Blog post by Stephen Wyber
European Commission Plans to Boost AI Must Not Ignore Efforts to Restrict Access to Data for Research
Two Conflicting Narratives on AI
There often seem to be two very separate discussions going on about Artificial Intelligence (AI).
On the one hand, there is the one, typically linked to debates around growth, competitiveness, and research, that emphasises the potential of the technology to advance a whole series of economic, societal, and scientific goals, and explores how to realise this.
And on the other, there is the one, usually closely linked to copyright law debates, that tends to see AI as a threat to the entertainment sector, and which argues that it must be either curtailed, or at least that ways must be found to generate windfall payments. This – it is argued – could come through the establishment of a licensing market for the rights to use (already legally accessed) materials for training.
While this is not the impression given by the media, this second discussion is in fact a very narrow one, a priori focusing on Generative AI models which produce creative outputs that potentially compete with human creators. These are just a subset of Generative AI in general, which in turn is just a subset of the vast array of Artificial Intelligence uses.
These two discussions rarely seem to interact, leading to conflicting policy agendas, with the EU supporting AI development on the one hand (through investment and other new initiatives), while holding it back with the other (by placing limits on the pool of material available for training, and so reducing the quality of results).
For the sake of research and scientific progress – and the wider benefits this brings – it is important to resolve this situation. The good news, of course, is that the dichotomy between a thriving AI sector and a thriving entertainment sector is a false one. But decisive action will be needed to ensure that we can achieve a more positive situation.
The AI Continent Action Plan: An Opportunity for Resolution?
One opportunity comes with consultations on the EU’s AI Continent Action Plan. These have their roots in efforts to realise the potential of the technology, and ensure that Europe keeps up with its international competitors. It sets out plans to invest in infrastructure and skills, as well as to simplify the implementation of the AI Act.
It does come close to the copyright-related debate when it talks about the need to increase access to ‘quality’ data, recognising that this is a key factor in enabling the training of high-performing AI. Quality (or veracity) is, of course, a factor, although what is needed will vary according to the need in mind. So too is velocity (processing power – addressed by infrastructure investment), but also volume. Whether the proposed steps really guarantee the right data, in the right quantities for the purpose is an open question.
There is also Apply AI, which looks to deliver on another priority from the AI Continent. This focuses on promoting uptake of AI in different sectors, with a particular emphasis on science. This recognises the urgency of the adoption of AI in order to accelerate progress, highlighting in particular the most recent Nobel Prize for Physics and Chemistry, which specifically went to researchers using AI. It also, of course, highlights questions around how to integrate it effectively into the research process, and notably rules around privacy, ethics, and more.
Here too, the question of the impacts of intellectual property rights is raised, without necessarily being addressed head-on. This is despite the growing evidence of rightholders seeking to limit text and data mining (and so AI) activities in research settings through licence terms, and even trying to prevent the use of common tools such as Copilot.
A False Dichotomy?
By touching on these questions, these documents open an opportunity to be clear about what researchers need, regardless of whether they are working purely in the public sector or engaged in research-industry collaborations.
Restrictions on possibilities to use legitimately accessed materials need to be recognised for what they are: restrictions on possibilities to advance science and innovation, and only used proportionately, for example, when ethical or privacy concerns are at stake. It is also vital to make sure that the arguments of the entertainment sector do not end up harming Europe’s long-term competitiveness and ability to provide for its population.
Crucially, there is no reason to think that such restrictions will even bring benefits to the creative sector. As highlighted, licensing schemes seem unlikely to benefit any more than the biggest players, while creating insurmountable barriers for science. Those countries that have taken a less reductive approach to AI – the US, Japan, Korea, Israel – have certainly not seen any obvious harm to their creative sectors.
A better way of achieving legitimate cultural policy goals would be to provide targeted support, potentially paid out of tax on the profits from AI companies themselves.
We therefore encourage readers to read about the EU Action Plan, and make the most of the opportunity to share views and experience, and in particular to be clear about the need to avoid unreasonable restrictions on data availability. The deadline for the survey on AI in Science is 5 June 2025.
Read more in our AI Primer and AI Principles.
17 April 2025
