Facebook bans deepfakes in the runup to 2020 election, but ‘shallowfakes’ still allowed

Facebook has announced a new policy change that will see the site ban deepfake content from its platform in the runup to the 2020 presidential election in November. A “deepfake” is the colloquial term for synthetic media, or media that has been created artificially using deep learning techniques.

Historically, deepfakes got their start by being used to create pornographic videos featuring people that never took part in the video in the first place, such as famous actors. But the technology is increasingly becoming embraced to spread political propaganda and misinformation.

If the AI or deep learning techniques behind a deepfake video are advanced enough, the synthetic media can make it look like someone—such as the president of the United States, or one of his opponents—is doing or saying something they, in reality, never said and did. Because of the very nature of their content, deepfakes have the potential to easily go viral online.

That’s why Facebook has announced a ban on them on their platform ahead of the 2020 elections. In a blog post, Facebook says that it will be removing media if it meets two criteria:

• The video has been artificially generated or edited to make it look like a person said something they, in fact, have not said.
• The video replaces or adds content that makes it look authentic by using AI or machine learning means.

But Facebook says its new deepfake policy does not apply to parody or satirical content, leaving room for users to argue their deepfakes should remain on the platform. The new policy also does not ban the posting of so-called “shallowfakes”—that is, authentic video that has simply been edited to rearrange the order of words or omit them entirely. Perhaps the most well-known example of a shallowfake is the manipulated video of House Speaker Nancy Pelosi that was edited in such a way to make her appear drunk.

But despite Facebook’s deepfake ban, the lack of a shallowfake ban and the fact that both types of videos can still be shared in WhatsApp, Messenger, and Instagram DMs mean the deepfake problem is probably far from over on Facebook’s platforms.