Should copyright protect AI generated work?

From NFTs to generative AI, the rate at which “game changing” technology is being rolled out iin recent years is remarkable. In a matter of months, generative AI models like ChatGPT have gone from something experimental and funny to something that is actively being seen as a threat to human livelihood.

“Things are moving extremely rapidly from something that's largely theoretical to a product that actually has real world application and that there's a consumer or client demand for”, James & Wells IP partner Ian Finch told NZ Lawyer. “That's happening to us as IP lawyers, in a field where we should be on top of that technology, and we were struggling to keep up with it. So you really do sympathise for all the fields, and in particular, the creatives who have seen a major and dramatic impact on their livelihoods extremely quickly. And there's not much that they can do to pivot to respond to that change”.

Finch pointed out that the rapid pace of AI advancement meant that lawmakers have largely been unprepared for the novel legal issues that came with this development – such as whether other people’s works or IP can be used to train AI.

“Is it acceptable to use that model to generate more works? Are you infringing on the original material if the original act of training was an infringement? Because it's not clear that the training wasn't an infringement, let alone the use of the product”, he explained. “And then on the back end, if you do use these tools to generate content, who owns it? Is it the company that owns the tool? Is it the person that's providing the instructions? Is it the person who provided the materials that train the tool? Or is it no one?”

Finch highlighted the recent case involving the AI program Device for Autonomous Bootstrapping of Unified Sentience (DABUS), which sought to determine whether an AI system can be recognised as a patent inventor. The case reached various courts worldwide, and while courts largely ruled that a patient inventor needed to be a “natural person”, Finch pointed out that the matter raised issues regarding ownership.

“You have those issues around getting computers to work in a semi-humanistic way, and there's the law out of step with the idea that you can actually use a computer as a tool to generate content. The concept of ownership needs to recognise that it may be creative input in itself to instruct a machine to produce a work”, he said.

“No one really envisioned it when they were drafting the legislation which would be relied on to enforce rights. Our Copyright Act in New Zealand (1994) is a pretty old piece of legislation by modern standards – even 10 years ago, it was criticised as being out of step with technology”.

What is copyright meant to protect?

James & Wells associate Sam Parkinson pointed to a significant case in London where Getty Images is seeking considerable damages for the inclusion of about 12 million of its images in the original training set for image generation AI Stable Diffusion.

“From a legal perspective, there's the issue of copyright; in particular, of potential trademark infringement because the generated images actually still included a really deformed version of the Getty Images watermark”, Parkinson told NZ Lawyer.

He also highlighted pushback from the games industry on AI use.

“Steam, by far the biggest digital marketplace for PC games, didn't allow a developer to list their game on it without being able to show that they had ownership and weren't infringing IP rights in the AI generated content”, Parkinson said. “Which, of course, they can't do – it’s very difficult given the current state of the law, the huge sizes of these data sets and the resources of small independent teams to determine whether or not what they're getting from these models is free to use”.

He added that cases being tackled in California have also touched on unfair competition laws. But ultimately, he believes that most of the legal issues related to AI content boil down to copyright – and what actually drives the production of new creative works.

“New Zealand is in a fairly unique position at the moment where the Copyright Act currently allows for computer-generated works to be protected – but when the Act was drafted, generative AI was not around. It is not clear how providing copyright protection to AI generated artworks would achieve the goal of incentivising the creation of new, original artistic works, so I think we are likely to see some changes here”, Parkinson said.

“The copyright system was designed and put in place to protect creatives and incentivise the creation of further creative works. If creatives are not incentivised to create new works, we are at risk of the system collapsing, and no new, original, human authored works being created, or at least none which are publicly shared for fear of them being vacuumed up to train an AI”.