Tech secretary rows back on Data Bill rights for AI companies
The UK government is rethinking proposals to allow AI companies to train models on copyrighted work without permission.
The government is rethinking its controversial new copyright proposals, which would enable AI companies to use copyrighted material to train models without copyright holders’ permission.
Those blanket rights would be tempered only by an opt-out for copyright holders, although it’s unclear how that would work in practice.
The proposals were to feature in the upcoming Data (Use and Access) Bill, intended as a package of measures to encourage AI growth in the UK, which also includes AI Growth Zones, intended to turn depressed regions of the UK into AI hubs. The Growth Zones would fast-track the infrastructure required to support energy-intensive AI data centres.
However, the proposal to let AI companies use copyrighted material has been strongly opposed by artists, which have lobbied government pointing out the economic cost of such an idea to the UK’s creative industries.
The Guardian, citing “a source close to Peter Kyle”, the technology secretary, suggested that the copyright holders’ opt-out was just one of several proposals he was considering, and no longer his preferred option.
Tacitly indicating that The Guardian’s “source” was Peter Kyle himself, the newspaper quoted the technology minister as saying: “We’re listening to the consultation and we are absolutely determined to get this right…
“We can’t pretend we can outlaw training in other countries who have their own copyright law, but we can build a system which works in the United Kingdom. Some elements of the consultation haven’t even featured in the debate so far.
“We will be working hard to come up with practicable solutions to the very complex issue of how we enable both the creative industries and UK AI companies to flourish. We will report to parliament on issues of transparency and licensing and try to find some common ground on which we can agree.”
Instead of broad rights and opt outs, ministers are now considering a way of encouraging licensing agreements between copyright holders and AI companies.
However, campaigners against the proposal remain sceptical over the government’s supposed change of emphasis.
Lady Kidron, a cross-bench peer in the House of Lords who has campaigned against the measure, pointed out that for Kyle’s new proposals to be meaningful “it must include an unequivocal commitment from the government to protect copyright holders.”
The fear is that copyright holders will only be superficially protected in the Data Bill when the eventual proposals emerge.
Copyright holders have already launched a slew of legal actions against AI companies over alleged breach of copyright, claiming that their material has been used to generate AI images, videos or music.
For example, Getty Images sued Stability AI for generating images that have simply been scraped from the company’s website – with the Getty Images watermarks even still visible in the AI-generated images.
Moreover, AI pioneer OpenAI has admitted it would be “impossible” to develop functional generative AI tools like ChatGPT without access to copyrighted material.
