Copyright and AI: Response by the CREATe Centre to the UK Government's Consultation

This submission responds to the UK government’s call for views on reform that ensures “that the UK’s legal framework for AI and copyright supports the UK creative industries and AI sector together”. The proposal of the government is for “a data mining exception which allows right holders to reserve their rights, supported by transparency measures”.

• In our view, the reservation of rights (opt-out) is difficult to implement technically, will increase costs and create hurdles to market entry. The default should remain an opt-in framework. However there needs to be much clearer scope for permitted research.
• We argue that the current data mining exception (s29A CDPA 1988: Copies for text and data analysis for non-commercial research) should be broadened to all research before market entry, at which point transparency and licensing obligations will kick in.
• In parallel, an equitable remuneration provision should be introduced that enables creatives (authors, artists, performers) to receive a share of licensing revenues negotiated by intermediaries (publishers, producers) with AI developers.

The submission first reviews the evidence relating to current workings of the data mining exception. It is a key obstacle of s29A that it does not allow knowledge transfer in partnerships between different academic institutions, let alone public-private partnerships and sharing training datasets with the wider research community. There is also uncertainty with respect to the database right, which presents considerable risks for research.

We then review the evidence relating to existing commercial agreements between rightsholders and AI developers for the training of models. Because of the need to access quality data and the legal uncertainty in major jurisdictions, there is a global move towards a licensing economy, regardless of the UK’s policy choices. We also find that the earnings of creatives have declined dramatically over the last decade, and that income from the new kinds of licensing deals does not translate transparently into royalties for authors and performers.

The third piece of evidence relates to the arrival of powerful open-source models that can be run locally (even on a laptop) rather than via platform type services. On the intermediary side (frameworks and computation tools such as Ollama and Hugging Face) there is currently no clear liability regime for harms resulting from models made available. Copyright enforcement for locally-trained models prior to commercialisation is challenging.  

Based on this evidence the submission then offers a detailed legal analysis that seeks to support three goals:

• removing obstacles for scientific research and education, recognising the specific imperatives of the humanities, social sciences and cultural sectors;
• improving and clarifying the licensing environment, so that access to quality content for R&D becomes easier; this must include transparency requirements;
• improving revenues of creatives (authors, artists, performers) as well as recognising their moral rights.

We propose solutions for a clearer data mining exception that avoids the fork of commercial vs non-commercial research required by EU law; for transparency obligations (addressing the vexed issue of extraterritoriality); and for introducing an equitable remuneration right for authors and performers. We also discuss the status of computer-generated works, terms of use of AI providers, and image rights.

Achieving a UK copyright environment that is “at least as attractive as in competing jurisdictions” requires a nuanced policy response.