Responding to Deepfakes and Disinformation

Did Tom Cruise really impersonate a snapping turtle on TikTok? Did former President Barack Obama actually reference the movie Black Panther and say the supervillain was right? In both cases, the answer is “no.” These videos are deepfakes. “Deepfakes”—a hybrid of the terms “deep learning” and “fake”—are videos, photos, or audio recordings that have been modified to make false content appear real. The technology can replace faces and speech to make it appear as if someone said or did something that never happened. The most sophisticated deepfake videos require thousands of images to train algorithms to recognize and then manipulate a face. In March 2021, realistic deepfake videos of Tom Cruise on TikTok fooled deepfake detection software, because the video algorithm used over 13,000 images of Tom Cruise that captured him from almost every angle. Although only sophisticated experts can make the most realistic deepfakes, anyone with an iPhone and a single photo can now use free apps to create simple deepfakes. These amateur deepfakes are stilted and noticeably computer-generated, but as technology to create realistic deepfakes becomes more accessible, soon anyone could have the power to wield deepfakes and interfere with other people’s identities and reputations. Deepfakes have profound potential to cause harm. The first deepfakes appeared around 2017 and comprised pornographic videos that swapped female celebrities’ faces onto other persons’ bodies. By 2019, a startup reported that approximately 96 percent of deepfakes are pornographic and disproportionately victimize women, casting them in humiliating, violent situations and exposing them to rape and death threats. Bad actors can also exploit others by using deepfakes to commit identity theft, blackmail, and fraud. Deepfakes are uniquely effective at spreading disinformation, which raises critical concerns for democracy and national security. Effective democratic discourse requires that voters start from the same foundation of facts, but deepfakes can lead individuals to live in their own subjective realities and exacerbate social divisions. Deepfake videos depicting public figures making incendiary comments or behaving inappropriately could also alter election outcomes. As deepfakes become more well known, public officials caught on camera can exploit a “liar’s dividend” and claim that a real video is a deepfake. Without clear methods to distinguish what is real from what is not real, the public may lose trust in media and other public institutions. In an initial effort to respond to deepfakes, the U.S. Congress adopted legislation in 2019 requiring the Director of National Intelligence to prepare annual reports about the national security impacts of deepfakes and foreign weaponization of deepfakes. Members of Congress have introduced other legislation to address deepfakes. In July, a bipartisan group of Senators introduced the Deepfake Task Force Act, which would establish a task force in the U.S. Department of Homeland Security to research and develop a government-wide plan to counter deepfakes. Earlier proposals included the Identifying Outputs of Generative Adversarial Networks Act, which would authorize the National Science Foundation to support research on deepfakes and the National Institute of Standards and Technology to develop measurements and standards for technologies to detect deepfakes. The Defending Each and Every Person from False Appearances by Keeping Exploitation Subject to Accountability Act would mandate that deepfakes identify themselves as altered media by containing embedded digital watermarks and including verbal or written disclosures that describe the alterations. In addition to these federal proposals, various states have passed deepfake legislation. And social media platforms have changed their terms of service in efforts to reduce harms to individuals and to protect the integrity of elections. In this week’s Saturday Seminar, scholars explore the problems created by deepfakes and the challenges for regulators seeking to address these problems