Posted on January 8, 2020

Facebook Bans Deepfakes Ahead of US Election, but Vows No Collateral Damage Against Satire or Parody Videos

RT, January 7, 2020

Facebook has announced a new policy to tackle machine learning and AI-manipulated ‘deepfake’ videos on the platform ahead of the 2020 US presidential election but claims satire and parody videos will be spared.

Under the new rules, announced Monday by Monika Bickert, Facebook’s head of global policy management, videos will be removed from Instagram and Facebook if they meet two criteria.

Videos will be automatically flagged by Facebook if they have “been edited or synthesised . . . in ways that aren’t apparent to an average person and would likely mislead someone into thinking that a subject of the video said words that they did not actually say” or if they are “the product of artificial intelligence or machine learning that merges, replaces or superimposes content onto a video, making it appear to be authentic.”

If a video is deemed to be false or partly false, Facebook will “significantly reduce” its distribution in users’ news feeds and reject it outright if it is posing as an ad.

The company is opting to leave such offending videos on the platform but marking them false, to help prevent them spreading under false pretences on other social media platforms and elsewhere on the internet.

In a bid to cover all its bases, Facebook said that even videos that do not meet these two criteria would still be eligible for “independent review” by the company’s third-party fact-checkers, which include 50 partners covering 40 languages.

Facebook gave no immediate, clear examples of what type of content would be in breach of such rules though it did state that the policy does not cover ‘shallow fakes’ or videos made using traditional video editing tools, while claiming that parody and satire videos would be exempted.

“This policy does not extend to content that is parody or satire, or video that has been edited solely to omit or change the order of words,” Bickert stated.

One example of the difference in content which contravenes the guidelines and that which does not would be videos shared last year of House Speaker Nancy Pelosi slurring her speech. This footage would apparently not contravene Facebook’s new ‘deepfake’ guidelines as it was merely slowed down, and the audio pitch-shifted, using traditional video editing techniques.

In 2018, Donald Trump shared a video of a Fox News segment which was edited in such a way as to make it appear as though Pelosi was fumbling her words. This would fall under “traditional video editing techniques” and would therefore be allowed on the platform.

But another video that went viral online purported to show Pelosi slurring far more heavily and was often posted with titles insinuating that she was drunk. This type of video falls somewhere in the middle as it has been both slowed down and pitch-shifted though not necessarily by an AI or machine learning algorithm, and would therefore likely be referred to the third party fact checkers for review.

However, the company added that if a politician posts something which breaches the deepfake guidelines, they may still allow it to remain, even if it was expressly intended to mislead, if Facebook deems the public interest in seeing it outweighs the potential harm.