HEARING BEFORE THE UNITED STATES HOUSE OF REPRESENTATIVES COMMITTEE ON ENERGY COMMERCE SUBCOMMITTEE ON CONSUMER PROTECTION COMMERCE January 8 2020 Testimony of Monika Bickert Vice President for Global Policy Management Facebook I Introduction Chairwoman Schakowsky Ranking Member McMorris Rodgers and distinguished members of the Subcommittee 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 at Facebook In that role I lead our efforts related to content 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 cybersecurity 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 Facebook is a community of more than two billion people spanning countries cultures and languages across the globe Every day members of our community express themselves on our platform in diverse ways having conversations and posting content from text and links to photos and videos We are proud of the wide array of expression on Facebook but we also recognize the important role we play in addressing manipulation and deception on our platform II Combating Manipulation and Deception Community Standards We publish Community Standards governing the types of content and behaviors that are acceptable on Facebook For example we prohibit hate speech harassment content posted by fake accounts and—under a new policy—misleading manipulated media including certain types of deepfakes When we become aware of content that violates our Community Standards through either proactive technical measures or reports we remove it Some types of misinformation—such as attempts to interfere with or suppress voting or participation in the census—violate our Community Standards and we work proactively to remove this type of harmful content We are mindful of our responsibility to respect freedom of expression but our Community Standards are clear that we remove content when it has the potential to contribute to offline physical harm 1 We recognize the risks of manipulated media Manipulated media can be made with simple technology like Photoshop or with sophisticated tools that use artificial intelligence or “deep learning” techniques to create videos that distort reality—usually called “deepfakes ” While these videos are still relatively rare on the internet they present a significant challenge for our industry and society as their use increases and we have been engaging broadly with internal and external stakeholders to better understand and address this issue Based on these conversations we have been considering a number of options regarding misleading manipulated media including deepfakes That is why we just announced a new policy that we will remove certain types of misleading manipulated media from our platform In particular under this policy which is part of our Community Standards we will remove videos that have been edited or synthesized using artificial intelligence or deep learning techniques in ways that are not apparent to an average person and that would mislead an average person to believe that a subject of the video said words that they did not say The policy is designed to prohibit the most sophisticated attempts to mislead people To be clear forms of misleading manipulated media that do not meet these criteria—such as videos that have been edited solely by splicing to omit or change the order of words or parodies or satires—are still subject to our other Community Standards and are eligible for fact-checking For example a synthesized video of a celebrity in which the celebrity is nude would violate our nudity policies Manipulated media may also be spread in a coordinated manner by fake accounts which would violate our policies against inauthentic behavior in such cases the content posted by such accounts would also be removed Misinformation We recognize that some types of misleading information lack quality and integrity despite not directly violating our Community Standards Our approach to such misinformation has several components including working with independent third-party fact-checkers to help reduce the spread of false news and other types of viral misinformation investigating AI-generated content and deceptive behaviors like fake accounts partnering with academia government and industry on tackling broad issues and exposing the bad actors behind these efforts People share millions of photos and videos on Facebook every day We know that this kind of sharing is particularly compelling because it is visual That said it also creates an opportunity for manipulation by bad actors Manipulated photos and videos can be factchecked by one of our independent third-party fact-checking partners who are certified through the non-partisan International Fact-Checking Network We now have over 50 partners around the world fact-checking content in over 40 languages and we are investing in ways to scale these efforts further Fact-checkers use their own expertise to determine which stories to review many of our third-party fact-checking partners focus on misinformation in images and videos This includes identifying when an image or video is being presented out of context using tools such as reverse image search or 2 utilizing video editing programs to identify when manipulation has occurred Factchecking partners are able to assess the truth or falsity of a photo or video by combining these skills with original reporting including outreach to technical experts academics or government agencies Once a fact-checker rates a photo or video as false or partly false we reduce its distribution in News Feed and reject it if it’s being run as an ad We also implement an overlaid warning screen on top of photos and videos marked as false If people try to share the content they will be notified of the additional reporting They will also be notified if content they have shared in the past has since been rated false by a factchecker Moreover in order to more effectively fight false news we also take action against Pages and domains that repeatedly share or publish content which is rated as false Such Pages and domains will see their distribution reduced as the number of offenses increases Their ability to monetize and advertise will be removed after repeated offenses Over time Pages and domains can restore their distribution and ability to monetize and advertise if they stop sharing false news We also use machine learning to assist in our fight against misinformation Algorithms cannot fundamentally tell what content is true or false but they do help in the process For example our machine learning models use various signals to identify content which might be false or partly false Comments expressing disbelief are one signal that helps inform our prediction as well as feedback from our community when people mark something as false news And we use model predictions to prioritize the content we show third-party fact-checkers Since there are hundreds of millions of pieces of content per week shared on Facebook we prioritize third-party fact-checkers’ time In addition to helping us predict content for fact-checkers to review machine learning helps us identify duplicates of debunked stories In turn fact-checker ratings help further train our machine learning model so it’s a cyclical process We are always improving our policies and enforcement practices and we will continue to closely monitor this issue and to consult with external stakeholders to ensure we’re taking the right approach Across the world we’ve been driving conversations with more than 50 global experts with technical policy media legal civic and academic backgrounds to inform our policy development As these partnerships and our own insights evolve so too will our policies toward manipulated media In the meantime we’re committed to investing within Facebook and to working with other stakeholders in this area to find solutions with real impact Coordinated Inauthentic Behavior The idea behind Facebook is to help bring communities together in an authentic way We believe that people are more accountable for their statements and actions when they use their authentic identities Fake accounts are often behind harmful and misleading content and we work hard to keep them off Facebook We took down over 5 billion fake accounts 3 in the first three quarters of 2019 and our technology stopped millions of additional attempts every day to establish fake accounts before they were created When we take down these accounts it’s because of their deceptive behavior like using networks of fake accounts to conceal their identity it’s not based on the actors behind them or what they say Our efforts to prevent coordinated inauthentic behavior focus on four areas First our expert investigators use their experience and skills in areas like cybersecurity research law enforcement and investigative reporting to find and take down the most sophisticated threats Second we build technology to detect and automatically remove the most common threats Third we provide transparency and reporting tools so users can make informed choices when they encounter borderline content or content that we miss We publicize our takedowns of coordinated inauthentic behavior for all to see and we also provide information about them to third parties for their review and share relevant data with researchers academics and others Fourth we work closely with civil society researchers governments and industry partners so they can flag issues and we can work to resolve them quickly Engaging with these partners regularly helps us improve the efficacy of our techniques and learn from their experiences Using this combination of tools we continually adapt our platforms to make deceptive behaviors much more difficult and costly When we conduct a takedown we identify the tactics the bad actors used and we build tools into our platforms to make those tactics more difficult at scale By continuing to develop smarter technologies enhance our defenses improve transparency and build strong partnerships we are making the constant improvements we need to stay ahead of our adversaries and to protect the integrity of our platforms III Partnering to Improve Deepfake Detection Deepfake techniques have significant implications for determining the legitimacy of information presented online Yet researchers in academia government and industry still lack strong data sets to analyze and benchmark this challenge We want to encourage additional research and development in this area and ensure there are better open-source tools to detect deepfakes That’s why Facebook has partnered with a cross-sector coalition of organizations including the Partnership on AI Cornell Tech the University of California Berkeley MIT WITNESS Microsoft the BBC and AWS among several others in civil society and the technology media and academic communities to build the Deepfake Detection Challenge The goal of the Challenge is to produce technology that everyone can use to better detect when AI has been used to alter a video in order to mislead the viewer The Deepfake Detection Challenge includes a data set and leaderboard as well as grants and awards to spur the industry to create new ways of detecting and preventing media manipulated via AI from being used to mislead others The governance of the Challenge will be facilitated and overseen by the Partnership on AI’s new Steering Committee on AI and Media Integrity which is comprised of members including Facebook and others in civil society 4 and the technology media and academic communities It’s important to have data that is freely available for the community to use That is why we constructed a new training data set specifically for this Challenge working with a third-party vendor to engage a diverse set of individuals who agreed to participate in creating the data set for the Challenge We are also funding research collaborations and prizes for the Challenge to help encourage more participation In total we are dedicating significant resources to fund this industry-wide effort We also recently launched an e-learning course as a partnership between the Facebook Journalism Project and Reuters Institute The course titled “Identifying and Tackling Manipulated Media ” aims to help newsrooms around the world equip themselves to identify manipulated media It includes real-world examples hypothetical cases and insights into the evolving technology used to create and detect manipulated media including deepfakes It teaches journalists about the various types of altered media and the ways in which newsrooms can confidently verify and publish truthful content from third-party sources Manipulated media presents a constantly evolving challenge and our hope is that by helping the industry and the AI community come together we can make faster progress in a way that benefits the whole of society IV Conclusion We recognize that both the issues and challenges in addressing manipulated media are rapidly evolving Experts have called on the industry to come together to develop a consistent approach across platforms As they have pointed out consistent enforcement across platforms is important to protect consumers from such content migrating from platform to platform We agree As our CEO Mark Zuckerberg has said we need to develop consistent industry standards on issues such as manipulated media We have encouraged the industry through our trade association to work together—specifically on manipulated media—in a more uniform way pushing for common standards and a consistent approach across platforms We welcome the opportunity to collaborate and partner with other industry participants and interested stakeholders including academics civil society and lawmakers to help develop such an approach Leading up to the 2020 US election cycle we know that combating misinformation including deepfakes is one of the most important things we can do We will continue to look at how we can improve our approach and the systems we’ve built including through continued engagement with academics technical experts and policymakers Thank you and I look forward to your questions 5