It is not ideal that AI companies are being impeded by copyright laws. That is the view of Markus Anderljung, Director of Policy and Research at an organization called the Centre for the Governance of AI. Speaking at a Westminster eForum policy conference on AI regulation recently, he said:
We should expect that there'll be a small handful of companies that produce the most capable, general-purpose AI systems. And much of the rest of the economy will end up using, adapting, and building on these models, rather than training large-scale general-purpose systems themselves.
So, what might that mean in terms of what kind of regulatory changes will be needed? Overall, existing regulations will become outdated. We will have to adjust our regulatory regimes as they are brought to bear.
Regulation will end up unnecessarily standing in the way of innovation and AI adoption. One example here is copyright law, which is currently is making it a lot harder to train AI systems than is ideal.
There it is yet again - the belief that innovation must never be impeded, come what may, and that copyright is just a shibboleth with no contemporary relevance.
There is much to unpack there, not least of which is the Director of a governance organization appearing to urge acceptance of adapting our economies to Big Tech’s whims. Which regulations ‘need’ to change is the critical issue here: who needs them to? Other people’s copyright is a mere obstacle in the path of trillion-dollar steamrollers, it seems.
I asked Anderljung why it is so important not to impede the innovation of corporations like Microsoft and Google. Surely, they can look after themselves without copyright conventions being dismantled so vendors can no longer be accused of theft? He said:
I think those concerns make total sense. However, one thing I’ve noticed is that currently, copyright law in the UK is unclear in terms of what duties companies have, and in some really unhelpful ways. So, if I am OpenAI, for example, and I've trained on data that is based in the UK, it is unclear if me hosting a copy of this model in the UK, in itself, makes it a copyright infringement. If there's some kind of breach of rights here, which I think is reasonably plausible, it's on the level of the system being trained. Some of those kinds of things need to be changed.
Who benefits?
Changed for whose benefit? The answer, implicitly, is AI companies. But why are their wishes paramount? On this specific point, Anderljung appeared to be speaking more like the emissary of an AI vendor, and less like the spokesman of a governance organization. This was a muddying of legal waters that are, in fact, perfectly clear.
This was established beyond reasonable doubt by the the UK Governent's House of Lords’ cross-party Communications and Digital (select) Committee last year. Having heard from all sides of the debate, including AI companies and their advocates, plus copyright experts, specialist lawyers, and industry organizations, the Lords’ judgement was stark: scraping copyrighted content for the purposes of training a commercial product is not acceptable. Period.
However, as the Committee later observed, then Prime Minister Rishi Sunak’s administration sat on its hands and did nothing, claiming that there were arguments both for and against – arguments that the Lords had already heard, of course, and which formed the basis of their judgement.
Chair of the Committee Baroness Stowell then accused Whitehall of tacitly siding with AI vendors by refusing – or neglecting – to act. As I explored in my recent report on the Hiroshima trade and technology Accord between the UK and Japan, the Sunak government’s inaction on US companies’ is in stark contrast to its attitude to IP theft by China and others, which the Accord condemns in forthright language.
Anderljung continued:
Then there is what one might call a realpolitik kind of concern. Currently, it is the case that companies that would otherwise be based in the UK, and would train their models in the UK, go to other jurisdictions, because the copyright laws are sort of clear and whatnot.
This is nonsense. Most of the companies questioned by the Lords Inquiry are American. They go where they are already based, and where copyright laws happen to be more ambiguous. This is precisely why a raft of lawsuits and class actions is ongoing in the US: litigants are seeking a definitive ruling on whether OpenAI and others’ use of unlicensed, copyrighted material – its wholsesale scraping in fact – constituted theft, as they believe.
Anderljung added:
And so, even if the UK sort of imposes requirements that are sort of more in some sense sort of respecting copyright holders more, it's not clear that will actually help those copyright holders, because actors [AI companies] might just go to other jurisdictions.
Once Anderljung had found the right words – he holds a Masters in the History and Philosophy of Science and Technology from Cambridge University, incidentally – this was a revealing comment.
Once again, it implied that his organization sees these issues through the lens of an AI company being the most important actor in the room. But copyright holders don’t just want to be paid by AI companies for the use of their proprietary content; they are also trying to protect their careers from being dis-intermediated by trillion-dollar corporations. It is about fair play, respect, ethics, and rejecting a wholesale abuse of trust.
Arts were low-hanging fruit; AI companies grabbed it because they thought they could get away with it for long enough for their actions to become a legal fait accompli, and the law would have to be retrospectively changed.
But that has not happened – yet. And it will only happen if people fail to protect important laws and long-established conventions – bearing in mind that, historically, copyright has always encouraged and protected innovation, never impeded it. After all, it is one reason why Big Tech companies spend millions of dollars suing each other!
Solution?
So, what is the solution in Anderljung’s view? He said:
My guess is that the sort of thing the UK should think about instead, in the copyright space, is a development that is maybe similar to what we saw in music, digital rights management, and whatnot. We are maybe in the Napster era of AI and copyright. And over time, I guess we'll start moving into more of the Spotify era, where it is possible for AI companies to, in an easy way, compensate some of the copyright holders.
At this point, I can hear generations of musicians screaming at their computers that the last thing they, or any other creative people, want is a Spotify-style solution, in which a platform makes billions of dollars while music makers receive as little as $0.001 per stream (far less in many instances). A solution that makes creators into micro-serfs of powerful, self-appointed data landlords that encourage people to consume other peoples’ content for free.
Anything but simply not steal people’s creative work in the first place. Remember: generative AI systems are not brilliant machine intelligences capable of making original work; they are merely recycling the data of talented human beings and selling it.
Such outcomes are only inevitable if there is a mass surrender of human agency in these discussions, which would be both craven and morally reprehensible. It also demands that a government sits on its hands and refuses to enforce copyright law, alas.
So here we are.
The grim reality is we face years more of this type of ‘opinion leadership’ from consultants and organizations that didn’t even exist until recently, and yet already have a platform at policy forums.
So, who or what is ‘the Centre for the Governance of AI’, a self-named entity with more than a touch of hubris, it seems, given that it is not a wing of government or of any other international organization. Despite that, it is ‘the Centre for Governance of AI’ no less. Says who? Says the Centre; it appointed itself.
The answer is that this descendant of the defunct Future of Humanity Institute at Oxford Universty is – temporarily it says – “fiscally sponsored” as one of a portfolio of ventures by a UK-registered charity, The Effective Ventures Foundation.
In 2023 alone, that charity received nearly £62 million in donations (roughly $80 million). But from whom? Two of its three Trustees were only appointed in 2024, and the other last year.
Who pays these entities, why, and using what money is an urgent question for policymakers to answer. Either way, the Centre is still establishing itself as an independent organization, and yet it is already at the heart of Westminster policymaking, giving keynotes about what we must do make life easier for Big Tech.
At $3 trillion, OpenAI investor and partner Microsoft is currently worth roughly the same as the entire UK economy; Google is worth $2 trillion, Amazon nearly as much, and Meta just north of $1.1 trillion. But apparently, we are inconveniencing them.
We really do need to stop caring about that.
My take
Alas, the queue of self-appointed consultants stretches into infinity, like those ‘policy institutes’ that are actually covers for Washington lawyers covertly representing multi-nationals (see diginomica, passim). Those types of organization, specifically, exist to nudge opinion towards acceptance of supposedly inevitable outcomes.
As for the Centre for the Governance of AI itself, its aims may be entirely benign, humanitarian, and philanthropic – there is no evidence to the contrary, and it conducts in-depth research on AI risk. But it is far from clear why, on copyright at least, its Director sounds like an emissary of vendors who want the law changed. I urge it to think again.
(I note for the record that its fiscal sponsor The Effective Ventures Foundation pays out nearly 100% of donations, and its portfolio includes The Centre for Effective Altruism, 80,000 Hours, Non-Trivial, and other organizations that present as being socially conscious.)
On AI specifically, the reason these matters are enraging to many people is that while the technology could help humanity reverse climate change, avert nuclear catastrophe, cure cancer, or use energy more efficiently – notwithstanding the inconvenient truth that data centres already use more energy than the world’s fourth largest economy, Japan – at present it is mainly being used to put creatives and knowledge workers out of work. Or at least, to permanently commoditize definitively human skills and turn them into AI vendors’ revenue streams.
Apparently, the world’s most urgent problem for AI to solve is artists, that most human of all activities. And we are told that copyright holders are impeding the only thing some people seem to care about on this burning planet: innovation and growth. The hell they are. They just want to be treated fairly.
Plus, we need to call time on vendors and politicians saying “innovation, innovation, innovation” as though it means something. Exactly what are companies referring to when they use that debased word? An idea, a whim, a monetized product, or some IP of their own, perhaps? Precisely how does obeying the law impede it – doesn’t that make it a bad idea? And can eight of the top ten most valuable companies in human history meaningfully claim obstruction when they have scraped copyrighted data – reams of it – without permission?
Given that the long-term impacts of AI are completely unknown, the risks unpredictable, and the safety questionable enough to host conferences on it and create international safety institutes, why are governments so willing to bow to AI vendors’ wishes? Are we so desperate for growth – and a productivity uplift that never appears – that we will now sacrifice centuries of IP convention to realise it?
We need to reset the debate in favour of fairness. Stop handing control over these decisions to self-appointed consultants – whose role in this, collectively, is at best deeply questionable – and let the rest of us speak.
