ARE cutting-edge developments in artificial intelligence built on a foundation of mass copyright infringement? We may start to find out in 2024. In the US, there are against AI firms, including OpenAI, Meta and Google parent company Alphabet. One of those, Getty Images vs Stability AI, is also being tried in the High Court in London.
Though each case varies on the specifics, accusations of copyright infringement are a common theme. AI firms infringed copyright, say the claimants, by taking intellectual property from the internet, like whole books or artworks, to train their models.
The people who own those works want to be compensated, while the AI firms very much don’t want to pay out. “Imposing the cost of actual or potential copyright liability on the creators of AI models will either kill or significantly hamper their development,” , which has significant holdings in AI companies, in a submission to the US Copyright Office in November.
Advertisement
Now, the courts are getting involved. The US class-action lawsuits, in which the plaintiffs represent a broader group, like any artist whose works were used to train a model, could have wide-ranging implications, so lawyers must first set out their arguments carefully to show they merit class-action litigation. A back-and-forth will then ensue between the plaintiffs – who will try out different arguments, some stronger than others – and the defendants, who will try to dismiss the arguments, either through previous cases or existing law.
“People say a lot of crazy things in a class-action pleading,” says at Emory University in Atlanta. “There’s no particular incentive to be judicious, to only run your best arguments, to only say things that you can actually back up with facts.”
Arguments that can’t be dismissed through this process will then be tried in court, but there is very little chance we will see such cases settled next year, says Sag. “Class-action lawsuits are never fast. The only way they’re fast is if they just completely flame out.” A similar legal claim – for copyright infringement over its book digitisation project, Google Books – took almost a decade to resolve, being decided in favour of the tech firm in 2015.
We might find out more from Getty Images vs Stability AI, which is being tried in the UK. Stock photo company Getty Images alleges that Stability AI, which is based in London, infringed copyright by training its AI image generator, Stability Diffusion, using Getty’s material.
There is little public information on Stability AI’s arguments for the defence. The firm did attempt to dismiss the lawsuit, claiming that the training of its model wasn’t done in the UK, but the High Court rejected this in December. Both parties have been regularly attending hearings to decide the shape that the case will take, says , head of artificial intelligence law at legal firm Gowling WLG in London, and the case should conclude in the second half of 2024. Gowling WLG isn’t involved in the proceedings.
If the claim proceeds to trial in London, Stability AI might find it tougher to mount a defence there than in the US. “There are very few exceptions that are suitable for training AI on expressive works in the jurisdiction of the UK,” says Hervey.
If Getty were to win, this could encourage other claimants to try to seek compensation from AI firms, but it might be hard to find a defendant. “Most of these [AI] companies have really come out of the west coast in the US,” says Hervey. “I’m not aware of any other juicy targets in this country who have openly trained on copyright materials.”
Developments in any of these cases might lead to some AI companies “jurisdiction shopping” – training their models in countries where the chance of legal action is lower. A ruling against Stability AI, as well as the current regulatory and legal landscape in the UK, could persuade firms to opt to do their AI training in places like the US, where the legal system appears more favourable, says Hervey.
