UK government rethinks plans for AI-training copyright exception


One of the big debates about creative AI technology at the moment concerns how these models are trained – and specifically whether, if they’re being trained on the commercial work of human artists, designers and musicians, whether they need proper licensing deals that pay royalties.

Last year, the UK emerged as a lightning rod for this debate, due to government consultation on proposals to provide a copyright exception for ‘text and data mining’ – meaning no permission would be required by developers training their models on copyright-protected works.

Cue a swift and unified backlash from British music industry bodies, later expanded on by BPI boss Geoff Taylor, who said the proposal “would essentially mean that anyone can train their AI on any music that’s ever been made, without seeking permission from anybody involved in the creation of that music”.

Rightsholders are also preparing their arguments against licence-less training. “The industry needs to be paid and our artists need to be paid based on the AIs that are learning off their music,” said Warner Music Group’s chief digital officer and EVP, business development Oana Ruxandra at the NY:LON Connect conference last month.

So, big news. In a debate about artificial intelligence and intellectual property rights in the UK’s parliament yesterday, minister for science, research and innovation George Freeman announced that the government is going back to the drawing board on its proposals.

“We have written around to make it clear to other Ministers that the proposals were not correct, that we have met with a huge response, which should have been picked up in the pre-consultation before the proposals were announced, and that we are looking to stop them,” said Freeman.

He promised “a rather deeper conversation” in the coming months with politicians, creative industry bodies, creators and AI companies to “ensure that we do not rush precipitately into a knee-jerk move that is wrong”. British readers may enjoy a wry smile at his follow-on comment that “one of the lessons from this is to try not to legislate in periods of political turmoil”.

In any case, it’s exactly what the music industry was hoping for. “The whole music industry has been united in its opposition to these proposals, which would have paved the way for music laundering and opened up our brilliant creators and rights holders to gross exploitation,” said UK Music boss Jamie Njoku-Goodwin. “We are delighted to see the back of a policy that risked irreparable damage to the global success story that is the UK music industry.”

This is just one country, although there may be a ripple effect around the world where other policymakers are considering their options on this issue. One longer-term challenge is that if legislation varies, and some countries DO provide the kind of copyright exception that’s now been rejected by the UK’s government, the training might just move to those markets.

Enough policy though, let’s talk partnerships. The positive future for creative AI is about collaboration between the creative industries and the technology firms (be they giant or startups) who are building creative AI models.

Permissions and licensing deals, as well as opt-outs for artists who don’t want their music to be used for training, will hopefully be at the heart of those partnerships, rather than a grudgingly-accepted legal obligation. The good actors in creative AI technology will be getting ahead of the legislation by figuring out those deals.

Written by: Stuart Dredge