People worry a lot about the various flavours of havoc that AI might wreak in the future, but one of the technology’s here-and-now problems is that of deepfakes, in particular those that depict actual women and girls in sexualised ways without their consent.

Public concern over the issue exploded earlier this year when people used generative AI tools to create explicit content featuring Taylor Swift. This kind of thing wasn’t really new - faked pornographic images have been a problem ever since Photoshop became a thing - but AI apps can make it particularly easy to do, and Swift’s unusually high profile ensured widespread consciousness.

Today, Meta’s independent (but Meta-funded) Oversight Board announced its findings on two deepfake cases that it took on in April. One was about an image on Facebook depicting a female American public figure in the nude, being groped - it is not clear if this figure was Swift, as the board didn’t name the people concerned. The other image, on Instagram, showed a female Indian public figure in the nude, depicted from behind.

The Board, which is sometimes referred to as Meta’s “Supreme Court”, gets the final say when it comes to specific cases such as these. It said today that Meta had done the right thing by taking down the picture of the American figure, despite the protestations of the poster.

However, Meta had initially ignored users’ complaints about the image of the Indian figure, only taking it down when the Board took on the case. Here, the Board decided Meta had been in the wrong.

Beyond specific cases, the Board can only make non-binding recommendations to Meta about its future content moderation. So, Meta will now have to make up its own mind about the Board’s deepfake recommendations, which mostly center on the company’s policy banning “derogatory sexualised photoshop”.

The Board reckons the word “derogatory” should be replaced by “non-consensual”, and the work “photoshop” by something that reflects the changing nature of the tech that people are using to make these images. The Board also wants Meta to start using the fact that content is AI-generated or AI-manipulated as a signal for lack of consent in its "Adult Sexual Exploitation" policy.

Meta said it would review the recommendations.

Of course, a social media company can’t do much more to those creating or sharing such content other than to ban them. So it’s also notable that, yesterday, the US Senate unanimously passed a bill that would let people portrayed in sexually explicit deepfakes sue the creators.

The Disrupt Explicit Forged Images and Non-Consensual Edits Act (DEFIANCE) Act would give victims the ability to claim up to US$150,000 in damages, plus an extra US$100,000 if the deepfake was connected to “actual or attempted sexual assault, stalking, or harassment”. It has a companion bill waiting in the House, and Senate Majority Leader Chuck Schumer (D-N.Y.) urged his counterparts there to take it up.

“By passing this bill, we are telling victims of explicit nonconsensual deepfakes that we hear them and we are taking action,” he said. But the House doesn’t have much time, as the August recess is just over a week away.

Over in the UK, the sharing of such content was already criminalised in last year’s Online Safety Act. The former Conservative government was set to make creation a crime too, but all new bills were dropped when the Conservatives called the election that led to their ouster at the start of this month.

Nonetheless, the new Labour government has also said it wants to criminalise the making of nonconsensual sexualised deepfakes - Deputy Prime Minister Angela Rayner has herself been a victim. Labour’s advisers have also recommended banning apps that are dedicated to making such imagery.

It's unclear how easy these laws will be to enforce, especially as the proliferation of open-source AI models makes it easy for people to set up “nudification” tools without the guardrails that AI providers like OpenAI put in place. But the threat of severe consequences, and action by social media companies, will at the very least make it clearer to people that this kind of behavior is socially unacceptable.

 - NY Times