What are Deepfakes?
Have you seen the 2019 video footage of President Obama swearing during a public service announcement? How about the one with Mark Zuckerberg announcing that he is deleting Facebook, which attracted 72 million views and led to outrage among viewers who believed the contentto be authentic? What about the Willy Wonka & the Chocolate factory clip in which Ryan Reynoldstakes real-star Gene Wilder’s place? If you did, andeven for a moment believed their surprising content to be genuine, then you were tricked.Welcome to deepfakes. While analog and digital fakes are not new, deepfakes leverage powerfultechniques from machine learning (ML) and artificial intelligence (AI) to manipulate or generatevisual and audio content with a high potential to deceive. The phenomenon gained its name from an anonymous user of the platform Reddit, who went by the name ‘deepfakes’ (deep learning þ fakes) and who shared the first deepfakes by placing unknowing celebrities into adult video clips. By sharing the computer code that produced the deepfakes, widespread interest spawned in the Reddit community and led to an explosion of fake content. The first targets of deepfakes were famous people, including actors (e.g., Emma Watson, Scarlett Johansson), singers (e.g., Katy Perry) and politicians (e.g., U.S. presidents Obama and Trump), whose faces were transposed without their permission onto others. One of the early deepfakes that showcased the power of AI and deep learning was 2017’s “Synthesizing Obama” (Suwajanakorn, Seitz, & Kemelmacher-Shlizerman, 2017), which featured an impressive use of lipsyncing technology based on existing audio footage.b Today, we could be watching the leader of one country convincingly deliver a speech by the leader of another country, or vice versa. Such deepfake trickery is alarmingly successful for two main reasons: believability and accessibility.