Group chatting on bridge with sheep.

Generative AI and intellectual property rights—the UK government's position

03 May 2023. Published by Helen Armstrong, Partner and Jani Ihalainen, Associate and Joshy Thomas, Knowledge Lawyer

The IPO is to produce a code of practice by the summer that will provide guidance to support AI firms in accessing copyright protected works as an input to their models.

Generative AI is currently an area of irresistible focus for the tech sector and creative industries, for investors, policymakers, the media and for society at large. Put simply, it creates text, images, music, speech code or video based on the prompts given to the system by its users. It can do this because its large language models (LLMs) such as ChatGPT are trained on vast amounts of data and use deep learning algorithms to process and understand natural language. The training data the LLMs are trained on is likely to be protected, at least in part, by intellectual property rights such as copyright and database rights. It is also possible that training data containing protected works will not only be used for machine learning but may also be part of the output.

The techniques used and outputs of generative AI systems are bringing into sharp focus important issues regarding infringement of works protected by intellectual property rights, content licensing and authorship and ownership of AI works.

Text and data mining—the current position

Text and data mining (TDM) refers to the automated computational analysis of information in digital form, such as text, sounds and images (often copyright protected works) to obtain data about those works. Machine learning is then applied to the data to enable dynamic learning from this input data. TDM is a hugely important element in advancing AI applications and the technique can often result in unauthorised use of copyright protected works leading to infringement. Consequently, there has been a TDM copyright exception in the UK since 2014, but only where access to the copyright protected work is lawful and where the computational analysis is for the purposes of research for a non-commercial purpose.

In June 2022, the UK Intellectual Property office (IPO) signalled its intention to introduce a new copyright and database exception that would allow TDM for any purpose including commercial use. The proposal did not permit rights holders to contract or opt out of the exception or to charge for UK licences for TDM. There was subsequent outcry by rightsholders, particularly from the publishing and creative industries sector, on the breadth of the exception and the proposal has been withdrawn pending an impact assessment on the implications for the creative industries.

The current position in the UK is therefore that AI systems trained on data protected by intellectual property rights such as copyright and database rights must do so under a valid licence or face the possibility of infringement proceedings.

Text and data mining—UK reforms

Alongside the March 2023 Spring budget, the government published Sir Patrick Vallance's Pro-Innovation Regulation of Technologies Review (PIRT) setting out his recommendations, as well as publishing the government's response to support innovation in generative AI. The PIRT acknowledged that government engagement with stakeholders had shown that the relationship between intellectual property law and generative AI is unclear and there was a lack of regulatory clarity as to the direction of UK reforms, particularly for AI firms deploying TDM techniques to generate new content.

The PIRT suggested TDM should be permitted for any purpose to include the use of publicly available content protected by intellectual property as an input to TDM (including databases). Intellectual property rights would apply to the output of any product. It also recommended a code of practice, a requirement for altered images to be labelled as generated or assisted by AI and the use of technological solutions for ensuring attribution and recognition, such as watermarking.

In its response and in support of the PIRT recommendations, the government proposed that the IPO will produce a code of practice by the summer that will provide guidance to support AI firms in accessing copyright protected works as an input to their models, while ensuring there are protections (e.g. labelling) on generated output. An AI business that commits to the code of practice will be offered a reasonable licence by the rights holder. If the code of practice is not adopted or if agreement is not reached, targeted legislation may follow. It will be interesting to see how this proposal impacts on the High Court copyright infringement proceedings commenced by Getty Images against Stability AI in January of this year (and other similar proceedings). In the proceedings Getty Images alleges that Stability AI unlawfully copied and processed millions of images protected by copyright and the associated metadata.

The attribution or labelling of works can be challenging as AI systems can produce new works from mere fragments of other works, such as words or parts of images, so the labelling of those words or sections will be a challenge. Any licence fees will also have to account for the degree of use, as a blanket fee for any labelled content could result in AI system developers paying costly licence fees for generated works.

The IPO is also to take forward the PIRT's enforcement recommendation to provide guidance (again, by the summer) to AI firms encouraging the development of AI tools which assist with enforcement. No indication has currently been given on any enforcement mechanisms and/or any penalties for the misuse of protected works in AI generated works.

The government's pro-innovation AI White Paper

At the end of March 2023, the UK government released its pro-innovation AI White Paper setting out the government's future intentions for the regulation of AI systems. For more detail on the AI White Paper and the UK's proposed approach to regulating AI, see our earlier post.

Intellectual property is mentioned only once in the White Paper, with the government repeating its intention to take forward Sir Patrick Vallance’s PIRT recommendations relating to intellectual property law and generative AI.  The main message of the AI White Paper, as set out firmly in the title, is of the government's desire to remove barriers to innovation in order to drive growth and prosperity and to strengthen the UK's position as a global leader in AI.

Since TDM plays such a significant role in advancing AI applications it will be interesting to see how the government marries this pro-innovation approach with the PIRT/government's licensing proposals for AI firms accessing copyright protected works as an input to their models, as mentioned above.  The UK government's commercial text and data mining exception was axed as a result of the robust response it received to its incredibly broad proposal and it has since promised to "engage seriously, cross-party and with the industry, through the IPO, to ensure that we can, when needed, frame proposals that will command the support required". The tension between those providing protected content and those using it is likely to remain until there is more flesh to the bones of the new TDM exception and licensing options.

Generative AI holds tremendous promise for a wide range of applications, but it also raises important questions about the applicability of intellectual property rights and ownership of generated works. As this technology continues to evolve, it will be important to develop new ownership models and explore where the infringement risks are in relation to this emerging technology.