On 18 March, after more than two years of deliberating what to do about artificial intelligence and copyright, the UK the need to “gather more evidence on how copyright laws are impacting the development and deployment of AI across the economy and the economic benefits of reform”.

Perhaps the reason for this delay has been minsters’ growing realisation that the issues concerning AI and copyright are far more complicated than the “Big Music v Big Tech” paradigm being presented in the media. As this highlights, more than 1,000,000 UK businesses use machine learning but, contrary to the hype, most will not be using GenAI models and virtually none creating music. Furthermore, what stands out is the wide array of copyrighted works being used – purchased databases, websites and social media posts – as well as taxpayer-funded research published in scholarly journals.

In the modern world, where everything we do with online material involves a copy being made of it by a computer and its network, how we define the scope of copyright law has major consequences for our ability to harness the full potential of digital technologies. In particular, in a research-intensive economy, how flexibly we can use all the information and data we have access to in order to generate new ideas, products, services and productivity gains is a matter of broad industrial strategy. But in that regard, the UK, with its restrictive framework, is lagging behind many other countries.

As I have previously written, even if they have legal access to online material, UK businesses have to seek an additional licence to analyse the facts and data that appear in the work – even though, paradoxically, copyright law does not protect facts and data. Even non-commercial bodies such as NHS trusts or universities are required to re-seek permission for everything used if they wish to share with another trust, university or commercial entity any data they have created for AI training, or any outputs containing copyright-protected work. At any scale, and at any speed, any licensing professional will tell you that this is an impossible task.

̽����Ƶ

By contrast, Asian governments in particular are updating their copyright law as part of a broader industrial strategy, with the emphasis on helping organisations use copyrighted works to which they have lawful access. In 2004, Singapore modernised its law to match the US and introduced fair use – the flexible copyright exception that provides so much support for creative, research and technology interests there. Taking a different path, in 2018, Japan introduced a forward-looking and flexible copyright provision directly aimed at supporting its domestic STEM and AI industries.

More recently, significant interventions have been made by the Indian and South Korean governments. The former has roundly rejected voluntary and collective licensing, proposing instead a mandatory statutory licence whereby rights holders will have no option to withhold use of their works. Many reasons are given.

̽����Ƶ

First, the Indian government’s highlights the impossibility of clearing rights at any scale and the vast range of copyright materials that are used in machine learning.

With a firm eye on the need for a competitive economy to encourage new market entrants, the report also highlights how a voluntary licensing regime cements incumbents with the deepest pockets. Given that AI, and IT markets more broadly, are already characterised by a tendency to create oligopolies, the need to clear legal obstacles and support the path to success for start-ups and scale-ups is clear.  

Another important aspect of the debate that the Indian paper faces up to better than the UK has is the link between licensing, the constraints this imposes on training datasets, and the risk of AI bias. As the report notes, “quality data from diverse sources is needed to develop good AI systems”.

South Korea introduced fair use into its copyright law in 2011, no doubt linked to its strong drive to promote its own AI industry. However, last month, its government brought out a guidance document explaining how copyright law may be interpreted to facilitate generative AI.

̽����Ƶ

Reflecting a similar position from the Israeli Ministry of Justice, it sets out that when the outputs are transformative and not intended to be substantially similar to the original work, generative AI is likely to be fair use. This is also a line of reasoning we see emerging from two recent important cases in the US involving and .

If the UK wants a vibrant research-intensive digital economy, it needs to learn from these examples and adopt either fair use rules or, like India, a compulsory licence calibrated to an organisation’s size, position and economic sector. Certainly, no licence fee should be payable at all for university or NHS research.

Some academics might object to their work being used to train commercial AI but, increasingly, academics are using AI in their own research and working life. Moreover, if the chief argument for the research budget is its long-term economic impact, there can surely be no case for stifling that impact by being precious about the uses of work we have authored but the taxpayer has funded.

Brexit allows such choices to be made – the question is whether the government values growth, science and innovation enough to make them.

̽����Ƶ

is a PhD student in the Centre for Intellectual Property Policy and Management at Bournemouth University. He was formerly head of intellectual property at the British Library.