a day ago
The government is moving too slowly on AI
Photo by Jack Taylor / Getty Images for SXSW London
When I first started grappling with the problem of AI and copyright as a minister in 2023, I grossly underestimated the difficulty: the two sides could surely be brought together to find compromises and workarounds. Sadly, it became ever clearer over months of talks that there was no mutually agreeable landing zone – we could satisfy the content creators or the AI labs, but not both.
As the government approaches the first anniversary of its own forlorn attempts to find the elusive middle ground and temperatures rise on both sides, I've been urging them to take a different approach.
First, let's think more clearly about where we want Jonathan Camrose Shadow minister for science, innovation and technology to get to. The eventual answer is surely a trusted, efficient marketplace for the use of copyrighted materials, where rights-holders can freely choose to license, sell or withdraw their property and developers can make rapid and affordable commercial choices. Prices will be set by supply and demand, with smaller rights-holders represented by collectives who process transactions on their behalf and distribute revenues.
Second, what are the barriers that prevent us building such a marketplace? Among a great many, two stand out: offshoring and transparency. The UK can make any laws it likes, but any AI developer who doesn't like them can offshore training activity to a jurisdiction where they can legally conduct the same training with the same content. Not only would the problem remain unsolved for our rights-holders, but we would be pushing AI activity out of the country.
Which brings us to transparency: there is a view that we could require any AI model used in the UK to declare all of the content used in its creation, thus solving the offshoring problem. I'm afraid this is wishful thinking.
How would this vast quantity of information be sought, verified and audited? How would we enforce it? How would we establish a direct link between the expressed output and the suspected input? What if the training material is not copyrighted but an imitation of an imitation (and so on) of copyrighted material?
I don't believe there is a form of words that can be made into a law that fixes AI and copyright in a way that satisfies all sides. I do, however, believe that a combination of technologies, standards and law can help us build the marketplace for copyright.
Subscribe to The New Statesman today from only £8.99 per month Subscribe
The key technology is going to be machinereadable digital watermarking that can be indelibly embedded in any file and would contain licence information.
Crucially, rights-holders would need to easily (or automatically) apply watermarks to all of their material. This starts to address the transparency problem, but it only works if everyone agrees to use the same design – or at least to choose from a limited number of designs. It would depend, in short, on the existence of globally agreed technical standards – hardly a novelty in the internet age. Armed with such standards, governments would be far better placed to make the laws to create a fair and trusted marketplace for copyrighted materials. Globally standardised digital watermarks solve the transparency problem and transparency solves the offshoring problem.
The government is moving too slowly. Technology is driving change more quickly now – it's time for boldness and agility. No country has solved this problem, and if we can then we have a chance to be a global leader in AI again.
This article first appeared in our Spotlight on Technology supplement, of 13 June 2025.
Related
