Google has renewed its campaign seeking to pressure the UK government to rewrite copyright law to the advantage of AI companies. So that it’s easier for those companies to make use of existing content, including music, when training AI models without having to go to all the trouble of securing and paying for licences from relevant creators and rightsholders.
Among other things, Google’s British lobbyists argue that providing rigorous protection for creators and rightsholders under UK copyright law isn’t as helpful as you might think, because American AI companies have access to all that content anyway, and can use it to train models in the US without securing any licences, because in the US AI training constitutes ‘fair use’.
Of course, as far as the music business and wider creative industries are concerned, AI training is definitely not fair use under US copyright law. But most tech companies are adamant that it is.
And as Google’s UK lobbyists ramp up their efforts here, the company’s US-based President Of Global Affairs, Kent Walker, has written a white paper on AI regulation in which he bullishly assumes that AI training is already fair use, before insisting that regulators and judges must ensure it stays that way.
Referring to all the art, music and content currently stored on the internet as “publicly available web data”, he says that using that data to train AI models “is a transformative, non-expressive use - like an art student taking inspiration from walking through a gallery - that should remain protected under fair use in the US”.
If AI training is fair use, then - within the US at least - AI companies can make use of any existing content when training their models without getting permission from any creators or rightsholders. If it’s not fair use, they will need to secure permission from all those creators and rightsholders through licensing deals.
Elsewhere in the world, where the fair use principle is not part of the copyright system, AI companies look for specific copyright exceptions covering text and data mining which they can possibly rely on to train models without having to secure any licences. In the UK, no such exception exists for commercial entities, though a new exception has been proposed twice in the last five years.
The current UK government officially abandoned its most recent plan to introduce a new exception to benefit AI companies in March this year, replacing that plan with a new policy that can basically be summarised as “we’re not really sure what we’re going to do”.
It seems ministers still desperately hope they can find a compromise position between AI companies and rightsholders, even though such a compromise position almost certainly doesn’t exist.
It’s in that context that, according to the Daily Telegraph, Google is continuing to pressure ministers to introduce more flexibility into UK copyright law, employing the argument that if AI companies can grab existing content for free in other countries, not allowing that to happen here will put the UK AI sector at a disadvantage.
The Telegraph quotes Google’s UK Head Of Public Policy Katie O’Donovan as saying, “when any company is building a data centre here or indeed building an AI company here”, as soon as they want to train a model, “they have to go to a different jurisdiction”. And that “puts a ceiling on the sophistication” of the UK’s AI economy, “that investment, that capacity, goes elsewhere”.
In the US, although Donald Trump’s White House is generally more sympathetic towards the AI companies, its official position is that the courts should decide whether or not AI training is fair use. And there are now dozens of lawsuits between rightsholders and AI companies that centre on that very dispute.
That includes a lawsuit filed by a group of independent musicians against Google, which will almost certainly ultimately centre on the fair use debate. Although currently Google is trying to get the case dismissed with a different argument, that the indie artists actually granted the company permission to use their music in AI training when they uploaded videos to YouTube.
Elsewhere in the world, both the creative industries and the tech sector continue to lobby hard. In some countries, including the UK and Australia, it felt like the creative industries had got the upper hand, fighting off proposals for new copyright exceptions. Though there are now growing concerns that the tech sector’s lobbyists are mounting a major push back that could as yet succeed.
Walker’s white paper talks about both fair use in the US and copyright exceptions elsewhere. While insisting that we need “balanced copyright frameworks” that “enable innovation” - by which he means companies like Google should get free access to online content for AI training - he does try to offer creators and rightsholders a few compromises in return.
He says “responsible model developers should give website owners choice and control over whether their content is used in model development via simple, machine-readable robots.txt tags”, such as his company’s existing Google-Extended control. Plus, he reckons, there are commercial partnerships to be agreed between tech companies and rightsholders on developing certain AI products.
He then focuses on the outputs of generative AI models, which is where he concedes copyright should kick in - ie where AI generated works clearly replicate an existing work that should be licensed, or prevented through guardrails in the AI model, or at least be subject to takedowns when spotted by rightsholders.
And, Walker writes, Google agrees that new regulation is needed when it comes to AI-generated content that exploits a person’s voice or likeness. “We have also supported regulatory proposals that can help protect individual voices and likenesses by establishing a balanced national standard against unauthorised digital replicas”, he explains, alluding to the No FAKES Act in the US.
Optimistic ministers in the UK government might see Walker’s proposals as forming the basis of the compromise between big tech and the creative industries that they’ve been desperately looking for. Although they’re almost certainly not.
The music industry and its allies across the other creative industries remain steadfast that - while we do also need new protections on the outputs of generative AI, including around digital replicas - permission is still needed for AI training on the input, and copyright law should clearly say so.
And if that argument can be won worldwide (even if that's a very big “if” indeed), then O’Donovan’s claim that UK-based AI companies will just move their training to the US and get their free ride under its copyright regime quickly falls down.
With the British government about to get a major revamp under a new Prime Minister, it’s not entirely clear where the AI copyright debate will go next in the UK. Although we can be certain both the creative industries and big tech will be lobbying ferociously as the debate continues.
