Ads
Throughout this year, we've been expanding our work around the
world to increase authenticity and transparency around political
advertising because we know how important it is that people understand
who is publishing the ads that they see. We have now launched our
publicly searchable Ad Library in over 190 countries and territories.
We allow advertisers to be authorized to purchase political ads and we
give people more information about ads that concern social issues,
elections, or politics. We require the use of these transparency tools
in over 50 jurisdictions, and we make them available for voluntary use
in over 140 others, to provide the option of greater transparency and
accountability.
We have added a variety of features to our ads transparency tools
to help journalists, lawmakers, researchers, and others learn more
about the ads they see, including information about how much candidates
have spent on ads. And soon we will also begin testing a new database
with researchers that will enable them to quickly download the entire
Ad Library, pull daily snapshots, and track day-to-day changes.
More Resources for Rapid Response for Elections
We have set up regional operations centers focused on election
integrity in California, Dublin, and Singapore. These hubs allow our
global teams to better work across regions in the run-up to elections
and further strengthen our coordination and response time between staff
in Menlo Park and in-country. These teams add a layer of defense
against fake news, hate speech, and voter suppression and work cross-
functionally with our threat intelligence, data science, engineering,
research, community operations, legal, and other teams.
Preventing the Spread of Viral Misinformation
On Facebook and Instagram, we work to keep confirmed misinformation
from spreading. For example, we reduce its distribution so fewer people
see it--on Instagram, we remove it from Explore and hashtags, and on
Facebook, we reduce its distribution in News Feed. On Instagram, we
also make content from accounts that repeatedly post misinformation
harder to find, for example by filtering content from that account from
Explore and hashtag pages. And on Facebook, if Pages, domains, or
Groups repeatedly share misinformation, we'll continue to reduce their
overall distribution, and we'll place restrictions on the Page's
ability to advertise and monetize.
Over the coming weeks, content across Facebook and Instagram that
has been rated false or partly false by a third-party fact-checker will
start to be more prominently labeled so that people can better decide
for themselves what to read, trust, and share. Labels will be shown on
top of false and partly false photos and videos, including on top of
Stories content on Instagram, and will link out to the assessment from
the fact-checker.
Much like we do on Facebook when people try to share known
misinformation, we are also introducing a new pop-up that will appear
when people attempt to share posts on Instagram that include content
that has been debunked by third-party fact-checkers.
In addition to clearer labels, we are also working to take faster
action to prevent misinformation from going viral, especially given
that quality reporting and fact-checking takes time. In many countries,
including in the US, if we have signals that a piece of content is
false, we temporarily reduce its distribution pending review by a
third-party fact-checker.
Fighting Voter Suppression and Intimidation
Attempts to interfere with or suppress voting undermine our core
values as a company, and we work proactively to remove this type of
harmful content. Ahead of the 2018 midterm elections, we extended our
voter suppression and intimidation policies to prohibit:
Misrepresentation of the dates, locations, times, and
methods for voting or voter registration (e.g., ``Vote by
text!'');
Misrepresentation of who can vote, qualifications for
voting, whether a vote will be counted, and what information
and/or materials must be provided in order to vote (e.g., ``If
you voted in the primary, your vote in the general election
won't count.''); and
Threats of violence relating to voting, voter registration,
or the outcome of an election.
We remove this type of content regardless of who it's coming from.
Ahead of the midterm elections, our Elections Operations Center removed
more than 45,000 pieces of content that violated these policies--more
than 90 percent of which our systems detected before anyone reported
the content to us.
In advance of the U.S. 2020 elections, we're implementing
additional policies and expanding our technical capabilities on
Facebook and Instagram to protect the integrity of the election.
Following up on a commitment we made in the civil rights audit report
released in June, we have now implemented our policy banning paid
advertising that suggests voting is useless or meaningless or advises
people not to vote. In addition, our systems are now more effective at
proactively detecting and removing this harmful content. We use machine
learning to help us quickly identify potentially incorrect voting
information and remove it.
We are also continuing to expand and develop our partnerships to
provide expertise on trends in voter suppression and intimidation, as
well as early detection of violating content. This includes working
directly with secretaries of state and election directors to address
localized voter suppression that may only be occurring in a single
state or district. This work will be supported by our Elections
Operations Center during both the primary and general elections.
Helping People Better Understand What They See Online
Part of our work to stop the spread of misinformation is helping
people spot it for themselves. That's why we partner with organizations
and experts in media literacy. We recently announced an initial
investment of $2 million to support projects that empower people to
determine what to read and share--both on Facebook and elsewhere.
These projects range from training programs to help ensure the
largest Instagram accounts have the resources they need to reduce the
spread of misinformation, to expanding a pilot program that brings
together senior citizens and high school students to learn about online
safety and media literacy, to public events in local venues like
bookstores, community centers, and libraries in cities across the
country. We're also supporting a series of training events focused on
critical thinking among first-time voters.
In addition, we're including a new series of media literacy lessons
in our Digital Literacy Library. These lessons are drawn from the Youth
and Media team at the Berkman Klein Center for Internet & Society at
Harvard University, which has made them available for free worldwide
under a Creative Commons license. The lessons, created for middle and
high school educators, are designed to be interactive and cover topics
ranging from assessing the quality of the information online to more
technical skills like reverse image search.
Question 4. Regarding the shared industry database of hashes linked
to content that promotes terrorism; I would like to understand the
thresholds for including certain content in the database. Who makes the
decision to include content in that database and how is that decision
made? What percent of that database concerns white nationalist or other
domestic extremist content?
Answer. Facebook's internal terrorism definition applies to a wide
range of terrorist actors, regardless of ideology or designation by
governments or intergovernmental entities. We have designated more than
200 white supremacist organizations under our broader Dangerous
Organizations policy. For the purposes of the hash-sharing database,
GIFCT uses the UN's Consolidated Sanctions List to identify groups for
which we will share hashes of any terrorist-related content or
propaganda found. Following the Christchurch attack, we have also
developed a Content Incident Protocol (CIP), which enables companies to
share hashes related to propaganda produced by attackers during a
terrorist attack. The CIP was deployed for the first time after the
October 9 attack in Halle, Germany.
Companies also agreed upon a basic taxonomy to describe the type of
content ingested into the hash-sharing database. The taxonomy includes
the following labels that are applied to the content when a company
adds hashes to the shared database:
Imminent Credible Threat: A public posting of a specific,
imminent, credible threat of violence toward non-combatants
and/or civilian infrastructure.
Graphic Violence Against Defenseless People: The murder,
execution, rape, torture, or infliction of serious bodily harm
on defenseless people (prisoner exploitation, obvious non-
combatants being targeted).
Glorification of Terrorist Acts: Content that glorifies,
praises, condones, or celebrates attacks after the fact.
Recruitment and Instruction: Materials that seek to recruit
followers, give guidance, or instruct them operationally.
New Zealand Perpetrator Content: The GIFCT set a new
precedent in the wake of the New Zealand terrorist attack. Due
to the virality and cross-platform spread of the attacker's
manifesto and attack video, and because New Zealand authorities
deemed all manifesto and attack video content illegal, the
GIFCT created a crisis bank to mitigate the spread of this
content.
GIFCT categorizes the content ingested based on these categories.
As of July 2019, the breakdown of the content in the database is as
follows:
Imminent Credible Threat: 0.4 percent
Graphic Violence Against Defenseless People: 4.8 percent
Glorification of Terrorist Acts: 85.5 percent
Radicalization, Recruitment, Instruction: 9.1 percent
New Zealand Perpetrator Content: 0.6 percent
______
Response to Written Questions Submitted by Hon. Jacky Rosen to
Monika Bickert
Question 1. The challenge for social media platforms prohibiting
certain types of behavior on their sites is creating clear and concise
rules for users to comply. Offensive conduct isn't a static issue, and
as technology has evolved, so have our definitions of what constitutes
abusive behavior such as cyberbullying and misinformation campaigns.
Can you explain to us how your companies come up with rules
regarding hateful speech and how those rules have evolved? What
are your guidelines for determining when charged rhetoric
crosses the line into becoming hate speech? For example, how do
you determine if rhetoric is anti-Semitic?
Answer. We do not allow hate speech on Facebook because it creates
an environment of intimidation and exclusion and in some cases may
promote real-world violence.
We define ``hate speech'' as a direct attack on people based on
what we call protected characteristics--race, ethnicity, national
origin, religious affiliation, sexual orientation, caste, sex, gender,
gender identity, and serious disease or disability. We also provide
some protections for immigration status. We define ``attack'' as
violent or dehumanizing speech, statements of inferiority, or calls for
exclusion or segregation. We separate attacks into three tiers of
severity, as described in our published Community Standards. For more
information, please see https://www.facebook
.com/communitystandards/hate_speech.
How closely do you work with outside groups, researchers,
and users to come up with definitions of what constitutes hate
and abusive speech and policies to deal with ambiguous cases?
For instance, have you worked with the Anti-Defamation League
or other groups combating hate when determining guidelines?
Answer. Facebook has partnerships with a broad range of U.S. and
international NGOs, academics, and experts who study organized hate
groups. These academics and experts share information with Facebook on
how organizations are adapting to social media and give feedback on how
Facebook might better tackle these problems.
Question 2. With almost three and a half billion social media users
worldwide--and one million users joining every day--social media
platforms have turned to a mix of machine learning and human moderators
to detect and take down hate speech, terrorist propaganda, cyber-
bullying, and disinformation. Machine learning can be a useful tool in
identifying objectionable content quickly, preventing it from
spreading. However, there are concerns about its ability to understand
the context of text or images, and the length of time it takes to train
systems with new data to recognize objectionable content.
Can you give us an estimate of how many content moderation
decisions are made by your machine learning systems? And can
you provide an estimated error rate for content flagged by
machine learning?
Answer. We don't have an either-or approach to reviewing content.
All content goes through some degree of automated review, and we use
human reviewers to check some content that has been flagged by that
automated review or reported by people that use Facebook. We also use
human reviewers to perform reviews of content that was not flagged or
reported to check the accuracy and efficiency of our automated review
systems. The percentage of content that is reviewed by a human varies
widely depending on the type and context of the content, and we don't
target a specific percentage across all content on Facebook.
Are there instances where machine learning is more effective
in flagging certain content than others? Does the error rate
change significantly from one type of content to another?
Answer. AI tools lend themselves toward identifying certain content
more easily than others. For example, we are better able to enforce our
nudity policies with automated tools than we are hate speech, due to
the linguistic and cultural nuances involved. One area in which we have
made significant progress is the detection of terrorist content. We
proactively detect 99 percent of the ISIS-and Al Qaeda-related content
that we remove before someone reports it. And we are committed to
continuing to improve our technology across different types of content.
______
Response to Written Questions Submitted by Hon. Shelley Moore Capito
to Nick Pickles
Question 1. I applaud Twitter's engagement and collaboration with
the Global Internet Forum to Counter Terrorism (GIFCT). As you
mentioned in your testimony, Twitter has already started partnering
with smaller tech companies to share best practices and continues to
partner with additional companies. What are some of the smaller
community groups Twitter has been working with and what are some of the
best practices you plan on sharing for combatting extremism?
Answer. Collaboration with our industry peers and civil society is
critically important to addressing common threats from terrorism
globally. In June 2017, we launched the Global Internet Forum to
Counter Terrorism (the ``GIFCT''), a partnership among Twitter,
YouTube, Facebook, and Microsoft.
The GIFCT facilitates, among other things: information sharing;
technical cooperation; and, research collaboration, including with
academic institutions. In September 2017, the members of the GIFCT
announced a significant financial commitment to support research on
terrorist abuse of the Internet and how governments, tech companies,
and civil society can respond effectively. Our goal is to establish a
network of experts that can develop platform-agnostic research
questions and analysis that consider a range of geopolitical contexts.
Technological collaboration is a key part of GIFCT's work. In the
first two years of GIFCT, two projects have provided technical
resources to support the work of members and smaller companies to
remove terrorist content.
First, the shared industry database of ``hashes''--unique digital
``fingerprints''--for violent terrorist propaganda now has more than
100,000 hashes. The database allows a company that discovers terrorist
content on one of its sites to create a digital fingerprint and share
it with the other companies in the forum, who can then use those hashes
to identify such content on their services or platforms, review against
their respective policies and individual rules, and remove matching
content as appropriate or block extremist content before it is posted.
Second, a year ago, Twitter began working with a small group of
companies to test a new collaborative system. Because Twitter does not
allow files other than photos or short videos to be uploaded, one of
the behaviors we saw from those seeking to promote terrorism was to
post links to other services where people could access files, longer
videos, PDFs, and other materials. Our pilot system allows us to alert
other companies when we removed an account or Tweet that linked to
material that promoted terrorism hosted on their service. This
information sharing ensures the hosting companies can monitor and track
similar behavior, taking enforcement action pursuant with their
individual policies. This is not a high-tech approach, but it is simple
and effective, recognizing the resource constraints of smaller
companies.
Based on positive feedback, the partnership has now expanded to 12
companies and we have shared more than 12,000 unique URLs with these
services. Every time a piece of content is removed at source, it means
any link to that source--wherever it is posted--will no longer be
operational.
We are eager to partner with additional companies to expand this
project, and we look forward to building on our existing partnerships
in the future.
Separately, Twitter provides training to civil society groups
around the globe that work on preventing and combating violent
extremism in their communities. These trainings aim to help credible
organizations amplify their voices using Twitter tools and cover a wide
range of best practices, which are summarized in our NGO training
handbook. Twitter does not advise on the specifics of the message, as
these partners are best placed to craft their own authentic content. We
are happy to provide a copy of the handbook upon request.
In addition, Twitter has helped amplify the voices and reach of
these organizations through in-kind assistance in the form of donated
advertising credit, both on Twitter and offline. Recently, for example,
Twitter donated advertising space in New York City to Parents for
Peace, an NGO founded and run by former extremists and families
impacted by extremism which aims to prevent radicalization.
Question 2. This Committee has held a number of hearings on the
rise and importance of artificial intelligence (AI) in today's digital
economy. AI has been invaluable in collecting and sorting massive
amounts of data. In the case of today's hearing, AI has become critical
in order to identify radicalization and terrorist threats. Each company
has identified key tools each company uses in identifying bad actors on
your platforms, but machine learning being one of the most critical.
What factors are given priority when determining radicalized or
terrorist content?
a. You also mention the importance of human expertise in
determining more nuanced cases. When does human expertise step in after
AI has identified or flags content?
b. After content has been flagged for law enforcement involvement,
what is the process that takes place afterward? Does that content get
sent to the FBI and then disseminated to state law enforcement?
Answer. Twitter's philosophy is to take a behavior-led approach,
utilizing a combination of machine learning and human review to
prioritize reports and improve the health of the public conversation.
That is to say, we increasingly look at how accounts behave before we
look at the content they are posting. This is how we seek to scale our
efforts globally and leverage technology even where the language used
is highly context specific. Twitter employs content detection
technology to identify potentially abusive content on the service,
along with allowing users to report content to us either as an
individual or as a bystander.
We suspended more than 1.5 million accounts for violations related
to the promotion of terrorism between August 1, 2015, and December 31,
2018. In 2018, a total of 371,669 accounts were suspended for
violations related to promotion of terrorism. We continue to see more
than 90 percent of these accounts suspended through proactive measures.
The trend we are observing year-over-year is a steady decrease in
terrorist organizations attempting to use our service. This is due to
zero-tolerance policy enforcement that allows us to take swift action
on ban evaders and other identified forms of behavior used by terrorist
entities and their affiliates. In the majority of cases, we take action
at the account creation stage--before the account even Tweets.
The long term challenge for industry is the availability and
sharing of training data for AI and machine learning models. Good
progress has been made in cross-industry collaboration on a number of
fronts, but this is an area where more can be done.
We have well-established relationships with law enforcement
agencies, and we look forward to continued cooperation with them on
these issues, as often only they have access to information critical to
our joint efforts to stop bad faith actors. The threat we face requires
extensive partnership and collaboration with our government partners
and industry peers. We have continuous internal coverage to address
requests from law enforcement around the world and have a portal to
swiftly handle law enforcement requests rendered by appropriate legal
process.
If we have a good faith belief that there is an imminent threat of
death or serious physical harm to an identifiable person or group, and
we have information that we believe is relevant to mitigating that
threat, we share such information with law enforcement. We become aware
of such threats through reports to our content moderation team, or
through an Emergency Request submitted by law enforcement.
Twitter does not have a role in how information is shared between
the FBI and other law enforcement entities.
______
Response to Written Questions Submitted by Hon. Richard Blumenthal to
Nick Pickles
NO HATE Act and Reporting
I have introduced legislation, the Jabara-Heyer NO HATE Act, which
would help states implement and train officers in the National
Incident-Based Reporting Systems. The NO HATE Act would also provide
grants to states to better address hate crimes by training law
enforcement, establish specialized units, create community relations
programs, and run hate crime hotlines.
Question. Do you support the Jabara-Heyer NO HATE Act?
Answer. Twitter believes that successfully combating violent
extremism requires a whole of society approach, including at the
grassroots and community-based level. Twitter has a positive working
relationship with law enforcement agencies, that play a key role in
preventing and addressing violent extremism. Twitter supports efforts
that seek to strengthen the ability of both community groups and law
enforcement agencies to more effectively address violent extremism.
I understand that it has taken some time for Google and Facebook to
establish reliable and timely channels to report threats made on your
platform to the proper authorities. Mr. Slater testified that Google
now has a strong relationship with the Northern California Regional
Intelligence Center, who has been effective at quickly getting reports
of threats into the right hands.
Question 1. Would you support adding measures to the Jabara-Heyer
NO HATE Act to expand the NCRIC model of integrated threat reporting
nationwide?
Answer. Twitter supports efforts that seek to strengthen the
ability of law enforcement agencies to more effectively address violent
extremism within appropriate legal frameworks.
Question 2. What steps would improve communications channels with
law enforcement to make sure the right information gets into the right
hands quickly?
Answer. Twitter has well-established relationships with law
enforcement agencies, and we look forward to continued cooperation with
them on these issues, as often only they have access to information
critical to our joint efforts to stop bad faith actors. The threat we
face requires extensive partnership and collaboration with our
government partners and industry peers. We have continuous internal
coverage to address reports from law enforcement around the world and
have a portal to swiftly handle law enforcement requests rendered by
appropriate legal process.
We encourage law enforcement and other government agencies to
consider whether it is possible to move more quickly to declassify
information that may be useful to industry in assessing and reacting to
the changing nature of bad actors.
Amplification of 8Chan and Other Hate Sites
We have seen over this year that fringe sites are a breeding ground
for racist and violent hate communities. However, extremists then use
mainstream platforms to recruit and amplify their hate and ideologies
to a larger audience. In particular, the site 8chan has had a repeated
role in multiple mass shootings this year. The perpetrators of
Christchurch mosque shootings, Poway synagogue shooting, and El Paso
massacre each posted manifestos to 8chan before their attacks. It is
also sites such as 8chan that facilitate campaigns of harassment and
terrorism that target the victims of mass shootings, such as the Sandy
Hook families. 8chan is currently offline after webhosting providers
finally cut their ties after the El Paso shootings. However, 8chan's
owner has said that he plans to revive the site as soon as this week.
Question 1. Has your company taken any steps to limit the spread of
8chan content, including the communities that hosted the manifestos of
shooters, on your platforms?
Answer. Twitter is committed to improving our ability to stop the
rapid spread of violent extremist content. We are implementing a number
of actions from lessons we learned from the Christchurch attack.
For example, the distribution of media immediately after the
Christchurch attack was manifestly different from how ISIS and other
terrorist groups had historically operated. These changes in the wider
threat environment require a renewed approach and a focus on immediate
crisis response.
In the immediate hours after the Christchurch shootings, an array
of individuals sought continuously to re-upload the content created by
the attacker, both the video and his manifesto, including this same
content hosted on third party services.
The Twitter rules make clear that we do not allow material to be
shared that threatens violence against an individual or group of
people, the glorification of violence or the promotion of terrorism. We
regard manifestos as falling under this rule.
As such, we will take action to limit the ability of individuals to
share materials, including raw video files and manifestos, wherever
they are posted, including on third party services.
Question 2. Please describe the specific steps you to restrict the
amplification of 8chan and other violent sites on your platforms,
including what sites you have taken action to restrict.
Answer. We are taking a number of steps, many in collaboration with
our GIFCT peers, to tackle the challenge of violent content that may
spread to our platform from another location on the Internet, including
but not limited to 8chan.
In addition to our commitment to the Christchurch Call, Twitter and
other leading websites recently voluntarily committed to the following
five distinct actions:
1. Terms of Service. First, we committed to updating our terms of
use, community standards, codes of conduct, and acceptable use
policies to expressly prohibit the distribution of terrorist
and violent extremist content. We believe this is important to
establish baseline expectations for users and to articulate a
clear basis for removal of this content from our platforms and
services and suspension or closure of accounts distributing
such content.
2. User Reporting of Terrorist and Violent Extremist Content.
Second, we committed to establishing one or more methods within
our online platforms and services for users to report or flag
inappropriate content, including terrorist and violent
extremist content. We will ensure that the reporting mechanisms
are clear, conspicuous, and easy to use, and provide enough
categorical granularity to allow us to prioritize and act
promptly upon notification of terrorist or violent extremist
content.
3. Enhancing Technology. Third, we committed to continuing to invest
in technology that improves our capability to detect and remove
terrorist and violent extremist content online, including the
extension or development of digital fingerprinting and AI-based
technology solutions.
4. Livestreaming. Fourth, we committed to identifying appropriate
checks on livestreaming, aimed at reducing the risk of
disseminating terrorist and violent extremist content online.
These may include enhanced vetting measures (such as streamer
ratings or scores, account activity, or validation processes)
and moderation of certain livestreaming events where
appropriate. Checks on livestreaming will necessarily be
tailored to the context of specific livestreaming services,
including the type of audience, the nature or character of the
livestreaming service, and the likelihood of exploitation.
5. Transparency Reports. Finally, we committed to publishing on a
regular basis transparency reports regarding detection and
removal of terrorist or violent extremist content on our online
platforms and services and ensuring that the data is supported
by a reasonable and explainable methodology.
In addition, all members of the GIFCT committed to the following
four collaborative actions:
1. Share Technology Development. We committed to working
collaboratively across industry, governments, educational
institutions, and NGOs to develop a shared understanding of the
contexts in which terrorist and violent extremist content is
published and to improve technology to detect and remove
terrorist and violent extremist content more effectively and
efficiently. This will include:
Work to create robust shared data sets to accelerate
machine learning and AI and sharing insights and learnings
from the data.
Development of open source or other shared tools to
detect and remove terrorist or violent extremist content.
Enablement of all companies, large and small, to
contribute to the collective effort and to better address
detection and removal of this content on their platforms
and services.
2. Crisis Protocols. We also committed to working collaboratively
across industry, governments, and NGOs to create a protocol for
responding to emerging or active events, on an urgent basis, so
relevant information can be quickly and efficiently shared,
processed, and acted upon by all stakeholders with minimal
delay. This includes the establishment of incident management
teams that coordinate actions and broadly distribute
information that is in the public interest.
3. Education. Third, we committed to working collaboratively across
industry, governments, educational institutions, and NGOs to
help understand and educate the public about terrorist and
extremist violent content online. This education includes
reminding users about how to report or otherwise not contribute
to the spread of this content online.
4. Combating Hate and Bigotry. Finally, we committed to working
collaboratively across industry to attack the root causes of
extremism and hate online. This includes providing greater
support for relevant research--with an emphasis on the impact
of online hate on offline discrimination and violence--and
supporting capacity and capability of NGOs working to challenge
hate and promote pluralism and respect online.
Testing of Consumer Platforms
Question 1. Please describe the process you use to test and
evaluate new consumer facing products, including algorithms designed to
promote forms of engagement. What methods are employed to assess the
impact of these products on individuals and groups, both for an
immediate and medium term response?
Answer. We want Twitter to provide a useful, relevant experience to
all people using our service. With hundreds of millions of Tweets per
day on Twitter, we have invested heavily in building systems that
organize content on Twitter alongside tools for individuals to control
their own experience. At the core of the Twitter service is the
individual's choice of which accounts to follow, and thus, are shown in
their home timeline. We want to help our customers to have an
informative and enjoyable experience on Twitter by doing some of the
work to surface content of interest.
With 335 million people using Twitter every month, in dozens of
languages and countless cultural contexts, we rely upon machine-
learning algorithms to help us organize content. Twitter uses a range
of algorithms and behavioral signals to determine how Tweets are
organized and presented in the home timeline, conversations, and search
based on relevance to individuals. Individuals can control themselves
whether to see their home timeline without any algorithmic processing
or instead with our suggested ranking. We are constantly iterating our
product to provide the best possible experience to all people using our
service.
A wide range of teams are involved in assessing potential product
and policy changes, from a broad range of perspectives. This includes
work to understand how algorithms are functioning.
Question 2. Do you ever identify unintended consequences of such
proposed products and then revise them or decide not to launch?
Answer. Yes. We are constantly iterating our product to provide the
best possible experience to all people using our service. Our teams
implement rigorous processes to think through all aspects of potential
product changes to ensure they respect consumer privacy and further our
goal of fostering healthy public conversation.
Question 3. What testing and measurement methodologies are
routinely used and how are the product evaluation teams selected?
Please submit any criteria you have developed for new or revised data
driven products or applications, including their intended impact,
demographic reach, and revenue potential.
Answer. We use a range of criteria to evaluate the success of our
work. For example, in April 2019, we published a range of metrics that
demonstrate the different ways we seek to measure our progress.
38 percent of abusive content that's enforced is surfaced
proactively to our internal teams for review instead of relying
on external reports from people on Twitter.
16 percent fewer abuse reports after an interaction from an
account the external reporter doesn't follow.
100,000 accounts suspended for creating new accounts after a
previous suspension during January-March 2019--a 45 percent
increase from the same time last year.
60 percent faster response to appeals requests with our new
in-app appeal process.
3 times more abusive accounts suspended within 24 hours
after a report compared to the same time last year.
2.5 times more private information removed with a new,
easier reporting process.
We continue to work with outside partners to develop a framework
for measuring healthy conversation, following an international call for
proposals. This will not be a quick or simple process, but we are
investing in the long-term health of the public conversation online.
______
Response to Written Questions Submitted by Hon. Tom Udall to
Nick Pickles
Question 1. We cannot talk about mass violence without talking
about the social and political climate that is dividing America. Most
recently, content that demonizes and spreads hate against immigrant
communities is proliferating across social media. This content is too
often indistinguishable from social media posts from some elected
representatives. How does your company define hate speech?
Answer. Twitter has a policy against hateful conduct. Under this
policy, people on Twitter are not permitted to promote violence against
or directly attack or threaten other people on the basis of race,
ethnicity, national origin, sexual orientation, gender, gender
identity, religious affiliation, age, disability, or serious disease.
We also do not allow accounts whose primary purpose is inciting harm
toward others on the basis of these categories.
We do not allow individuals to use hateful images or symbols in
their profile image or profile header. Individuals on the platform are
not allowed to use the username, display name, or profile bio to engage
in abusive behavior, such as targeted harassment or expressing hate
toward a person, group, or protected category.
Under this policy, we take action against behavior that targets
individuals or an entire protected category with hateful conduct.
When determining the penalty for violating our hateful conduct
policy, we consider a number of factors including, but not limited to
the severity of the violation and an individual's previous record of
rule violations. For example, we may ask someone to remove the
violating content and serve a period of time in read-only mode before
they can Tweet again. Subsequent violations will lead to longer read-
only periods and may eventually result in permanent account suspension.
If an account is engaging primarily in abusive behavior, or is deemed
to have shared a violent threat, we will permanently suspend the
account upon initial review.
Question 1a. And how do you address situations when content that
meets that definition comes from a political leader?
Answer. When it comes to the actions of world leaders on Twitter,
we recognize that this is largely new ground and with important
implications. We understand the desire for our decisions to be ``yes/
no'' binaries, but it's not that simple. The actions we take and
policies we develop will help set precedents around online speech and
we owe it to the people we serve to be deliberate and considered in
what we do.
Twitter's mission is to provide a forum that enables people to be
informed and to engage their leaders directly. We also have a
responsibility to the people who use Twitter to better explain why we
make the decisions we make, which we will do here.
We assess reported Tweets from world leaders against the Twitter
Rules, which are designed to ensure people can participate in the
public conversation freely and safely. We focus on the language of
reported Tweets and do not attempt to determine all potential
interpretations of the content or its intent.
Direct interactions with fellow public figures, comments on
political issues of the day, or foreign policy saber rattling on
economic or military issues are generally not in violation of the
Twitter Rules. However, if a Tweet from a world leader does violate the
Twitter Rules but there is a clear public interest value to keeping the
Tweet on the service, we may place it behind a notice that provides
context about the violation and allows people to click through should
they wish to see the content. We announced this in June 2019.
Our goal is to enforce our rules judiciously and impartially. In
doing so, we aim to provide more insight into our enforcement decision-
making, to serve public conversation, and protect the public's right to
hear from their leaders and to hold these same leaders to account.
Question 2. Knowing that the problem of extremism and mass violence
extends beyond the screen, I would like you to describe your
partnerships with communities and organizations around the country to
fight against extremism and hate. What are you doing to promote their
voices on your platforms? And what makes them effective?
Answer. Twitter works around the globe to support civil society
voices and promote positive messages. Twitter provides regular
trainings to local, credible groups on five continents on how to
amplify their content using our tools. In addition, we have provided
pro-bono advertising to groups to enable their messages to reach
millions of people. When we at Twitter talk about the health of the
public conversation, we see the principles of civility, empathy, and
mutual respect as foundational to our work. We will not solve problems
by removing content alone. We should not underestimate the power of
open conversation to change minds, perspectives, and behaviors.
Question 3. We are entering another election year and we know that
foreign actors have amplified divisive rhetoric on social media and, in
some cases, orchestrated actual protests. What specific actions are you
taking to prepare for 2020 to prevent Russia and other foreign actors
from trying to inflame racial and political tensions through social
media?
Answer. The public conversation occurring on Twitter is never more
important than during elections, the cornerstone of democracy. Any
attempts to undermine the integrity of our service is antithetical to
our fundamental values and undermines the core tenets of freedom of
expression.
We remain vigilant about malicious foreign efforts to manipulate
and divide people in the United States and throughout the world,
including through the use of foreign disinformation campaigns that rely
in certain instances upon the use of deepfakes. In April 2019, we
issued a new Twitter policy regarding election integrity governing
different categories of manipulative behavior and content related to
elections. First, an individual cannot share false or misleading
information about how to participate in an election. This includes but
is not limited to misleading information about how to vote or register
to vote, requirements for voting, including identification
requirements, and the official announced date or time of an election.
Second, an individual cannot share false or misleading information
intended to intimidate or dissuade voters from participating in an
election. This includes but is not limited to misleading claims that
polling places are closed, that polling has ended, or other misleading
information relating to votes not being counted.
Third, we do not allow misleading claims about police or law
enforcement activity related to polling places or elections, long
lines, equipment problems, voting procedures or techniques that could
dissuade voters from participating in an election, and threats
regarding voting locations. Finally, we do not allow the creation of
fake accounts which misrepresent their affiliation, or share content
that falsely represents its affiliation to a candidate, elected
official, political party, electoral authority, or government entity.
If we see the use of any manipulated content to spread
misinformation in violation of our policies governing election
integrity, we will remove that content.
Additionally, we make available a unique comprehensive archive of
removed Tweets and media associated with suspected state-backed
information operations. Our industry peers, academics and policymakers
can leverage the range of signals we publish including links, media,
and account indicators. The data sets we have published so far include
more than 30 million Tweets and more than one terabyte of media.
Further, information sharing and collaboration are critical to
Twitter's success in preventing hostile foreign actors from disrupting
meaningful political conversations on the service. We have well-
established relationships with law enforcement agencies active in this
arena, including the U.S. Federal Bureau of Investigation's Foreign
Influence Task Force and the U.S. Department of Homeland Security's
Election Security Task Force. We look forward to continued cooperation
with federal, state, and local government agencies on election
integrity issues because in certain circumstances only they have access
to information critical to our joint efforts to stop bad faith actors.
On Election Day in the 2018 U.S. midterms, Twitter participated
virtually in an operations center convened by the U.S. Department of
Homeland Security. The operations center also convened officials from
the U.S. Department of Justice, the Federal Bureau of Investigation,
and the Office of the Director of National Intelligence, in addition to
federal, state, local, and private sector partners. In the lead up to
Election Day, and throughout the course of the day itself, Twitter
remained in constant contact with officials throughout all levels of
government. We plan to do the same in the 2020 U.S. election period.
We also worked in close collaboration with the National Association
of Secretaries of State (NASS) and the National Association of State
Election Directors (NASED). Founded in 1904, NASS is the Nation's
oldest, nonpartisan professional organization for public officials, and
is open to secretaries of states and lieutenant governors in the 50
states, D.C. and territories. In February 2019, Twitter participated in
a panel discussion convened by NASS on the Role of Social Media in
Democracy and their New Voters Forum, broadcast on C-Span.
Question 4. Regarding the shared industry database of hashes linked
to content that promotes terrorism; I would like to understand the
thresholds for including certain content in the database. Who makes the
decision to include content in that database and how is that decision
made? What percent of that database concerns white nationalist or other
domestic extremist content?
Answer. Collaboration with our industry peers and civil society is
critically important to addressing common threats from terrorism
globally. In June 2017, we launched the Global Internet Forum to
Counter Terrorism (the ``GIFCT''), a partnership among Twitter,
YouTube, Facebook, and Microsoft.
The GIFCT facilitates, among other things: information sharing;
technical cooperation; and, research collaboration, including with
academic institutions. In September 2017, the members of the GIFCT
announced a significant financial commitment to support research on
terrorist abuse of the Internet and how governments, tech companies,
and civil society can respond effectively. Our goal is to establish a
network of experts that can develop platform-agnostic research
questions and analysis that consider a range of geopolitical contexts.
Technological collaboration is a key part of GIFCT's work. In the
first two years of GIFCT, two projects have provided technical
resources to support the work of members and smaller companies to
remove terrorist content.
As reported in GIFCT's first transparency report, published in July
2019, the GIFCT Hash Sharing Consortium has reached over 200,000 unique
pieces of terrorist content. Companies often have slightly different
definitions on ``terrorism'' and ``terrorist content.'' The taxonomy
includes the following labels that are applied to the content when a
company ads hashes to the shared database.
Imminent Credible Threat (ICT): A public posting of a
specific, imminent, credible threat of violence toward non-
combatants and/or civilian infrastructure.
Graphic Violence Against Defenseless People: The murder,
execution, rape, torture, or infliction of serious bodily harm
on defenseless people (prisoner exploitation, obvious non-
combatants being targeted).
Glorification of Terrorist Acts (GTA): Content that
glorifies, praises, condones, or celebrates attacks after the
fact.
Recruitment and Instruction (R&I): Materials that seek to
recruit followers, give guidance, or instruct them
operationally.
New Zealand Perpetrator Content: The GIFCT set a new
precedent in the wake of the New Zealand terrorist attack. Due
to the virality and cross-platform spread of the attacker's
manifesto and attack video, and because New Zealand authorities
deemed all manifesto and attack video content illegal, the
GIFCT created a crisis bank to mitigate the spread of this
content.
The following shows the breakdown of how much content has been
ingested into the shared database of hashes based on the above
taxonomy.
Imminent Credible Threat: 0.4 percent
Graphic Violence Against Defenseless People: 4.8 percent
Glorification of Terrorist Acts: 85.5 percent
Radicalization, Recruitment, Instruction: 9.1 percent
New Zealand Perpetrator Content: 0.6 percent
More information can be found here: https://gifct.org/transparency/
In addition to these efforts by the GIFCT, in 2018 Twitter began
working with a small group of companies to test a new collaborative
system. Because Twitter does not allow files other than photos or short
videos to be uploaded, one of the behaviors we saw from those seeking
to promote terrorism was to post links to other services where people
could access files, longer videos, PDFs, and other materials. Our pilot
system allows us to alert other companies when we removed an account or
Tweet that linked to material that promoted terrorism hosted on their
service. This information sharing ensures the hosting companies can
monitor and track similar behavior, taking enforcement action pursuant
with their individual policies. This is not a high-tech approach, but
it is simple and effective and recognizes the resource constraints of
smaller companies.
Based on positive feedback, the partnership has now expanded to 12
companies with which we have shared more than 12,000 unique URLs. Every
time a piece of content is removed at source, it means any link to that
source--wherever it is posted--will no longer be operational.
We are eager to partner with additional companies to expand this
project, and we look forward to building on our existing partnerships
in the future.
______
Response to Written Questions Submitted by Hon. Jacky Rosen to
Nick Pickles
Question 1. The challenge for social media platforms prohibiting
certain types of behavior on their sites is creating clear and concise
rules for users to comply. Offensive conduct isn't a static issue, and
as technology has evolved, so have our definitions of what constitutes
abusive behavior such as cyberbullying and misinformation campaigns.
Can you explain to us how your companies come up with rules
regarding hateful speech and how those rules have evolved? What
are your guidelines for determining when charged rhetoric
crosses the line into becoming hate speech? For example, how do
you determine if rhetoric is anti-Semitic?
How closely do you work with outside groups, researchers,
and users to come up with definitions of what constitutes hate
and abusive speech and policies to deal with ambiguous cases?
For instance, have you worked with the Anti-Defamation League
or other groups combating hate when determining guidelines?
Answer. We draft and enforce the Twitter Rules to keep people safe
on our service, and to protect the health of the public conversation.
The Twitter Rules apply to everyone. In general, we create our rules
with a rigorous policy development process; it involves in-depth
research, analysis of the behavior of individuals on Twitter,
historical violation patterns, and immersion in academic material.
We appreciate these issues are complex and we value the input of
external voices in developing our approach. As part of our internal
development process, we consult with a wide range of stakeholders and
we focus on the risk of gaming, subverting, or otherwise abusing our
policies and product changes. We supplement this work with
conversations with outside experts and organizations where appropriate.
For example, many scholars have examined the relationship between
dehumanization and violence. In September 2018, we tried something new
by asking the public for feedback on a policy before it became part of
the Twitter Rules. Our goal was to test a new format for policy
development whereby the individuals who use Twitter have a role in
directly shaping our efforts to protect them. We wanted to expand our
hateful conduct policy to include content that dehumanizes others based
on their membership in an identifiable group, even when the material
does not include a direct target.
We asked for feedback to ensure we considered a wide range of
perspectives and to hear directly from different communities and
cultures who use Twitter around the globe. In two weeks, we received
more than 8,000 responses from people located in more than 30
countries.
Following our review of public comments, in July 2019, we expanded
our rules against hateful conduct to include language that dehumanizes
others on the basis of religion.
We also work with outside groups, including those represented on
the Twitter Trust and Safety Council, of which the Anti-Defamation
League is a member. These groups are able to provide input on a range
of policy and product approaches, both as part of the council and in
direct conversations with teams at Twitter.
Question 2. With almost three and a half billion social media users
worldwide--and one million users joining every day--social media
platforms have turned to a mix of machine learning and human moderators
to detect and take down hate speech, terrorist propaganda, cyber-
bullying, and disinformation. Machine learning can be a useful tool in
identifying objectionable content quickly, preventing it from
spreading. However, there are concerns about its ability to understand
the context of text or images, and the length of time it takes to train
systems with new data to recognize objectionable content.
Can you give us an estimate of how many content moderation
decisions are made by your machine learning systems? And can
you provide an estimated error rate for content flagged by
machine learning?
Are there instances where machine learning is more effective
in flagging certain content than others? Does the error rate
change significantly from one type of content to another?
Answer. Twitter's philosophy is to take a behavior-led approach,
utilizing a combination of machine learning and human review to
prioritize reports and improve the health of the public conversation.
That is to say, we increasingly look at how accounts behave before we
look at the content they are posting. This is how we can scale our
efforts globally and leverage technology even where the language used
is highly context specific. Twitter employs extensive content detection
technology to identify potentially abusive content on the service,
along with allowing users to report content to us either as an
individual or as a bystander.
For abuse, this strategy has allowed us to take three times the
amount of enforcement actions on abuse within 24 hours than this time
last year. We now proactively surface over 50 percent of abusive
content we remove using our technology compared to only 20 percent a
year ago. This reduces the burden on individuals to report content to
us. Since we started using machine learning three years ago to reduce
the visibility on abusive content:
80 percent of all replies that are removed were already less
visible;
Abuse reports themselves have been reduced by 7.6 percent;
The most visible replies receive 45 percent less abuse
reports;
100,000 accounts were suspended for creating new accounts
after a previous suspension during January through March 2019--
a 45 percent increase from the same time last year;
60 percent faster response to appeals requests with our new
in-app appeal process;
3 times more abusive accounts suspended within 24 hours
after a report compared to the same time last year; and
2.5 times more private information removed with a new,
easier reporting process.
Machine learning plays an important role across a multitude of our
product surface areas. Making Twitter healthier also requires making
the way we employ machine learning more fair, accountable, and
transparent.
In many areas, machine learning is not sufficiently accurate to
utilize in content removal decisions. For example, machine learning is
not well suited to address sarcasm, innuendo, satire, or distinguish
news coverage from propaganda broadcasts. A human role in these content
decisions is essential to protect vulnerable groups, public debate, and
free expression.
As machine learning evolves, there are some challenges that are
more difficult than others. Often this is tied to the availability of
training data for models. Rare events create fewer opportunities to
obtain training data that hinder future efforts to identity similar
incidents, while those that happen frequently offer greater data to
train models. This is an area where further industry collaboration is
essential, as the availability of training data is fundamental to the
ability of companies to develop machine learning models that are better
able to identify and remove different types of problematic content.
We welcome efforts to increase collaboration in this area, both
with industry and governments. Increased efforts to foster technical
collaboration will enable us to build upon work already done, and
policymakers can support these efforts with greater legal protections
for companies sharing content of this nature.
______
Response to Written Questions Submitted by Hon. Shelley Moore Capito
to Derek Slater
Question 1. I represent a state that has one of--if not--the
highest rates of drug overdoses deaths, the vast majority of which are
due to opioids. I appreciate Google's recent announcement to launch new
tools to help connect individuals recovering from opioid addiction with
treatment resources. Does Google plan on doing similar initiatives to
combat radicalization, like providing resources to mental health
services?
Answer. We hope that Google's growing, multi-faceted efforts to
address the opioids epidemic can help West Virginia and families
nationwide grapple with substance use issues. These efforts include
prevention, like making it easier for people to find medication
disposal sites in their communities, to the treatment resources you
mentioned, and providing uplifting resources for people in long term
recovery for substance abuse.
Like substance use, mental health treatment generally is complex
and stigmatized, with 50 percent of individuals not receiving needed
treatment for depression and 1 in 5 not receiving needed treatment for
PTSD. In partnership with the National Alliance on Mental Illness
(NAMI), Google enabled people searching for information on mental
health conditions including depression and PTSD to understand the
likelihood of having these conditions by taking brief, clinically
validated surveys (PHQ-9 for depression and PC-PTSC-5 for PTSD). We
also help users to find resources to take action toward recovery
including directing them to instant access to the National Suicide
Hotline by phone or chat if they are having suicidal thoughts.
Question 2. This Committee has held a number of hearings on the
rise and importance of artificial intelligence (AI) in today's digital
economy. AI has been invaluable in collecting and sorting massive
amounts of data. In the case of today's hearing, AI has become critical
in order to identify radicalization and terrorist threats. Each company
has identified key tools each company uses in identifying bad actors on
your platforms, but machine learning being one of the most critical.
What factors are given priority when determining radicalized or
terrorist content?
a. You also mention the importance of human expertise in
determining more nuanced cases. When does human expertise step in after
AI has identified or flags content?
b. After content has been flagged for law enforcement involvement,
what is the process that takes place afterward? Does that content get
sent to the FBI and then disseminated to state law enforcement?
Answer. We use a mix of people and technology to address terrorist
and violent extremist content on our platforms. We apply our most
advanced machine learning research to train new ``content classifiers''
to help us more quickly identify and remove extremist and terrorism-
related content. This can be challenging: a video of a terrorist attack
may be informative reporting by a news agency, or glorification of
violence if uploaded in a different context by a different user. Human
reviewers play a key role in making nuanced decisions about the line
between violent propaganda and newsworthy speech. Our efforts to
address this content have also included consultation with dozens of
experts in subjects like terrorism, violent extremism, civil rights,
and free speech.
The Stored Communications Act allows Google and other service
providers to voluntarily disclose user data to governmental entities in
emergency circumstances where the provider has a good faith belief that
disclosing the information will prevent loss of life or serious
physical injury to a person. When we have a good faith belief that
there is a threat to life or serious bodily harm made on our platform
in the United States, the Google CyberCrime Investigation Group (CCIG)
will report it to the Northern California Regional Intelligence Center
(NCRIC). In turn, NCRIC quickly gets the report into the hands of
officers to respond. Our team is staffed on a 24/7/365 basis to respond
to these emergency disclosure requests (EDRs).
In other cases, law enforcement agencies at the Federal and state
levels make emergency requests to Google for user data in situations
involving danger of death or serious physical injury to any person. As
illustrated in our transparency report covering government requests for
user data, the number of EDRs submitted from agencies in the U.S.
almost doubled from 2017 to 2018. We have grown our teams to
accommodate this growing volume and to ensure we can quickly respond to
emergency situations that implicate public safety.
______
Response to Written Questions Submitted by Hon. Richard Blumenthal to
Derek Slater
NO HATE Act and Reporting
I have introduced legislation, the Jabara-Heyer NO HATE Act, which
would help states implement and train officers in the National
Incident-Based Reporting Systems. The NO HATE Act would also provide
grants to states to better address hate crimes by training law
enforcement, establish specialized units, create community relations
programs, and run hate crime hotlines.
Question 1. Do you support the Jabara-Heyer NO HATE Act? I
understand that it has taken some time for Google and Facebook to
establish reliable and timely channels to report threats made on your
platform to the proper authorities. Mr. Slater, you testified that you
now have a strong relationship with the Northern California Regional
Intelligence Center, who has been effective at quickly getting reports
of threats into the right hands.
Question 2. Would you support adding measures to the Jabara-Heyer
NO HATE Act to expand the NCRIC model of integrated threat-reporting
nationwide?
Question 3. What steps would improve communications channels with
law enforcement to make sure the right information gets into the right
hands quickly?
Answer. Addressing questions 1-3: We appreciate your work in this
area and share your interest in getting threat information to law
enforcement so they can take immediate action. We have worked with law
enforcement to create efficient processes with NCRIC and think we have
all made important strides together. While we have not yet taken a
position the Jabara-Heyer NO HATE Act, we support the intent and we are
interested in learning more about how it could be amended to include an
expansion of the NCRIC model. Google is supportive of efforts to expand
the approach spearheaded by NCRIC to allow for greater geographic
coverage, handle overflow work, and to make the process more robust. It
is also important that NCRIC continue to receive the necessary funding
to continue building its capacity and effectiveness.
Amplification of 8Chan and Other Hate Sites
We have seen over this year that fringe sites are a breeding ground
for racist and violent hate communities. However, extremists then use
mainstream platforms to recruit and amplify their hate and ideologies
to a larger audience. In particular, the site 8chan has had a repeated
role in multiple mass shootings this year. The perpetrators of
Christchurch mosque shootings, Poway synagogue shooting, and El Paso
massacre each posted manifestos to 8chan before their attacks. It is
also sites such as 8chan that facilitate campaigns of harassment and
terrorism that target the victims of mass shootings, such as the Sandy
Hook families. 8chan is currently offline after webhosting providers
finally cut their ties after the El Paso shootings. However, 8chan's
owner has said that he plans to revive the site as soon as this week.
Question 4. Has your company taken any steps to limit the spread of
8chan content, including the communities that hosted the manifestos of
shooters, on your platforms?
Question 5. Please describe the specific steps you to restrict the
amplification of 8chan and other violent sites on your platforms,
including what sites you have taken action to restrict.
Answer. Answering questions 4-5: We take a number of steps to
address harmful content across our platforms, regardless of the source,
including the following:
Removing content from hosted platforms: Hate speech is not
allowed on YouTube and other Google hosted platforms, and we
are bringing significant attention to detection and removal of
hateful content on our platforms. We have a number of policies
that work together to disallow hateful content--our hate speech
policy, our harassment policy which disallows malicious attacks
against individuals, and our general policy that disallows
incitement to violence. Our policies would be applicable to
``manifestos'' from those who commit violent acts.
Removing financial incentives: In addition, our longstanding
policies prevent ads from running on violative content,
including hate speech. On YouTube, channels that have shown a
history of brushing up against our hate speech policies (even
if they haven't crossed the line), will be suspended from our
YouTube Partner program.
Reducing recommendations of borderline content: Besides removal
of content on YouTube, we also take other steps to curb
potentially harmful content. Several months ago, we began
reducing visibility of borderline content (which comes close to
but doesn't quite violate our rules) or content that can
misinform users in harmful ways. This will be a gradual change,
but this approach is already starting to bear fruit. In the
U.S. we've seen a 50 percent drop of views from recommendations
to this type of content, meaning quality content has more of a
chance to shine.
COPPA Settlement and Children's Privacy
Question 6. How will Google implement its new promises under its
COPPA-related consent decree regarding data collection on content in
which it has a direct financial and curatorial relationship, including
Google Preferred?
Answer. We are making a number of changes to how we treat data on
children's content on YouTube to address the concerns reflected in the
FTC's investigation.
In order to identify this category of content, we will be requiring
creators to tell us when their content is made for kids. We will also
use machine learning to find videos that clearly target young
audiences, for example those that have an emphasis on kids characters,
themes, toys, or games and use it as a signal that will help us define
the child directed content at YouTube.
We will treat data from anyone watching made for kids content on
YouTube as coming from a child, regardless of the age of the user. This
means that we will limit data collection and use on videos made for
kids only to what is needed to support the operation of the service. We
will stop serving personalized ads on this content entirely, and some
features will no longer be available on this type of content, like
comments and notifications.
Advertising inventory on content made for kids will be available
for certain reservation ad buys, such as custom packs and sponsorships,
but we currently don't have plans to include it in reservation buys
like the Google Preferred Lineup and Breakout Video packages.
Question 7[1]. When Google says it will only collect data on videos
made for kids for ``what is needed to support the operation of the
service,'' what specifically will it gather and how will it be used?
Answer. On content that is identified as made for kids, we are
putting in place limitations on the data we collect and use as
described above. This means:
Limiting the collection of personal information like name,
address, or contact information. We will collect user activity
information such as when users watch a video or click on an
advertisement, and information about their device such as IP
address.
Disabling features including comments, sharing features,
notification requests, and add-to-playlist features. Actions
such as Subscribe or Like may still be enabled for users
logged-in to their Google accounts, but would have limited
functionality.
Prohibiting the serving of personalized advertising or
remarketing ads.
Using data only to support the operation of the service, which
includes: performing actions you request (e.g., playing a video);
respecting your settings (e.g., preferred language/country which allows
us to surface e.g., videos in French for users in France); preventing
fraud and abuse; personalizing users' experience on YouTube (e.g.,
recommending relevant content based on watch history), depending on
their account settings; and serving contextual advertisements.
Question 7[2]. What role, if any, will the Google Marketing
Platform play in this regard?
Answer. The changes that we're implementing to how we collect and
use data on content identified as made for kids, as described above,
will apply across our ad products that may be used to serve advertising
on YouTube.
Question 8. Will Google expand its new kid privacy safeguards to
its other child directed services, such as the Play Store?
Answer. We have already made a number of improvements this year as
part of the revamp of the Designed for Families Program, which launched
in May 2019. These policy changes build on our existing efforts to help
ensure that apps for children have appropriate content, show suitable
ads, and handle personally identifiable information correctly; they
also reduce the chance that apps not intended for children could
unintentionally attract them. Developers who have children as part of
their target audience must meet stringent policy requirements in their
apps concerning both content safety, ads appropriateness, and privacy
protections. We will also be double checking apps to make sure that
they are not seeking to attract children but attempting to avoid these
requirements. We take action when we identify developers who do not
fulfill these policy requirements and remove their content from the
Play store when appropriate.
Question 9. Will the new fund for children's content creators
principally fund non-commercial and ad free content aimed for kids,
families and for education?
Answer. We are currently determining the criteria for distributing
the funds, working with children's media experts to ensure we are
funding high-quality content with a global reach. We will look to fund
enriching content similar to the type of content we featured in our
Creating for Families Field Guide. When complete, the content will
appear on our existing platforms, which are supported through
contextual advertising so that they can remain free and accessible for
all families, regardless of ability to pay.
Question 10. How will Google deal with influencer and unboxing
videos aimed at children?
Answer. We do not currently allow paid promotional content on
YouTube Kids. See here and here for more details because there is not
yet an industry consensus on what an appropriate disclosure for such an
audience would require. On the main YouTube service, however, which is
intended for a wider audience, paid promotional videos are generally
permitted. Such promotions must be disclosed as paid promotions and
must abide by all YouTube ads policies (including restrictions around
targeting kids under 13). These policies are outlined in our Help
Center here. We are also seeking guidance among industry partners and
regulators to determine what an appropriate disclosure for kid
appealing content might look like in this context.
There is a wide range of content that is sometimes called
`unboxing'--ranging from videos of kids playing with toys (or indeed
even cardboard boxes, making them into pirate ships or castles) to an
adult showing off a Chewbacca mask. Therefore not all unboxing is
necessarily commercial, not all of it is compensated, and the audiences
are diverse. When creators of unboxing videos are compensated in any
way by an advertiser, Google would consider the video to be a paid
promotion and the creator would be required to mark it as such under
our policies. Unboxing videos that are not motivated by any
consideration or connection with an advertiser are not commercial, and
are outside of the scope of such policies. We are currently working
with experts to understand how to treat this content and whether we
need to update our policies or practices.
Question 11. Are the YouTube changes all global?
Answer. Yes, the changes we are making consistent with the FTC
settlement will be implemented globally.
Testing of Consumer Platforms
Question 12. Please describe the process you use to test and
evaluate new consumer facing products, including algorithms designed to
promote forms of engagement. What methods are employed to assess the
impact of these products on individuals and groups, both for an
immediate and medium term response?
Answer. Google's evaluation and testing processes reflect the
diversity of our consumer-facing products and offerings, ranging from
laptops to operating systems or search engines-typically involving
multiple rounds of testing, experiments, and reviews with product,
engineering, trust-and-safety, legal, policy, and privacy experts.
Those reviews are designed to verify that the product functions as
expected, to explore unintended consequences, and to address possible
risks.
For instance, to help ensure Search algorithms meet high standards
of relevance and quality, we have a rigorous process that involves both
live tests and thousands of trained external Search Quality Raters from
around the world. Search Quality Raters follow strict guidelines that
define our goals for Search algorithms and are publicly available for
anyone to see. The ratings provided by Search Quality Raters help us
benchmark the quality of our results so that we can meet a high bar for
users of Google Search all around the world.
In addition to the Search quality tests, we conduct live traffic
experiments to see how real people interact with a feature, before
launching it to everyone. Results from these experiments undergo a
review by experienced engineers and search analysts, as well as other
legal and privacy experts, who then determine whether the change is
approved to launch. In 2018, we ran over 654,680 experiments, with
trained external Search Raters and live tests, resulting in more than
3,234 improvements to Search. For more information on this process and
our methods, please refer to www.google.com/search/howsearchworks.
YouTube also conducts robust evaluation and testing processes ahead
of launching new features or policies. Given our scale, it's important
that we roll out new offerings and product changes incrementally so we
can monitor performance and feedback from users. Creators can also
submit feedback directly through YouTube Studio.
YouTube's development of platform policies provides another
example. At YouTube, we have developed robust ``Community Guidelines''
that set the rules of the road for what we don't allow. We are
constantly evaluating these policies and their enforcement,
incorporating feedback from experts and trends we see on the platform;
we made 30 updates to our policies in the last year alone. For
instance, we strengthened our hate speech policy in June by
specifically prohibiting videos alleging that a group is superior in
order to justify discrimination, segregation or exclusion based on
qualities like age, gender, race, caste, religion, sexual orientation
or veteran status. When evaluating our approach towards hateful content
we consulted with dozens of experts in subjects like violent extremism,
civil rights, and free speech.
In addition to our teams focused on policy development and
enforcement, we've also established an intel desk to help us detect
emerging trends in how people and organizations may try to misuse our
platform. It identifies new trends in harmful content by synthesizing
leads from third party intel vendors, internal trend data, social
listening, and other relevant inputs.
Question 13. Do you ever identify unintended consequences of such
proposed products and then revise them or decide not to launch?
Answer. Yes. Where our reviews identify significant unintended
consequences for users or society that we cannot adequately resolve in
time for our planned launch date, we may postpone temporarily or
indefinitely. This is a normal part of doing business for each of the
products and services that we operate.
Question 14. What testing and measurement methodologies are
routinely used and how are the product evaluation teams selected?
Please submit any criteria you have developed for new or revised data
driven products or applications, including their intended impact,
demographic reach, and revenue potential.
Answer. Our evaluation criteria for new products or services are a
direct function of their intended goals. For instance, our hardware
product launch process will involve in-depth reviews for the quality,
durability, and resilience of each individual component, in line with
the best industry standards, whereas the same processes would not make
sense for software, storage, or computing services. When it comes to
our ranking algorithms, as mentioned above, we use a number of methods
including side-by-side experiments with trained raters and live
experiments to test whether a given change to our products represents a
tangible improvement for our users. Our criteria are both quantitative
and qualitative, aimed at measuring not just the changes in user
behavior, but also whether a given change advances the goals outlined
in our Search Quality Rater Guidelines.
______
Response to Written Questions Submitted by Hon. Tom Udall to
Derek Slater
Question 1. We cannot talk about mass violence without talking
about the social and political climate that is dividing America. Most
recently, content that demonizes and spreads hate against immigrant
communities is proliferating across social media. This content is too
often indistinguishable from social media posts from some elected
representatives. How does your company define hate speech? And how do
you address situations when content that meets that definition comes
from a political leader?
Answer. Hate speech is not allowed on YouTube and other Google
hosted platforms, and we are bringing significant attention to
detection and removal of hateful content on our platforms. We enforce
those policies regardless of a speaker's political persuasion.
We have a number of policies that work together to disallow hateful
content--our hate speech policy, our harassment policy which disallows
malicious attacks against individuals, and our general policy that
disallows incitement to violence. For instance, our hate speech policy
on YouTube specifically prohibits: ``Content that encourages or
glorifies violence against individuals or groups, or whose primary
purpose is to incite hate against individual or group based on
attributes including age, ethnicity, disability, gender, nationality,
race, immigration status, religion, sex, sexual orientation, and
veteran status.'' It's important to note that YouTube takes action
against hateful content, not based on speakers.
We don't allow content that dehumanizes individuals or groups with
these attributes, claims they are physically or mentally inferior, or
praises or glorifies violence against them. We also don't allow use of
stereotypes that incite or promote hatred based on these attributes, or
racial, ethnic, religious, or other slurs where the primary purpose is
to promote hatred. Our policy prohibits content that alleges the
superiority of a group over those with any of the attributes noted
above to justify violence, discrimination, segregation, or exclusion.
We also do not allow content that denies that a well-documented,
violent event took place.
In enforcing our hate speech policy, we consider the purpose of the
video. If users are posting educational, documentary, scientific, or
artistic content related to hate speech, we encourage them to be
mindful to provide enough information so viewers understand the
context, such as through an introduction, voiceover commentary, or text
overlays, as well as through a clear title and description. We give
users tips and tools for adding context on YouTube.
Question 2. Knowing that the problem of extremism and mass violence
extends beyond the screen, I would like you to describe your
partnerships with communities and organizations around the country to
fight against extremism and hate. What are you doing to promote their
voices on your platforms? And what makes them effective?
Answer. In 2016, we launched YouTube Creators for Change, an
initiative dedicated to amplifying the voices of role models who are
tackling difficult social issues with their channels. From combating
hate speech, to countering xenophobia and extremism, to simply making
the case for greater tolerance and empathy toward others, these
creators are helping to foster productive conversations around tough
issues and make a positive impact on the world.
As part of their commitment to the program, Creators for Change
Ambassadors and Fellows receive mentorship and promotional support to
aid the creation of their Impact Projects--films that tackle a wide
range of topics, from self-acceptance and showing kindness to others,
to celebrating cultures and advocating for global empathy.
Creators for Change is a global program that thrives through its
many local chapters. From providing education on the dangers of fake
news, to helping create safe spaces for making content that addresses
hate speech, these chapters empower thousands of young people to drive
positive social change across Europe, the Middle East and the Asia-
Pacific region.
We have produced annual reports detailing information about the
Creator Ambassadors and the billions of views their content has
generated, see https://www.youtube.com/creators-for-change/.
In addition, because technology alone is not a silver bullet, we
have greatly increased the number of independent experts in YouTube's
Trusted Flagger program. Machines can help identify problematic videos,
but human experts still play a role in nuanced decisions about the line
between violent propaganda and religious or newsworthy speech. While
many user flags can be inaccurate, Trusted Flagger reports are accurate
over 90 percent of the time and help us scale our efforts and identify
emerging areas of concern. We will expand this program by adding 50
expert NGOs to the 63 organizations who are already part of the
program, and we will support them with operational grants. This allows
us to benefit from the expertise of specialized organizations working
on issues like hate speech, self-harm, and terrorism. We will also
expand our work with counter-extremist groups to help identify content
that may be being used to radicalize and recruit extremists.
Finally, we would also note that Jigsaw, a project of Google's
parent company Alphabet, created the Redirect Method--a way to use
AdWords targeting tools and curated YouTube videos uploaded by people
all around the world to confront online radicalization. For example, it
focuses on the slice of ISIS' audience that is most susceptible to its
messaging, and redirects them towards curated YouTube videos debunking
ISIS recruiting themes. This open methodology was developed from
interviews with ISIS defectors, respects users' privacy and can be
deployed to tackle other types of violent recruiting discourses online.
Question 3. We are entering another election year and we know that
foreign actors have amplified divisive rhetoric on social media and, in
some cases, orchestrated actual protests. What specific actions are you
taking to prepare for 2020 to prevent Russia and other foreign actors
from trying to inflame racial and political tensions through social
media?
Answer. Although we found limited activity on our platforms in 2016
and during the 2018 midterms, we understand the existence of this
threat, and take the integrity of our elections very seriously. We have
a team dedicated to ensuring the integrity of election-related content
and ads across our platforms, including combating potential foreign
influence.
We've taken various key steps to combat election interference. For
example, we have policies that prohibit misrepresentation and other
forms of abuse, and we have devoted significant resources to enforcing
our policies, and we have conducted vulnerability testing across key
products and made several changes to safeguard our products from being
used to confuse voters, such as through manipulation of search features
(e.g., WebAnswers, Knowledge Panels). We also worked closely with
others in industry and government election integrity task forces to be
able to identify threats and respond quickly.
We are approaching the 2020 election with vigilance and commitment.
We expect to once again establish a war room with dedicated full-time
employees to provide 24/7 monitoring and rapid escalation of any issues
in the days before and after the elections. We continue to provide
regular updates on our work to that end.
This is in addition to our broader efforts to ensure the integrity
of our elections. For example, we've trained 1,000 campaign
professionals last year about online security, and we've released
``Protect Your Election,'' a suite of digital tools designed to help
election websites and political campaigns protect themselves from
digital attacks.
Question 4. Regarding the shared industry database of hashes linked
to content that promotes terrorism; I would like to understand the
thresholds for including certain content in the database. Who makes the
decision to include content in that database and how is that decision
made? What percent of that database concerns white nationalist or other
domestic extremist content?
Answer. In working together to build technological solutions that
will prevent and disrupt the spread of terrorist content online, the
largest cross-platform advancement supported by the Global Internet
Forum to Counter Terrorism (GIFCT) has been the creation of a Hash
Sharing Consortium. The consortium shares ``hashes'' (or digital
fingerprints) of known terrorist images and videos. The image or video
is ``hashed'' in its raw form and is not linked to any source original
platform or user data. Hashes appear as a numerical representation of
the original content and can't be reverse engineered to create the
image and/or video. A platform needs to find a match with a given hash
on their platform in order to see what the hash corresponds with.
It is up to each consortium member how they utilize the database
and how they contribute to it, depending on their own terms of service,
how their platform operates, and how they utilize technical and human
capacities.
Companies often have slightly different definitions for
``terrorism'' and ``terrorist content''. For the purposes of the hash
sharing database, and to find an agreed upon common ground, founding
companies in 2017 decided to define terrorist content based on content
relating to organizations on the UN Terrorist Sanctions lists.
Companies also agreed upon a basic taxonomy around the type of content
ingested relating to these listed organizations. The taxonomy includes
labels that are applied to the content when a company adds hashes to
the shared database.
GIFCT released its first transparency report in 2019; it includes
the specific taxonomy used by the Hash Sharing Consortium and the
respective percentages of each category, available at: gifct.org/
transparency.
______
Response to Written Questions Submitted by Hon. Jacky Rosen to
Derek Slater
Question 1. The challenge for social media platforms prohibiting
certain types of behavior on their sites is creating clear and concise
rules for users to comply. Offensive conduct isn't a static issue, and
as technology has evolved, so have our definitions of what constitutes
abusive behavior such as cyberbullying and misinformation campaigns.
Can you explain to us how your companies come up with rules
regarding hateful speech and how those rules have evolved? What
are your guidelines for determining when charged rhetoric
crosses the line into becoming hate speech? For example, how do
you determine if rhetoric is anti-Semitic?
Answer. We are investing in the policies, resources and products
needed to live up to our responsibility and protect the YouTube
community from harmful content. Over the past few years, this work has
focused on four pillars: removing violative content, raising up
authoritative content, reducing the spread of borderline content and
rewarding trusted creators. Thanks to these investments, videos that
violate our policies are removed faster than ever and users are seeing
less borderline content and harmful misinformation. As we do this,
we're partnering closely with lawmakers and civil society around the
globe to limit the spread of violent extremist content online.
We review our policies on an ongoing basis to make sure we are
drawing the line in the right place: in 2018 alone, we made more than
30 policy updates. One of the most complex and constantly evolving
areas we deal with is hate speech.
YouTube has always had rules of the road, including a longstanding
policy against hate speech, but we've been taking a close look at our
approach towards hateful content in consultation with dozens of experts
in subjects like violent extremism, supremacism, civil rights, and free
speech. Based on those learnings, we made several updates.
In June 2019, we updated YouTube's hate speech policy by
specifically prohibiting videos alleging that a group is superior in
order to justify discrimination, segregation or exclusion based on
qualities like age, gender, race, caste, religion, sexual orientation
or veteran status. This would include, for example, videos that promote
or glorify Nazi ideology, which is inherently discriminatory.
Additionally, we will remove content denying that well-documented
violent events, like the Holocaust or the shooting at Sandy Hook
Elementary, took place.
Human reviewers remain essential to both removing content and
training machine learning systems because human judgment is critical to
making contextualized decisions on content. The total number of people
across Google working to address content that might violate our
policies is over 10,000. Our trust and safety teams manually review
millions of videos, helping train our machine-learning technology to
identify similar videos in the future.
How closely do you work with outside groups, researchers,
and users to come up with definitions of what constitutes hate
and abusive speech and policies to deal with ambiguous cases?
For instance, have you worked with the Anti-Defamation League
or other groups combating hate when determining guidelines?
Answer. Regarding the definition of hate speech, we operate in 190
countries, and hate speech laws vary by country. We respect the law as
required in each country, and will block illegal hate speech content in
a given country to comply with its applicable local laws. In addition,
our hate speech policy is part of the YouTube Community Guidelines,
which we enforce globally. That policy prohibits content that promotes
violence against individuals or groups based on certain attributes,
such as race, religion, disability, gender, age, or veteran status. We
enforce those policies regardless of a speaker's political persuasion.
The YouTube Trusted Flagger program is an important part of our
work with external experts. The program was developed by YouTube to
help provide robust tools for individuals, government agencies, and
non-governmental organizations (NGOs) that are particularly effective
at notifying YouTube of content that violates our Community Guidelines.
Trusted flaggers have expertise in at least one policy vertical, flag
content frequently with a high rate of accuracy, and are open to
ongoing discussion and feedback with YouTube about various content
areas.
We have made improvements to YouTube's flagging tools, based on
feedback from our Trusted Flagger network. In addition to our bespoke
tools for Trusted Flaggers, we designed a dashboard that allows any
user to check the status of flags they have submitted. The dashboard
tells users if the content they flagged is active, removed, or
restricted.
Consultation with external experts is a key aspect of how we
develop our approach to tough issues and how we evaluate our guidelines
and enforcement mechanisms.
Question 2. With almost three and a half billion social media users
worldwide--and one million users joining every day--social media
platforms have turned to a mix of machine learning and human moderators
to detect and take down hate speech, terrorist propaganda, cyber-
bullying, and disinformation. Machine learning can be a useful tool in
identifying objectionable content quickly, preventing it from
spreading. However, there are concerns about its ability to understand
the context of text or images, and the length of time it takes to train
systems with new data to recognize objectionable content.
Can you give us an estimate of how many content moderation
decisions are made by your machine learning systems? And can
you provide an estimated error rate for content flagged by
machine learning?
Are there instances where machine learning is more effective
in flagging certain content than others? Does the error rate
change significantly from one type of content to another?
Answer. As you might imagine, it takes a combination of both
machine learning and human review to effectively review content and we
actively monitor the success of both efforts. With human review, we
check to see what decisions are being made by reviewers and update our
guidelines if they are not clear or not meeting expectations. And we
are constantly working to improve machine learning.
The profound impact of YouTube's updated hate speech policy update
is already evident in the data released in YouTube's Q2-2019
transparency report: the number of individual video removals for hate
speech saw a 5x spike to over 100,000, the number of channel
terminations for hate speech also saw a 5x spike to 17,000, and the
total comment removals nearly doubled in Q2 to over 500 million due in
part to a large increase in hate speech removals. And because of our
ability to remove this content quickly, videos that violate our
policies generate a fraction of a percent of the views on YouTube. For
example, the nearly 30,000 videos we removed for hate speech over the
last month generated just 3 percent of the views that knitting videos
did over the same time period.
______
Response to Written Questions Submitted by Hon. Richard Blumenthal to
George Selim
NO HATE Act and Reporting
I have introduced legislation, the Jabara-Heyer NO HATE Act, which
would help states implement and train officers in the National
Incident-Based Reporting Systems. The NO HATE Act would also provide
grants to states to better address hate crimes by training law
enforcement, establish specialized units, create community relations
programs, and run hate crime hotlines.
Question 1. Do you support the Jabara-Heyer NO HATE Act?
Answer. ADL strongly supports the Jabara-Heyer NO HATE Act.
Since 1990, the FBI has been collecting and reporting hate crime
data, required by the Hate Crime Statistics Act of 1990 (HCSA). While
the FBI HCSA data provides the best national snapshot of bias-motivated
criminal activity in America, it is clearly incomplete. For example, in
2017, the most recent data available, only 2,040 of the 16,149
reporting law enforcement agencies--less than 13 percent--reported one
or more hate crimes to the FBI. The remaining 87 percent of
participating agencies affirmatively reported zero hate crimes to the
FBI, including 92 cities with populations over 100,000. And more than
1,000 law enforcement agencies did not report any data to the FBI,
including 9 cities over 100,000. The entire state of Mississippi
reported one hate crime in 2017, Alabama reported 9, and Arkansas
reported 7. By contrast, two cities that have focused on effective hate
crime response, Boston and Seattle, reported 140 hate crimes and 234,
respectively.
Studies have shown that more comprehensive, complete hate crime
reporting can deter hate violence.\1\ Better data will assist in proper
allocation of police resources and personnel--preventing crimes and
reassuring victims. And better data will advance police-community
relations. Improved data collection will necessarily require outreach
and expanded networking and communication with targeted communities, as
well as more training for law enforcement personnel in how to identify,
report, and respond to hate violence.
---------------------------------------------------------------------------
\1\ See ``Investigation of Hate Crimes; Model Policy; Concepts &
Issues Paper; Need to Know. . . .'' from the IACP Law Enforcement
Policy Center, September 2016, https://www.the
iacp.org/sites/default/files/2018-08/HateCrimesBinder2016v2.pdf.
---------------------------------------------------------------------------
I understand that it has taken some time for Google and Facebook to
establish reliable and timely channels to report threats made on your
platform to the proper authorities. Mr. Slater testified that Google
now has a strong relationship with the Northern California Regional
Intelligence Center, who has been effective at quickly getting reports
of threats into the right hands.
Question 2. Would you support adding measures to the Jabara-Heyer
NO HATE Act to expand the NCRIC model of integrated threat reporting
nationwide?
Answer. While law enforcement collaboration should be part of any
conversation on improving responsiveness to hate crimes, I am not
prepared to comment specifically on the NCRIC model. Overall, it is
important that the Jabara-Heyer NO HATE Act should remain focused on
one thing--improving reported hate crime data. Threats may or may not
be criminal activity.
Question 3. What steps would improve communications channels with
law enforcement to make sure the right information gets into the right
hands quickly?
Answer. Improving hate crime data requires at least two efforts--
law enforcement authorities ready and willing to collect the data, and
members of the targeted communities ready and willing to contact the
police to report that they have been the victims of bias-motivated
violence. The Department of Justice should incentivize state and local
law enforcement to more comprehensively collect and report hate crimes
data to the FBI, with special attention devoted to large underreporting
law enforcement agencies that either have not participated in the FBI
HCSA program at all or have affirmatively and not credibly reported
zero hate crimes.
If marginalized or targeted community members--including
immigrants, people with disabilities, LGBTQ community members, Muslims,
Arabs, Middle Easterners, South Asians and people with limited language
proficiency--cannot report, or do not feel safe reporting hate crimes,
law enforcement cannot effectively address these crimes, thereby
jeopardizing the safety of all. Such efforts could be supported through
the promotion of model policies and best practices and the passage of
legislation designed to improve hate crime data collection and
reporting legislation, such as the Jabara-Heyer NO HATE Act.
Incentives can encourage police departments to report their hate
crime data, and help overcome negative publicity that can accompany
hate crime reporting. Police departments need to have the support of
the community when their hate crime numbers increase; an increase may
well indicate improved police-community relations, increased trust in
police, public confidence that they will respond seriously to hate
crime reports.
Lastly, law enforcement must be encouraged to create relationships
with community members who may be privy to threat information.
This past December, in Monroe, Washington, a clearly troubled young
man made a series of anti-Semitic rants and violent posts online. He
bragged about planning to ``shoot up an (expletive) school'' in a video
while armed with an AR-15-style weapon, and on Facebook posted that he
was ``shooting for 30 Jews.'' Fortunately, these posts came to the
attention of the Anti-Defamation League, which was able to tip off the
FBI. The ADL's vigilance prevented another Parkland or Tree of Life
attack.
Question 4. I take it that the ADL does not report every terrible
or obscene comment on the Internet to the FBI. Can you tell me about
the process and criteria that your organization uses to identify
threats, such as in Monroe?
Answer. Investigative researchers at ADL encounter hundreds, if not
thousands, of posts daily by individuals who make extreme and
threatening comments on various online platforms. When we find such
threats, we delve deeper into that person's online footprint. If that
person displays photos showing his/her weapons and expresses a desire
to use those weapons against a community and we can identify either the
person or where he/she lives, we will report that person's comments to
law enforcement. In addition, if we see that this individual is citing
literature that promotes violence or previous violent acts as
inspiration, we are more likely to report the person. These individuals
often post their comments online hoping to receive support and
encouragement to carry out acts of violence. Individuals we report to
law enforcement are not just talking about hating a group of people--
they want to take action. For example, we reported Dakota Reed in
Washington State for making threats to carry out a mass killing of
Jews, and also reported the comments of Corbin Kauffman of Leighton,
Pennsylvania, in March 2019. In Kauffman's case, in addition to posting
violent comments and pictures of weapons, he posted a photo of himself
carrying out an act of anti-Semitic vandalism. Law enforcement was able
to identify him because of the various clues we provided about his
identity.
______
Response to Written Questions Submitted by Hon. Tom Udall to
George Selim
Question 1. Mr. Selim, in response to the recent spate of mass
shootings, this administration has floated the dangerous idea of
monitoring persons with mental illness to try to predict gun violence.
This policy, however, would only serve to further stigmatize people
with mental illness--who are more likely to be victims of crimes than
perpetrators. It is also a disingenuous way of talking about gun
violence without having to talk about guns.
What do you think about this proposal? In addition, can you
describe what indicators of mass violence or extremism actually look
like, based on the best research?
Answer. It is wrong to assert that people with mental health
disabilities, including those with perceived mental health
disabilities, are inherently dangerous and the cause of our Nation's
gun violence problem or that targeting them will solve our country's
gun violence problem is wrong. In fact, many studies have demonstrated
that people with disabilities, including mental health disabilities,
are far more likely to be victims of gun violence than perpetrators.
Blaming persons with mental health disabilities is counterproductive, a
distraction from the real problem, and can result in stigmatizing
people with mental health disabilities and the disability community as
a whole.
There is no one path to extremism, and there are no specific
indicators that can act as predictors of extremist actions. Some
academic sources have explored whether a combination of factors may
indicate an over-arching risk of radicalization, and I direct you to
those sources, such as this NIJ overview: https://www.ncjrs.gov/
pdffiles1/nij/251789.pdf
Question 2. You were the director of the Office for Community
Partnerships, and led the Countering Violent Extremism Task Force under
the Department of Homeland Security. These offices were responsible for
providing grants to anti-extremist groups and combatting domestic
terrorism through interagency partnerships. Unfortunately, it appears
these offices have been gutted by the current administration. We cannot
fight against white supremacy and violent domestic extremism without
partnering with communities, civil society, and federal, state, and
local governments.
In your opinion, should funding be reinstated to support these
initiatives, and why?
Answer. Yes, funding should be reinstated and scaled much higher.
In light of how domestic terrorism laws differ from those of
international terrorism, there are fewer law enforcement resources at
the government's disposal, and prevention therefore is a key
undertaking for the government. However, in light of the current
administration's inadequacies and singular focus on Islamist-motivated
forms of extremism, entities outside government must take the lead in
preventing extremist violence. A public-private effort--with Congress
funding research universities, technology companies, non-profit expert
organizations, and state and local government partners--could provide
the critical boost that prevention efforts need while also avoiding
misgivings many have about the implications of an overly-federalized
effort.
______
Response to Written Questions Submitted by Hon. Jacky Rosen to
George Selim
As we discuss the spread of hate online, I want to turn our focus
to combating anti-Semitism in the digital sphere. Last year we saw the
deadliest attack on the Jewish community in American history, when
eleven people were killed at the Tree of Life Synagogue in Pittsburgh.
Perhaps unsurprisingly, the shooter was linked to numerous anti-Semitic
postings on a fringe social networking site called Gab. And hours after
a gunman opened fire at a synagogue in Poway, California, a violently
anti-Semitic letter from the shooter appear on 8chan and Facebook, with
links to the letter later showing up on Twitter and other social media
sites, spreading his hateful ideas across the world.
Mr. Selim, online forums such as 8chan and Gab do very little to
police their site from hateful and violent speech.
Question 1. What role do these sites play in perpetuating mass
violence and domestic terrorism in our country, AND
Answer. Fringe web communities play a critical role in the
dissemination of hate and extremist content. Perhaps the most important
contributor to the subculture of the alt right is the so-called
``imageboard,'' a type of online discussion forum originally created to
share images. One of the most important is 4chan, a 15-year-old
imageboard whose influence extends far beyond the alt right, as a key
source of Internet memes. Its/pol subforum is a dark place, an anarchic
collection of posts that range from relatively innocuous to highly
offensive.
Over time, 4chan has become home to many racists and open white
supremacists. Some of its imitators, such as 8chan, lean even more
towards racism and white supremacy. Parts of Reddit, a popular website
that contains a massive collection of subject-oriented discussion
threads, also share the chan subculture, as do parts of Tumblr.
In April 2019, ADL released a report, a collaboration between
Network Contagion Research Institute and ADL's Center on Extremism
(COE), analyzing the similar ideological motivations and online
activity of the perpetrators of the Pittsburgh and Christchurch
massacres. Both killers announced their violent plans to their
preferred Internet forums, Gab and 8chan, and were consumed by the
white supremacist conspiracy theory of ``white genocide,'' which is
frequently referenced on both sites.
Both Gab and 8chan are rife with white supremacist, hateful, anti-
Semitic bigotry. Image boards such as 4chan are totally anonymous,
without user names, allowing participants to say or post whatever they
want, no matter how offensive, without fear of being exposed. Many take
full advantage to engage in some of the most crude and blatant
offensive language online, taking aim at many targets. The chan
subculture has a strong tendency to portray all such content as a joke,
even when not intended to be, resulting in a strong ``jkbnr'' (``just
kidding but not really'') atmosphere. The alt right has also absorbed
an even darker aspect of chan subculture: online harassment campaigns
against people who have angered them.
Question 2. In the immediate aftermath of deadly attacks motivated
by hate, how should mainstream social networks such as Facebook and
Twitter interact with these fringe sites to stop the spread of
manifestos, letters, and other hateful writings?
Answer. While the most extreme forms of online content thrive on
websites like 8chan, Gab, and 4chan, larger social media platforms like
Facebook, Twitter, and YouTube need to remain vigilant. Extremists
leverage larger mainstream platforms to ensure that the hateful
philosophies that begin to germinate on message boards like Gab and
8chan find a new and much larger audience. Twitter's 300 million users
and Facebook's 2.4 billion dwarf the hundreds of thousands of users on
8chan and Gab. Extremists make use of mainstream platforms in specific
and strategic ways to exponentially increase their audience while
avoiding content moderation activity that Facebook and Twitter use to
remove hateful content. These include creating private pages and
events, sharing links that directly lead users to extreme content on
websites like 8chan, as well as using coded language called dog
whistles to imply and spread hateful ideology.
To address this, mainstream platforms should limit the ways they
are spreading hateful messages from smaller platforms. Fringe platforms
like Gab and 8chan openly cater to users interested in spreading hate
and conspiracies, and a considerable amount of their content would
violate mainstream platforms' terms of service. As a result, mainstream
platforms must aim to decrease cross-users' ability to recruit and
spread hate and should increase the friction for users between their
platforms and fringe platforms.
Beyond their community guidelines and content moderation policies,
features available on social media platforms need to be designed with
anti-hate principles in mind. Companies need to conduct a thoughtful
design process that puts their users first and incorporates society's
concerns before, and not after, tragedy strikes. Today, the most
popular method of developing technology tools is through a Software
Prototyping approach: an industry-wide standard that prompts companies
to quickly release a product or feature and iterate on it over time.
This approach completely devalues the impact of unintended design
consequences. For example, the Christchurch shooter used Facebook's
livestreaming feature to share his attack with the world. The feature
could have been designed to limit or lock audiences for new or first-
time streamers or prevent easy recording of the video.
Question 3. Earlier this year, ADL's Center on Technology and
Society called on technology companies, including several of those
testifying here before us, to release ``transparency reports''
providing details on how they define and identify hate speech, how they
moderate hateful content, and the efficacy of these techniques.
Mr. Selim, can you discuss why such reports are useful?
Answer. Knowledge on the efficacy of platforms' content moderation
efforts at dealing with the problem of white supremacist activity
remains extremely limited. Meaningful transparency will allow
stakeholders to answer questions such as: ``How significant is the
problem of white supremacy on this platform?'' ``Is this platform safe
for people who belong to my community?'' ``Have the actions taken by
this tech company to improve the problem of hate and extremism on their
platform had the desired impact?''
We can conduct external research to evaluate their efforts, but
companies often do not share user data, limiting opportunities to
collect and use data for research. Alternatively, we can review
transparency reports on content moderation efforts published by
technology companies, but these too offer very limited information.
Mainstream social media platforms have a few potentially relevant
metrics to the issue of extremism, especially white supremacist
extremism, that they share in their regular transparency reports.
Though each platform provides its own metrics on extremist activity,
the metrics published are limited across the board, they are self-
reported by the companies, and we have no real way of knowing what
content has been put into which category outside of the brief
descriptions given by the platforms as part of their reporting.
In order to truly assess the problem of hate on social platforms,
technology companies must provide meaningful transparency with metrics
that are agreed upon and verified by trusted third parties and that
give actionable information to users, civil society, government and
other stakeholders. Until technology platforms are willing to actively
engage external parties and meaningfully address their concerns through
greater transparency efforts, our ability to understand the extent of
the problem of hate and extremism online, or how to meaningfully and
systematically address it, will be extremely limited.
Have such reports been released?
Answer. Mainstream social media platforms publish transparency
reports. These include Facebook, Twitter, and YouTube.
If we look at the published metrics characterized as being related
to terrorism (Facebook reported 6.4 million pieces of content related
to terrorist propaganda removed from January to March 2019), this may
seem relevant. However, typically, social platform platforms define
terrorism in terms of Al Qaeda and ISIS-related activity and do not
include white supremacist violence or activity as part of the terrorism
classification. White supremacist extremist content could be
categorized as hate speech or violent content on a platform, but at the
same time, so could a wide variety of other types of content not
associated with extremism or white supremacy.
Moreover, when Facebook claims in their transparency report that
they took action on four million pieces of hate speech from January to
March 2019, we still have no sense of how that compares to the level of
hate speech reported to them, what communities are impacted by those
pieces of content or whether any of that content is connected with
extremist activity on their platform.
YouTube provides more granularity, sharing a number of different
categories of content reported by users as well as the amount of
content in each category that YouTube actioned. That being said, the
names of the categories actioned by YouTube differs from those reported
by users, making a comparison between what is reported and actioned
impossible, and providing in the end the same level of opaqueness as
Facebook's report.
Twitter's transparency report on the other hand provides both the
users reported to the platform and users actioned by the platform in
identical categories, but does not provide any information on the
amount of content reported versus amount actioned, making the scale of
their activity similarly opaque.
With almost three and a half billion social media users worldwide--
and one million users joining every day--social media platforms have
turned to a mix of machine learning and human moderators to detect and
take down hate speech, terrorist propaganda, cyber-bullying, and
disinformation. Machine learning can be a useful tool in identifying
objectionable content quickly, preventing it from spreading. However,
there are concerns about its ability to understand the context of text
or images, and the length of time it takes to train systems with new
data to recognize objectionable content.
Question 3. Mr. Selim, earlier this year the ADL announced a
partnership with the Network Contagion Research Institute (NCRI) to
research how extremism and hate speech spread on social media. NCRI
uses machine learning to expose hate on digital platforms. Can you talk
tell us about your findings?
Answer. On October 27, 2018, Robert Bowers perpetrated the
deadliest attack against Jews in American history when he stormed a
Pittsburgh synagogue armed with an assault rifle and three handguns.
Shouting ``All Jews must die,'' Bowers killed eleven people in their
place of worship. Within months, Brenton Tarrant perpetrated the
deadliest attack against Muslims in New Zealand's history when he
slaughtered 50 people gathered for prayer at two mosques. In the wake
of these horrific crimes, Jewish and Muslim communities worldwide and
concerned citizens across the globe began searching for clues about
attacks that seemed to come out of nowhere.
In hindsight, though, these killings should not have been
surprising. Both attackers were enmeshed in online communities that
exposed them to content designed to make them hateful and potentially
violent. Bowers was a member of a fringe online community called Gab,
which, like similar online forums, is a bastion of hatred and bigotry.
Gab has seen a surge in racist and anti-Semitic postings since the 2016
presidential election. Tarrant, too, was part of a fringe online
community called 8chan, one of the most notoriously hateful online
communities on the internet.
Platforms like these force us to reassess our understanding of how
violence may be inspired by such hateful echo chambers. Even more
broadly, as we have recently reported, mainstream platforms can
sometimes push such individuals from an open community, such as
Twitter, into fringe environments like Gab that foster acceptability of
dangerous views.
In September 2018, the Network Contagion Research Institute and its
partners published a study, also detailed in a Washington Post article,
which indicates that the state of online echo chambers of hate is far
worse than many may imagine. Analyzing more than 100 million comments
and tens of millions of images posted between July 2016 and January
2018 to Gab and 4chan's ``politically incorrect'' message board (/pol/
), the NCRI performed the largest quantitative study to date regarding
the rise of anti-Semitism and white nationalism on these popular white
supremacist web communities. The study shows that anti-Semitic slurs
and content doubled on these platforms after the election of President
Donald Trump. During the same timeframe, these web communities also
showed a dramatic surge in the expression of racism, including a
substantial increase in the use of the n-word slur.
NCRI's research also shows that these web communities influence the
spread of hateful memes and images to more mainstream networks like
Twitter and Reddit. This research (along with other studies) shows an
uptick in hateful rhetoric on fringe web communities in the wake of
significant political events or highly publicized extremist violence.
Relatedly, some studies have similarly demonstrated that ethnic hate
expressed on social media can cause surges in real-life hate crimes.
The implications of this online-offline dynamic are highly concerning.
On Gab, Bowers demonstrated how online propaganda can feed acts of
violent terror. On 8chan, Tarrant showed how violent terror can itself
create online propaganda. In both cases, the shooters strongly signaled
back to their fringe web communities with their criminal acts, as
though they were including them as knowing co-conspirators. In both
cases, the participation of these fringe web communities proves to be
key to the scope, sensationalism, and ideological thrust of the act.
Moreover, both shooters claim the same twisted notion of ``white
genocide''--or the imminent destruction of the white race by Jews and
people of color--as the motive behind their terrorist acts, suggesting
a shared ideological motivation. In fringe online communities, many
members indoctrinate other users based on the conspiracy propaganda of
a ``white genocide'' not online violent extremists of other ideologies
spreading a grievance used to justify their malign views.
The evidence overwhelmingly suggests that such platforms can serve
to spread modern violent extremism in ways that could not have been
predicted from the early days of social media. Gab and 8chan fan the
flames of bigotry and hatred and organize violent fantasies in online
communities even as they fuel them in the real world.
There is no telling who else on Gab or 8chan may take cues from
Bowers and Tarrant and act on the violent ideologies they derive from
these online communities. In essence, these platforms serve as round-
the-clock white supremacist rallies, amplifying and fulfilling their
vitriolic fantasies.
[all]