UK Spring budget: IP vs generative AI – back to the drawing board

The UK Government has endorsed recommendations to announce a clear policy position on the relationship between IP law and AI. However, a solution that pleases both the tech and creative industries will be hard to find, as shown by the backlash to recent proposals to permit commercial text and data mining.

Background 

In 2022, Ministers claimed that “unleashing the power of AI is a top priority in the plan to be the most pro-tech Government ever”. In the recent Spring 2023 Budget speech, UK Chancellor Jeremy Hunt doubled down on that sentiment, promising various measures to strengthen the UK’s position on AI. This included a promise to deliver on Sir Patrick Vallance’s recommendation, in his “Pro-innovation Regulation of Technologies Review”, to announce a clear policy position on the relationship between IP law and generative AI.

This recommendation stems principally from the fact that using text and data mining methods on publicly available content without permission is currently unlawful in the UK, and it is unclear if and how that might change.

Why is TDM so important for AI? 

Text and data mining (TDM) means using computational techniques to analyse digital content. TDM methods are significant in AI development, as mined datasets are widely used to train AI systems, e.g. machine learning solutions that rely on access to high quality data to learn (like ChatGPT).

However, TDM methods typically involve copying the underlying content to create these datasets and can therefore infringe copyright. Currently, in the UK, TDM can only legitimately be performed on third party content for non-commercial research or with permission, e.g. under a bespoke licence or open access scheme. 

Déjà vu? 

In June 2022, following its second consultation on AI and IP, the UK Intellectual Property Office (UKIPO) proposed broadening existing copyright exceptions to permit any commercial TDM. Under that proposal, rightsholders would not have been able to opt-out, contract-out, or charge for licences for use of TDM methods on their lawfully accessed content.

The proposal was welcomed by the tech sector but received forceful criticism from the creative industries as it would weaken copyright and undermine licensing opportunities and was a “landgrab” going beyond even what tech companies asking for. Ultimately, the Government confirmed that the proposal had been axed.  

So what now? 

The UK Government has endorsed Sir Valance’s recommendation to announce a clear policy position on the relationship between AI and generative IP. It also published a separate response to it. The response says that the UKIPO will, with stakeholders input, produce a code of practice “by the summer” to support AI firms in accessing copyright works as an input to their models. Moreover, “An AI firm which commits to the code of practice can expect to be able to have a reasonable licence offered by a rightsholder in return”. This may be followed up with legislation if the code is not adopted or agreement not reached. It will also support rightsholders by ensuring there are “protections (e.g. labelling) on generated output”.

So, the UKIPO must, by the summer, come up with a solution that removes barriers faced by AI firms in accessing copyright content, but also satisfies the creative industries that their rights and business models will not be undermined. This will not be easy.

Comment

There is urgency here. Sir Vallance notes that there is a 12–24-month window to “make the UK one of the top places in the world to build foundational AI companies, with other countries moving faster to provide clarity and a friendly regulatory environment for innovators”. He also acknowledges, crucially, that restricting training data access would put the UK at a disadvantage. Essentially, if other countries provide a more tech-friendly regulatory environment, including by permitting commercial TDM, that’s where AI companies will go.

Given the Government’s most recent response (and express reference in it to “licences” for TDM), it appears that an improved licensing environment is its current preferred solution. This might entail collective (or even compulsory) licensing or model licensing terms, all of which could simplify the process for engaging with multiple rightsholders.

This response also suggests that the UK does not plan to adopt the EU’s chosen solution, i.e. to permit commercial TDM but to allow rightsholders to opt-out, including by “machine readable means”. 

We eagerly await the UKIPO’s proposals.