AI and Copyright: What Next in the UK?

Chaired by Professor Sir Christopher Floyd, honorary Professor at UCL Faculty of Laws and former Lord Justice of Appeal, the event examined the complex interplay between copyright and Artificial Intelligence (AI), and how this could be developed in the future to encourage innovation in AI, whilst preserving the role of IP in promoting human creativity. The context for the discussion was the UKIPO’s recent call for views on AI and IP, conducted in the Autumn of 2020 with the UKIPO’s response published in March 2021 (read our summary and analysis of the UKIPO response here).

Sir Christopher began proceedings by outlining two key copyright issues addressed by UKIPO’s call for views; the application of copyright to the outputs of AI systems and issues arising from the use of copyright works as inputs to train AI systems.

Regarding AI outputs, the potential for AI systems to generate works which would be subject to copyright protection if created by a human gives rise to a definitional problem, namely: where is the diving line between a work authored by a human using an AI as a tool (often referred to as ‘AI assisted works’) and a work which has been created by an AI without a human author (referred to as ‘AI generated works’)? Wherever this line is drawn, if we accept that some works could (either now or in future) be ‘AI generated’ we then need to decide whether those works should be protected by an IP right, and if so, whether this should be traditional copyright or a new related right. Sir Christopher pointed out that the application of the UK’s current answer to this question, affording copyright protection to computer-generated works (which dates from the last major reform of UK copyright law in 1988), had been left uncertain following subsequent developments in the law relating to requirement for originality. The UKIPO’s call for views was therefore opportune.

Sir Christopher also highlighted the timely nature of the UKIPO’s call for views on the use of copyright works as inputs to train AI systems. A significant divergence of opinions had emerged from the responses to the call for views. On the one hand, many rights holders considered use of their works to train an AI system to be an activity which engaged their copyright and should only be undertaken with their authorisation. On the other, the act of training an AI system could be considered analogous to the process of exposing a human mind to a work as part of a human learning process such as learning to play an instrument. The resulting reproduction of the work in human memory is not an act which falls within the scope of copyright, leading some working in AI development to question whether the copies created when training an AI system should also fall outside the scope of copyright.

Joining Sir Christopher to discuss these issues were Alexandra Condon (head of Policy and Public Affairs at PRS for Music), Tobias McKenney (Copyright Counsel at Google) and Professor Dr Martin Senftleben (Faculty of Law at the University of Amsterdam). The panel was also joined by Margaret Haig, head of Copyright Operations at the UKIPO, who provided insight into the UKIPO’s progress and intended next steps.

The discussion between the panel provided a fascinating insight into the views of rights holders, AI developers and others on these issues. The session was recorded and can be viewed online here. Rather than providing a blow by blow account of the views expressed, this article focuses on the UKIPO’s next steps and our analysis of the potential direction of travel for UK copyright policy.

The UKIPO’s position on copyright and AI

The UK Government’s aim is to create an environment which places the UK at the forefront of the AI and data revolution. With a view to translating that ambition into copyright policy, the copyright sections of the UKIPO’s call for views focused on the following issues:

(a) Whether existing liability rules can be relied on to manage infringement of copyright works by AI systems;

(b) Whether AI-generated creations should be protected by copyright at all;

(c) Whether copyright (and/or patents) adequately protect AI software; and

(d) Whether copyright is a barrier to machine learning or not.

The call for views received 92 responses. In relation to liability arising from the use of AI systems to infringe copyright, there was general agreement that existing rules were sufficient. On the remaining issues, the responses expressed a range of positions with two poles. On the one hand, there was enthusiasm for the possibility of intelligent computers supporting humans in research, creativity and inventions. On the other, there was concern that the use of AI technology risks removing the human from the creative process.

The UKIPO therefore wants its next steps to encourage innovation in AI technology while preserving the role of IP in promoting human creativity and innovation. It also wants to ensure that any future developments are based on evidence and, to that end, wants to further engage with experts, AI developers and rights holders. As a result, the UKIPO intends to consult on the following issues during Autumn 2021:

(a) Reviewing the ways in which copyright owners license their works for use with AI, and consult on measures to make this easier, including improved licensing or copyright exceptions, to support innovation and research.

(b) Consult on whether to limit copyright in original works to human creations (including AI-assisted creations).

(c) Consult on whether or not to replace the existing copyright protection for computer-generated works with a related right, with scope and duration reflecting investment in such works.

The UKIPO will also consider whether or not there is a need to take action to reduce confusion between AI works and human works, and the risk of false attribution.

Expected next steps

The three areas identified by the UKIPO for consultation had already been highlighted in its response to the call for views, and do not come as a surprise. However, we now have a timeframe, with the consultation expected in Autumn 2021. If the UKIPO follows a similar timeframe for the consultations as it did for its call for views, we could expect the consultations to run from September to November or October to December, with the UKIPO’s response and potentially draft legislation published for further consultation in Q1 or Q2 of 2022. While these estimated timings are somewhat speculative, they suggest that any changes to UK copyright law are unlikely to take effect until the second half of 2022 or first half of 2023.

In terms of the likely direction of travel, some areas are easier to predict than others. Regarding whether AI generated works should be protected by copyright, there has already been strong opposition from rights holders which has not been matched by pressure from AI developers or elsewhere to ensure that such protection exists. Given the UKIPO’s current emphasis on the role of IP in promoting human creativity and the complexity of re-defining the originality requirement for AI generated works without upsetting the existing law, barring a substantial change of direction in response to further consultation, it appears unlikely that UK copyright law will be substantively amended to ensure AI generated works are protected by copyright. There also appears to be an emerging acceptance that the scope and policy rationale behind the UK’s current provisions relating to computer-generated works (in particular section 9(3) of the Copyright, Designs and Patents Act 1988 (“CDPA”)) are unclear. Unless the UKIPO can be persuaded that section 9(3) plays a useful role in encouraging and protecting a certain category of work, its revocation appears to be a reasonably likely outcome of the consultation.

Will we see a new sui generis right for AI Generated works?

Much harder to predict is whether a new sui generis right will be introduced to protect AI generated works. The urgency of reaching an answer is currently unclear as there are divergent views on whether AI generated works (i.e. works produced by AI without human intervention) are a realistic technical possibility either now, or in the short to medium-term future. Some take the view that all works created using AI for the foreseeable future will be AI assisted works (where a human author is merely using an AI as a tool in the creation of a work) and there is no need for an immediate answer to the copyright issue. Others believe that AI generated works are already being created and we need to decide now whether they should be protectable.

As a result, it may be a challenge for the UKIPO to obtain clear evidence in support of the need for a new right, or of any potential harm or unintended consequences which a new right could cause. The responses to the call for views from rights holders have expressed strong disquiet about extending copyright protection to AI generated works, believing it to be a threat to human creativity. However, its less clear whether the same rights holders will also object to a sui generis right, with a narrower scope of protection. From the AI developers side, there does not yet appear to be strong pressure to adopt a sui generis right, although this may be based on a belief that all works created by their AI systems are currently protectable by copyright on the basis that they are AI assisted and not AI generated works. We will have to wait and see what comes out of the consultation once the UKIPO puts forward a more concrete set of proposals.

Where are we headed for text and data mining in the UK?

The outcome of the consultation relating to licensing and copyright exceptions for the use of copyright works in AI training is also hard to predict. There are clearly strong views on the part of rights holders, who see copyright right works used to train commercial AI systems as a raw material which should be paid for by AI developers in the same way as any other. Equally, there are concerns that without a commercial text and data mining exception, the UK will be seen as a less attractive jurisdiction for AI research and development than other jurisdictions which have provided commercial text and data mining exceptions. This includes the EU, which has recently enacted an “opt out” commercial text and data mining exception under the Digital Copyright Directive, and jurisdictions such as the US and Israel, which permit some forms of commercial text and data mining under fair use doctrines.

However, there are hints that the UKIPO may want to think beyond a simple exception vs no-exception dichotomy. In particular, its intention to review the manner in which copyright owners licence their works for use with AI suggests that it intends to assess whether there are currently any failures in the market for licensing copyright works for text and data mining purposes. Examples of situations where AI developers have been refused a licence to use a particular set of copyright works for text and data mining, or the work required to obtain licences from multiple parties whose works are present in a dataset, making a project commercially impossible to deliver, will therefore be of particular interest to the UKIPO. Depending on the evidence presented on the current licensing market, measures to facilitate or even mandate, licensing for commercial licensing in particular scenarios could be considered.

While we will have to wait and see what policy options are put forward by the UKIPO, one possible approach could be adapting the approach taken under section 66 CDPA which provides an exception for lending to the public of certain works, which is subject to “such reasonable royalty or other payment as may be agreed or determined in default of agreement by the Copyright Tribunal” and does not apply where a licensing scheme already exists for those works.

It is also interesting to note that when looking at issues relating to licensing and copyright exceptions for use of works as inputs to AI systems, the UKIPO is considering several issues relating to data access which overlap with work currently being undertaken by the European Commission. For example, in a recently published Inception Impact Assessment for a proposed Data Act (read our summary here) the Commission suggested that one policy option under consideration is a business to business fairness test to avoid unilaterally imposed unfair conditions for data access and use and/or laying down data access and use rights, potentially on the basis of fair, reasonable, proportionate, transparent and non-discriminatory terms for non-personal data, which could be accompanied by model contractual terms. While the Commission’s approach isn’t focused on the use of copyright works for training AI systems, their thinking around the licence terms which are imposed when granting access to data could provide the UKIPO with food for thought.