Contribution
I rise not just as a member of the Science, Innovation and Technology Committee or as co-chair of the APPG on artificial intelligence, but as an academic with first-hand experience of building AI models. I will try to make a case—go with me on this one—that protecting our creative industry and the high-quality output it produces is vital not just to our vibrant creative sector, but to our rapidly evolving AI sector.
Back in the mists of time, when I was building a model for my PhD, I resisted the temptation to build a large dataset from any relevant data I could find. Instead, I chose to produce a smaller, high-quality dataset, which was risky, as it might not have been enough to create a reliable model and I might not have passed my PhD. My risk paid off. My model worked better than expected and, ever since, I have adhered to the principle that it is the quality of our data, not its quantity, that matters.
To mangle a metaphor, some say that data is the new oil or gold. Taking that metaphor forward, let us do a thought experiment. Let us say that the UK discovered vast reserves of gold, making us the second biggest provider globally. What should we do? Would we look at gold-hungry organisations and give them the gold for free, in the hope that they will invest in the UK? I should hope not.
By some estimates, the UK creative industry is the second largest globally. It is our gold. Should we give away this valuable asset for free? I hope not, for the sake of our creatives, but also for our new AI industry, in which this Government are rightly investing. The unlicensed and illegal use of copyrighted content for generative AI development has been equated by some with a form of theft. Not only is it unfair to make such acquisition legal via an opt-out system, but it risks creating a future of fool’s gold data as our creative industry loses control of its work and moves elsewhere, or simply gives up.
High-quality data is essential for the success of generative AI, but, as with gold—to overkill the metaphor—its reserves may be finite. Researchers predict that at the current rate, generative AI developers will have used all the publicly available stock of human-created text data between 2026 and 2032. In other words, we could run out next year. This means that, with limited new high-quality data, innovation and AI growth will be hindered, inviting model collapse.
Model collapse happens when generative AI models start training on their own lower-quality data. This is the fool’s gold of AI. If we wish to be competitive, both in the creative and AI industries, we as a country need to set ourselves above our competitors, not below them. We do this not by giving away our most valuable assets for free, but by protecting them so that we can keep generating more and more new high-quality data.
Generative AI developers, like all AI developers, are extremely data-hungry and keen to mine, mine, mine. If all we have left to offer them is fool’s gold—if our world-class British artists have to look elsewhere to make a living—the AI developers will likewise look elsewhere. Protecting our creative industry by not allowing the free use of data for model training purposes is therefore the right thing to do not only for our creative industry, but for our AI industry.
I therefore ask the Government to consider a longer-term view and retain the UK’s current copyright framework, to place the onus on generative AI developers to seek a licence for our creatives’ data—with a possible caveat for academic research—and to expand it to cover all generative AI models marketed in the UK. In conjunction, we must require meaningful transparency on data usage in a form that is accessible to artists and regulators. That would allow for enforceable regulation and enable the actual data creators, our creatives, to seek redress. I resist, as I always do, claims that that would inhibit innovation and growth. Based on my experience as a researcher and AI ethicist, I reject the notion that regulation inhibits innovation. It simply does not. Transparency of model development and data sources plus enforceable regulations are vital to encourage high standards and good quality AI.
By retaining our copyright framework and pioneering a licensing approach, we would guarantee that rights holders have control over their own creative output. Such an approach would give confidence to our creatives, allowing them to pursue sustainable, well-paid and productive careers, and provide enjoyment to us all, income and growth for the country and the high-quality output that is so valuable to generative AI developers. By taking that approach, we will remain what we are today—the gold standard of the creative industries—and that will enable us to rise up as the gold standard for AI development.