Facebook Just Banned “Deepfake” Videos, But What Exactly Are They?

Ahead of the 2020 presidential election, Facebook announced an updated policy to remove videos modified by artificial intelligence known as “deepfakes.” In a major play for the company that has come under scrutiny for allowing fake political ads to remain, the policy’s scope already presents a myriad of questions regarding its implementation. 

The announcement came from Facebook’s head of global policy management Monika Bickert who explained the criteria a misleading video on Facebook or Instagram must meet in order to be removed. Bickert also explained what happens to the videos that are reported, but don’t meet the requirements.

The new policy doesn’t cover videos edited with conventional tools known as “shallow fakes,” which have been the category most examples of damaging, manipulated media in recent years falls under. In May 2019, a video altered to make House Speaker Nancy Pelosi appear to be unwell and slurring her words as if she might be drunk.

The video went viral, and at one point, President Donald Trump’s personal lawyer Rudy Giuliani even shared it. Despite the obvious malicious intent of the video, the editing techniques used are not sophisticated. The video was slightly slowed down and the pitch of Pelosi’s voice was altered. Had it come out after this new Facebook policy, it would not qualify for removal.

“If we simply removed all manipulated videos flagged by fact-checkers as false, the videos would still be available elsewhere on the internet or social media ecosystem,” Bickert said. “By leaving them up and labelling them as false, we’re providing people with important information and context.”

Deepfake videos are new category of edited video that has emerged in the last few years. Creators of the video use sophisticated editing techniques such as artificial intelligence and machine learning to alter raw footage. The technology is advance enough to make the subject appear to say something they never said. Most importantly, the technology is developed enough that the alterations appear near seamless from the original footage.

According to the Facebook announcement, their new policy has two main criteria for deepfake videos. First: It must be edited or synthesized in such a way that it is not readily apparent to the average viewer. Second: the video is a product of artificial intelligence or machine learning that either merges, replaces, or superimposes content onto a video in an attempt to make it appear authentic. 

Deepfake videos are dangerous not simply because the information is misleading or outright false, but because the editing is so sophisticated that the majority of people cannot detect that the video is the product of heavy editing.

It becomes increasingly difficult for viewers to discern the accuracy of a video if footage can be manipulated to this great of an extent. Technology in this area is rapidly advancing, and while it still requires a considerable amount of skill and processing power to make a convincing deepfake video, it is becoming easier and more accessible.

Not necessarily. The new policy not only overlooks simpler but equally pervasive and effective techniques, it also allows exceptions based on the source of the video.

According to these new terms and conditions, content is allowed to remain online – whether it is fake or not – if it is judged to be newsworthy. Furthermore, all content posted by politicians is automatically viewed as such.

“That policy means that even an AI-created deepfake video expressed intended to mislead could still remain on the social network, if it was posted by a politician,” said Nick Clegg, Facebook’s vice president of global affairs and communications. Videos that are “manipulated for purposes of parody or satire” are also exempt which could lead to heated debates over whether a video is labeled as deceptive or merely altered for comedic effect.