Forcing UK creatives to ‘opt out’ of AI training risks stifling new talent, Cambridge experts warn

The UK government should resist allowing AI companies to scrape all copyrighted works unless the holder has actively ‘opted out’, as it puts an unfair burden on up-and-coming creative talents who lack the skills and resources to meet legal requirements.

This is according to a new report from University of Cambridge experts in economics, policy and machine learning, who also argue the UK government should clearly state that only a human author can hold copyright – even when AI has been heavily involved.

A collaboration between three Cambridge initiatives – the Minderoo Centre for Technology and Democracy, the Bennett Institute for Public Policy, and ai@cam – the report argues that unregulated use of generative AI will not guarantee economic growth, and risks damaging the UK’s thriving creative sector. 

If the UK adopts the proposed ‘rights reservation’ for AI data mining, rather than maintaining the legal foundation that automatically safeguards copyright, it will compromise the livelihoods of many in the sector, particularly those just starting out, say researchers.

They argue that it risks allowing artistic content produced in the UK to be scraped for endless reuse by offshore companies.

“Going the way of an opt-out model is telling Britain’s artists, musicians, and writers that tech industry profitability is more valuable than their creations,” said Prof Gina Neff, Executive Director at the Minderoo Centre for Technology and Democracy.

“Ambitions to strengthen the creative sector, bolster the British economy and spark innovation using GenAI in the UK can be achieved – but we will only get results that benefit all of us if we put people’s needs before tech companies.”

'Ingested' by technologies

Creative industries contribute around £124.6 billion or 5.7% to the UK’s economy, and have a deep connection to the tech industry. For example, the UK video games industry is the largest in Europe, and contributed £5.12 billion to the UK economy in 2019.

While AI could lead to a new generation of creative companies and products, the researchers say that little is currently known about how AI is being adopted within these industries, and where the skills gaps lie.

“The Government ought to commission research that engages directly with creatives, understanding where and how AI is benefiting and harming them, and use it to inform policies for supporting the sector’s workforce,” said Neil Lawrence, DeepMind Professor of Machine Learning and Chair of ai@cam.

“Uncertainty about copyright infringement is hindering the development of Generative AI for public benefit in the UK. For AI to be trusted and widely deployed, it should not make creative work more difficult.”

In the UK, copyright is vested in the creator automatically if it meets the legal criteria. Some AI companies have tried to exploit ‘fair dealing’ – a loophole based around use for research or reporting – but this is undermined by the commercial nature of most AI.

Now, some AI companies are brokering licensing agreements with publishers, and the report argues this is a potential way to ensure creative industries are compensated.

While rights of performers, from singers to actors, currently cover reproductions of live performances, AI uses composites harvested from across a performer’s oeuvre, so rights relating to specific performances are unlikely to apply, say researchers.

Further clauses in older contracts mean performers are having their work ‘ingested’ by technologies that didn’t exist when they signed on the dotted line.

The researchers call on the government to fully adopt the Beijing Treaty on Audio Visual Performance, which the UK signed over a decade ago but is yet to implement, as it gives performers economic rights over all reproduction, distribution and rental.

"The current lack of clarity about the licensing and regulation of training data use is a lose-lose situation. Creative professionals aren't fairly compensated for their work being used to train AI models, while AI companies are hesitant to fully invest in the UK due to unclear legal frameworks,” said Prof Diane Coyle, the Bennett Professor of Public Policy.

“We propose mandatory transparency requirements for AI training data and standardised licensing agreements that properly value creative works. Without these guardrails, we risk undermining our valuable creative sector in the pursuit of uncertain benefits from AI."

'Spirit of copyright law'

The Cambridge experts also look at questions of copyright for AI-generated work, and the extent to which ‘prompting’ AI can constitute ownership. They conclude that AI cannot itself hold copyright, and the UK government should develop guidelines on compensation for artists whose work and name feature in prompts instructing AI.

When it comes to the proposed ‘opt-out’ solution, the experts it is not “in the spirit of copyright law” and is difficult to enforce. Even if creators do opt out, it is not clear how that data will be identified, labelled, and compensated, or even erased.

It may be seen as giving ‘carte blanche’ to foreign-owned and managed AI companies to benefit from British copyrighted works without a clear mechanism for creators to receive fair compensation.

“Asking copyright reform to solve structural problems with AI is not the solution,” said Dr Ann Kristin Glenster, Senior Policy Advisor at the Minderoo Centre for Technology and lead author of the report.

“Our research shows that the business case has yet to be made for an opt-out regime that will promote growth and innovation of the UK creative industries.

“Devising policies that enable the UK creative industries to benefit from AI should be the Government’s priority if it wants to see growth of both its creative and tech industries,” Glenster said.