Navigating intellectual property law in the generative AI revolution

Simon Barker and Chris Musgrave of Freeths discuss Generative AI and how intellectual property law developed out of an old world order is addressing challenges presented by a new technology and how it may have to adapt.

Some might say we are on the brink of a 21^st century industrial revolution. Like a steam hammer from the 19^th century AI may be set to change our lives forever; mechanising human tasks, changing and replacing jobs, the goods and services we consume and how we consume them. AI hardly needs introduction. It is widely commented on in the popular press whether as some autonomous beast, a catalyst for a dystopia, or more realistically (at least for now) a form of machine learning that will be a driver for future economic success. Whatever your view, change is coming, and the law will have to adapt too.

Both the EU, with its proposed EU AI Act, and the UK are looking at regulation, but these are focused on protecting fundamental values around freedom, democracy, and privacy. Whether the final regulations will address intellectual property rights is yet to be seen.

Indeed, the impact on intellectual property is a special case because AI presents new ways of generating works and inventing novel ways of doing things. While the English courts have found that a machine cannot be named as the inventor of a patent[1], AI may be used to help develop inventions. Whether AI systems are patentable themselves can be a difficult question to answer because computer programs are excluded (at least under English law), although there has been some recent guidance from the courts which is positive for AI developers[2]. The issue is what technical contribution the invention makes over and above the computer program itself[3]. Computer code is more readily protected as a literary work under the law of copyright.

It follows that, in the field of intellectual property the more recent controversy is in the development of new generative AI systems that have been trained using other people’s copyright works and data, the works AI systems produce, and the relationship with the law on copyright. The case of Getty Images v Stability Ai Ltd brought in the US and the UK[4] highlights some of the difficult issues that the courts and legislators are going to have to wrestle with.

Firstly, do AI systems infringe third party rights in the works or data that they use for training the AI system? Getty allege that Stability Ai have infringed their copyrights by reproducing the whole or substantial parts of their works in training their product, Stable Diffusion. They also allege infringement of a database right by extracting a substantial part of a database, comprising millions of images and data taken by scraping Getty Images websites. The matter is complicated in the UK proceedings by the need for Getty to prove that the location of the alleged infringements also took place in the UK.

Secondly, there is the question of whether there is nevertheless infringement by importing Stable Diffusion into the UK and whether by possessing or dealing in the product also amounts to an infringement, regardless of whether the “training” took place here. This raises a novel point because liability rests on whether Stable Diffusion can be described as an “article” under the provisions of the Copyright, Designs and Patents Act 1988 (“CDPA”). The counterargument is that an article does not encompass anything intangible like an AI system but rather something physical like a book or a photograph. Whether that is the case is now reserved for trial.

Further, Stable Diffusion has an “image-to-image” feature that enables a user to input an image and then have the system produce a synthetic image in response, which corresponds to the whole of or none of the visual prompt. The allegation appears to be that in producing the synthetic image there is infringement of Getty’s copyright work by authorising acts of reproduction or communication to the public of an infringing copy. It follows, as a matter of principle, that there is a big question as to whether the output of AI systems trained on other peoples’ work can infringe their intellectual property rights.

In a move that would promote the AI industry and the perceived economic advantages it might bring, the UK government sought to loosen the legislative framework by permitting reproduction of text and data for training AI systems.

However, in response to lobbying from the creative sector the government has abandoned those plans so that acts of copying text and data are only permitted for non-commercial research.

However, navigating AI and whether the existing law on copyright is fit to deal with it doesn’t begin and end with the issues raised in the Getty case. We have said the purpose of AI is to reduce the expense of human effort in the creation of things, but to fully realise the value of created works in intellectual property terms a user will want to derive a competitive advantage by owning and controlling the output. Many AI systems will purport to assign rights or at least a licence them to the user.

However, it is not at all clear whether copyright would subsist in the output of a generative AI system or who would be the owner of such rights if it did. Literary, musical and artistic works have to be original for copyright to subsist, which means in European terms that they have to be “of the author’s own intellectual creation”[5], or applying the UK traditional approach “require skill, labour and judgment”. How that applies to a computer-generated work is one thing but, as if things were not complicated enough, post-Brexit there could be scope for argument as to which test applies. In a recent case led by one of our colleagues, Lord Justice Arnold in the Court of Appeal[6] indicated that the European approach was more demanding.

Either way it seems there needs to be some human creative input to have a chance at satisfying the test, and where the courts cannot identify a human author, it may not be able to satisfy the test. 

In relation to computer-generated works section 9(3) of the CDPA applies in the UK. It says that in the case of literary, dramatic, musical, or artistic works the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken. There seems to be two possibilities: the author of the AI system or the user of it. 

Since the author is the first owner of copyright (or their employer if employed) it is important to identify the author, but ownership may be inherently uncertain if that is difficult to do, unless it is addressed in contractual terms with AI system supplier. In a case in 2006[7]  the English court found that a computer games programmer was the author of the works produced by the program he wrote rather that the player, because the player had no real creative input. However, the case may be distinguished from AI systems generally where the user may be much more involved in the production of the generated work.

We shall have to see how the legislators and courts deal with these challenges but like the industrial revolution before it, AI is with us and there is no going back.[8]

---

This article was originally produced for the Corporate Live Wire Intellectual Property Expert Guide 2024 Expert Guides - Corporate LiveWire - Corporate LiveWire – JANUARY 2024.

[1] Thaler v The Comptroller-General of Patents, Designs And Trade Marks [2020] EWHC 2412 (Pat) (21 September 2020)

[2] Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks [2023] EWHC 2948 (Ch) (21 November 2023) (bailii.org)

[3] Aerotel Ltd v Telco Holdings Ltd Macrossans Patent Application [2006] EWCA Civ 1371

[4] Getty Images (US) Inc, Getty Images International UC, Getty Images (UK) Limited, Getty Images Devco UK Limited, Istockphoto LP, Thomas M Barwick Inc v Stability Ai Ltd [2023] EWHC 3090 (Ch)

[5] Infopaq International A/S v Danske Dagblades Forening Case C-5/08

[6] THJ Systems Ltd v Sheridan [2023] EWCA Civ 1354

[7] Nova Production ltd v Mazooma Games Ltd [2007] RPC 25

[8] This article was prepared in December 2023. The draft EU AI Act wording has now been agreed with obligations in relation to copyright law and the Supreme Court has confirmed the cited position in Thaler v The Comptroller-General of Patents, Designs and Trade Marks.