Ai, Ai, Ai – It’s just not fair use!

Generative AI has a rocky relationship with copyright in the US – with The New York Times, Getty Images and a class action of artists each each suing AI companies for copyright infringement.

Generative AI produces content based on inputted word stimuli. You could order a large language model, like Chat GPT to write a poem about cats or you could demand a diffusion model, like Midjourney, to generate an image of a cat wearing a suit. 

Machine learning algorithms in generative AI models train on massive data sets scraped from websites. Scraping involves copying content to make derivative works that are communicated to the public in generative AI outputs. This is copyright infringement in most jurisdictions unless a defence or exception applies.

The Washington Post says that AI’s Future could hinge on one thorny legal question – the US fair use defence. I spoke to intellectual property barrister and UTS lecturer, Jane Rawlings, who agrees.

The defence can consider whether the purpose and character of a use of copyright work is transformative.

Fair use can also consider whether generative AI outputs supplant the markets and value of the copyright work – be it news for The New York Times, or photos for Getty Images. Consider how Google’s new AI Overviews will render news articles obsolete.

However, Rawlings points out that recent cases are at odds. The Authors Guild v Google Books case found ‘snippets’ of copied content didn’t undermine the market for books and provided a fundamentally different service. 

But in the Andy Warhol Foundation for the Visual Arts v Goldsmith case, competing with an original work weighed against fair use. 

The balance between the interests of copyright owners and users is murky and it is unclear where AI companies will fall in the US where, at the time of publication, 29 copyright lawsuits are unfolding.

But how would similar lawsuits fare in Australia?

Australia’s fair dealing exceptions are narrower. The exceptions for research or study might apply, but would likely fail since the purpose can’t be to develop a commercial product. This means copyright owners may have more protection from AI scraping in Australia.

It seems to me that the horse has bolted for copyright law to hold AI companies accountable. Even if damages are awarded to plaintiffs like The New York Times, which has a convincing case, so much data has already been scraped and industries are already permanently changed.

So what do you think? Is copyright law sufficient to regulate the data scraping required to run generative AI models? Or do we need a new type of intellectual property regime for AI, which upholds fair credit, compensation and consent?