For years AI companies trained on the world’s books, journalism, music, and art without asking or paying. This week Parliament, Meta’s lawyers, the Intellectual Property Office and, briefly, the Vatican all weighed in on whether that can go on.
Since generative AI arrived in public life, one question has hung over it more stubbornly than any other, which is what it owes to the people whose work it learned from. For most of that time the answer has been settled by default. The models were trained on whatever could be found, the companies that built them maintained that this was lawful, and the writers, journalists, artists and publishers whose work had been absorbed were left to find out after the event and argue about it in court.
It helps, when following all this, to think of content as land. For years it lay open like common ground, and the AI companies snuck across it, gathered whatever they liked and built very large and very valuable things with the harvest. The people who own the land have finally noticed, and the dispute that has followed has every ingredient of a proper English property row, with complaints about trespass, haggling over rent, a sudden enthusiasm for title deeds, and, rather lost in the noise, the question of what happens to the people who actually live there.
Last week gave a fairly complete picture of how things are starting to change. In the space of a few days, British publishers asked the Prime Minister to back legislation against covert scraping, Meta rewrote its commercial terms to give itself broad rights to crawl the websites of its business customers, Google signed paid agreements with CNN, Axios and others, an Oxford consultation urged the Government to build a proper licensing market, and the Intellectual Property Office approved a stamp certifying that a book was written by people. Taken together, they suggest the argument has moved on from whether AI companies should pay for what they use to the more practical matter of who sets the terms, and the uncomfortable answer, for now, seems to be ‘whoever moves first’.
The most immediate complaint is about who is wandering on to the property uninvited. Britain’s news publishers wrote to the Prime Minister this week asking for his backing for the Automated Online Software (Access and Transparency) Bill, which takes aim at what the trade has taken to calling stealth bots – the crawlers that harvest websites without announcing who sent them or what they are for – and the Bill was listed for its second reading in the Commons within days of the letter landing.
Meta, meanwhile, has been helpfully demonstrating why publishers are nervous. Under its revised commercial terms, from 30th October any organisation with a business account on Facebook, Instagram, or WhatsApp will be deemed to have agreed that the company may crawl its associated websites and use whatever it finds to train its products, and the only way to decline is to stop using those business tools altogether. Meta says nothing will change in practice and that publishers can still block its crawlers in the usual way, but the lawyers who read the terms for Nieman Lab pointed out that the clause hands over material that would cost many millions to license on the open market, which is a remarkable thing to acquire by updating a settings page.
The argument has also gone global, and picked up a theological dimension along the way. In Australia, where copyright reform is under review, Father Paolo Benanti, who advises Pope Leo on artificial intelligence, warned that letting foreign AI companies train on Australian work without permission would amount to a “new form of colonialism”, a phrase with the rare advantage, in technology policy, of being understood instantly by everyone in the room. It is certainly a strong phrase, but it points to something meaningful about the imbalance between the handful of companies that build these models and the far larger number of countries and creators whose work supplies them.
If trespass is the stick, rent (i.e., licensing) is the carrot, and last week there was much more carrot on offer than usual. CNN, Axios and a handful of other publishers signed up to Google’s AI programme for news, under which they will be paid for the use of their journalism, joining a lengthening queue of outlets that have concluded it is better to be paid something by the machines than to be read by them for nothing.
The difficulty here is that a market built one deal or one licence at a time tends to reward those with the size and leverage to negotiate. That is good news for the big newsrooms and much less so for the regional paper, the specialist publisher, or the individual author, who may have neither the lawyers nor the bargaining power to strike terms of their own.
This is the gap that The Oxford Consultation on Copyright and AI, published this week by the Institute for Ethics in AI, sets out to close. Drawing on more than forty contributors from the creative industries, technology, law, and academia, it argues that Britain could become the world’s marketplace for licensing material to AI developers, provided the Government stops dithering and sets clear rules on consent, transparency, and fair payment.
It is an attractive idea, not least because Britain happens to be unusually good at the two things such a market most needs – producing creative work and drafting contracts – and the report suggests the same machinery could one day serve industries as far from the arts as pharmaceuticals and defence. Its warning, however, is that the opportunity will not wait, and the Government’s record gives it reason to say so. Last year’s proposal to let AI developers use work unless creators opted out met fierce resistance and was dropped after a long stand-off between the Commons and the Lords, which left the underlying question exactly where it started.
There is a more hard-nosed reading of the economics as well. The Times made the case that Britain’s best route to an AI dividend may be to let Silicon Valley spend the eye-watering sums needed to build the models, and carry the risk that comes with them, whilst British companies such as Next get on with using them. The two positions are more aligned than they might look. A country that mostly adopts AI, instead of building it, still has every reason to want a fair and predictable licensing regime, because its own creators and companies are among those whose work the models depend on.
When the law moves slowly, people start drawing up their own paperwork, and the creative industries have been doing so with some vigour. The Society of Authors, working with the Creators’ Rights Alliance and partner bodies around the world, launched a global charter for the ethical use of generative AI, framed quite openly as an attempt to set standards in the vacuum left by governments that have yet to set any. Books by People, meanwhile, announced that the Intellectual Property Office had approved its new stamp certifying that a book was written by human beings, which means the next novel you pick up may carry the same reassuring label as a bag of carrots from the farmers’ market, and it can surely be only a matter of time before someone at a dinner party lets it be known that they read nothing but free-range fiction.
Still, none of this settles the legal question underneath – and in the largest market that question is still with the courts. In the United States, a federal appeals court last month upheld Thomson Reuters’ win in one of the first major AI training cases, book publishers are pursuing a class action alleging that Meta trained its models on pirated books, and a good many other suits are working their way through the system. Until judges decide whether training on copyrighted work counts as fair use, every licensing deal and every charter is being drawn up without knowing what the law will finally require. This is splendid news for lawyers and rather less splendid for everyone else.
It would be a mistake to treat all of this purely as a matter of rights and revenue. One of the most thoughtful pieces of recent weeks, an episode of The Daily from The New York Times, followed mathematicians as AI systems began to do work they had dedicated their careers learning to master, and asked what becomes of a person’s sense of purpose when that happens. Writers and artists came into it too, and it serves as a useful corrective to all the talk of licences and levies, because for most people who create things for a living the worry was never only whether they would be paid for their work. It was also whether anyone would still want them to do it.
For organisations outside publishing and the creative industries, it is tempting to see this as someone else’s problem. It isn’t. Almost every organisation now publishes huge amounts content, from research and reports to case studies and the thoughtful essays executives post on LinkedIn, and all of it is open to AI crawlers.
The questions publishers are wrestling with this month – about whether to protect their content, which bots to let in and, if so, how much to charge them – will reach corporate communications teams soon enough, quite possibly in the form of a policy change that arrives without much notice and even less explanation.
And if you’re wondering why I’m so deep in the detail on all this, it’s because of our work with the Copyright Licensing Agency (CLA), which licenses the reuse of published work on behalf of the authors, publishers, and visual artists who made it. For an organisation like that, the questions in this piece fill the working week, and these days they fill a good part of mine too.
Blog
