For a still nascent technology, generative artificial intelligence already has an impressive resume. It can compose music, summarise wads of legal documents in seconds and generate television adverts based on minimal descriptive input. To become even cleverer, weed out errors and broaden its uses, AI models will need to continuously ingest human-generated content to train on. But the legal framework required to facilitate this symbiosis between man and machine has fallen woefully behind. That puts the long-term development of the technology, and the individuals and companies who feed it with unique data and insights, in harm’s way.
Generative AI models owe their capabilities, so far, to the reams of text, sounds, images and videos posted online. Much of this has been scraped without the consent of the original creators. A lack of clarity over how copyright laws apply to gen-AI training has also fomented protests and litigation battles around the world. Model developers tend to argue that “fair use” exemptions, which allow the use of copyrighted material under specific conditions, for instance by researchers using short, cited excerpts, are applicable. Artists, musicians and the media strongly disagree. They allege that AI companies are breaching their rights to intellectual property protections, since they go beyond merely excerpting their data.
With legal cases ensuing across America and disagreements in Europe over how the EU’s AI Act applies, Britain has taken a welcome initiative to end the ambiguity. Last week it closed a consultation into plans for the future of copyright and AI. But the UK government is also caught between wanting to be attractive for AI companies to scale and drive economic growth, while also protecting its world-class creative industry.
Though Prime Minister Sir Keir Starmer last week suggested that plans are not set in stone, the consultation did indicate that the government favours allowing AI firms to use copyrighted work to train their models without consent, unless the owner opted out. That approach would be a mistake. It tilts the playing field against content creators, overturning a default right, which has stood for centuries, that no one should be able to profit from another’s established IP. Though the opt-out mechanism is used in the EU, the systems needed to process and enforce them, across numerous platforms and use cases, are patchy.
Legislators around the world should recognise that building a sustainable and fast-growing gen-AI ecosystem depends on the strength and trust of those that produce the source data. Indeed, enabling tech firms to absorb their content, against their will, to build highly scalable competitors against them, undermines the creative and innovative incentives of individuals and companies in the first place.
There is a better way forward: supporting licensing markets. Remunerated consent between creators and AI companies gives content makers control over their copyright (it is opt-in by design) and compensation for their work, which incentivises their efforts. It also gives AI models sustained access to high-quality data, free from legal wrangling. Many creative businesses, including this newspaper, have already struck individual content licensing deals with AI companies. Moving from ad hoc deals to a broader market for training licenses is the next step. Governments can help by supporting industry-led transparency standards for how training data is used alongside the development of software to process and track licenses.
As it reviews its consultation responses, Britain’s government now has an opportunity to set a global standard for how AI and human creativity can coexist. If it wants to create a competitive environment to attract AI companies that actually endures, developing a free and fair market for data is a win-win solution.