Artificial intelligence and intellectual property: copyright and patents—a response by the CREATe Centre to the UK Intellectual Property Office’s open consultation

Abstract

Martin Kretschmer is Professor of Intellectual Property Law; Bartolomeo Meletti is Creative Director; Luis H Porangaba is Lecturer in Intellectual Property Law—all at CREATe, School of Law, University of Glasgow. The authors are named in alphabetical order, contributed equally and should all be cited. This article The UK Intellectual Property Office consulted between October 2021 and January 2022 on policy options for intellectual property (IP) law interventions that could ‘secure the UK’s position amongst the global AI superpowers’, in line with the government’s national AI Strategy (September 2021) and the vision ‘to make the UK a global hub for innovation by 2035’ (UK Innovation Strategy, July 2021). This article reproduces the submission by the Copyright and Creative Economy (CREATe) Centre at the University of Glasgow. We show that policymakers are in a difficult position to assess reform proposals relating to artificial intelligence (AI) and IP because evidence remains scarce. With respect to computer-generated works and patent inventorship, we urge caution. There is no evidence that new rights are needed. The onus of proof needs to lie with the proponents of much discussed proposals, such as offering AI copyright authorship in the guise of computer-generated works or granting AI inventorship under patent law. With respect to text and data mining (TDM), we see a straightforward opportunity to stimulate UK innovation and improve the transparency of AI systems by opening up the current ‘Hargreaves’ exception to all users (Copyright, Designs and Patents Act 1988: s 29A, Copies for text and data analysis for non-commercial research). More generally, the UK’s research and innovation environment would in our view benefit from a technologically neutral, open-ended user exception (akin to the copyright doctrine of ‘fair use’ in the USA). ‘Artificial intelligence (AI) is a transformative technology, which is already revolutionising many areas of our lives. Unleashing the power of AI is a top priority in the plan to be the most pro-tech government ever’. Thus opens modestly the consultation on Artificial Intelligence and Intellectual Property: Copyright and Patents, conducted by the UK Intellectual Property Office (IPO) between 29 October 2021 and 7 January 2022.1 The Consultation sought ‘evidence and views’ on three specific areas: – Copyright protection for computer-generated works without a human author. These are currently protected in the UK for 50 years. But should they be protected at all and if so, how should they be protected? – Licensing or exceptions to copyright for text and data mining (TDM), which is often significant in AI use and development. – Patent protection for AI-devised inventions. Should we protect them and if so, how should they be protected? We consider each of these legal issues in turn, reproducing the given policy options at the beginning of each section, even where we would have preferred to frame the discussion differently. We then proceed to assess the existing evidence. The text is an authentic reproduction of CREATe’s submission to the Consultation. In the structured format of the Consultation, there was no space to evaluate fully the evidence for our preferred policy option of a technologically neutral, open-ended user exception (‘fair use’).2 With no counterpart in most jurisdictions, s 9(3) of the Copyright, Designs and Patents Act 1988 (CDPA) is rather unique, if not problematic. Indeed, the effective operation of this provision may depend upon other aspects of copyright law which, following Brexit, remain unsettled. By providing 50-year protection to ‘authorless’ computer-generated literary, dramatic, musical or artistic (LDMA) works, s 9(3) poses the complex legal question of what originality standard should be applied. There is an apparent inconsistency with the EU standard of ‘an author’s own intellectual creation’, which relies on creative choices made by an individual,3 for example. The standard of ‘originality’ applicable to computer-generated outputs that do not reflect human creative input is a matter for UK law alone.4 In more than 30 years, s 9(3) was only ever considered in a single court decision,5 which did not address the originality issue. Determining the author of computer-generated works—that is, the ‘person by whom the arrangements necessary for the creation of the work are undertaken’—is no straightforward matter either. In Nova Productions, the Court of Appeal found such a person to be the author of the computer program rather than the user. However, this decision concerned a simple two-dimensional video game, offering limited guidance on the issue of AI-assisted outputs. Furthermore, as the experience with other types of subject matter (eg sound recordings) suggests, the notion of ‘arrangements necessary’ is not resolved, nor is it clear if the ‘person’ making such arrangements can be a legal entity (ie a firm).6 The introduction of a related right of reduced scope and duration referred to as option 2 may lead to an issue of cumulation, with the same subject matter attracting rights of different kind, as the recent experience with databases suggests. The potential costs of additional IP rights typically are of two kinds: higher prices and loss of innovation. In the UK, the Hargreaves (2011) and Gower (2006) Reviews recommended making the policy process more transparent and rigorous.7 IP rights, once created, have proved almost impossible to remove.8 In a period of rapid technological and industrial change, the standards of evidence required therefore must be particularly high. A fundamental point relates to the onus of proof. Advocates of new rights need to evidence what the potential costs are, who will carry them and that the costs are necessary and proportionate to the claimed benefits. The UK government should carefully consider whether the complex legal questions attendant on AI outputs must really be addressed at such an early stage, as an attempt to anticipate issues that have not emerged yet. There is no conclusive evidence showing that the current copyright framework provides suboptimal incentives for the creation of AI-generated works, let alone the existence or sudden emergence of market failure requiring legislative intervention. The UK IPO’s Impact Assessment is framed by a utilitarian discourse which, if unaccompanied by market-based evidence, may seem all too speculative. At present, the role of most AI tools is largely limited to the execution stage of creative production, with human authors retaining control over the conception and redaction phases.9 From a creative and legal perspective, AI applications such as Grammarly are not very different from editing or motion graphics software such as Adobe Photoshop or After Effects. In all such cases, the computer (or AI) carries out the work under the instruction and control of a (human) creator. UK copyright already affords protection to outputs generated by or through such applications so long as they fall within one of the categories of protected works and meet the originality standard.10 There is no real need for a dedicated, sui generis provision dealing with copyright subsistence in computer-generated works. Unless strong evidence emerges that AI users, developers and businesses indeed do rely on s 9(3), we recommend that the UK government removes protection for computer-generated works (Option 1). Extracting information from copyright-protected materials should not be considered a copyright-relevant act. We, therefore, recommend that the UK should avail herself of recently acquired post-Brexit freedoms to foster innovation by adopting a TDM exception for any use (Option 4). In addition, the introduction of a technologically neutral, open-ended exception (akin to the fair use doctrine in the USA) should be explored. With regard to TDM, more evidence is available than for the issue of computer-generated works. Indeed, empirical research indicates that in jurisdictions with more permissive copyright frameworks11 and robust research exceptions, more data mining-related research is conducted.12 Higher firm revenues in information industries, computer system design and software publishing as well as increased, higher-quality scholarly output appear to be found in countries with more open user-friendly provisions such as the US fair use clause.13 The scope of the UK exception for text and data mining (s 29A CDPA) is rather narrow and uncertain, creating confusion, for example in the context of the widespread practice of data scraping.14 While Option 4 seems the most conducive to innovation in research and business, we would favour two other options which the UK government may not have considered: (i) excluding from the scope of exclusive rights TDM and other acts of extracting informational value from protected works; or (ii) introducing a technologically neutral, open-ended exception akin to the fair use doctrine in the USA. We understand that option (i) could be effected by judicial interpretation, especially in a post-Brexit context allowing UK courts to depart from the jurisprudence of the Court of Justice of the European Union (CJEU). Lord Hoffmann’s speech in Designer’s Guild, for example, makes it clear that copyright protection should not extend to the ideas underlying LDMA works.15 If ‘a literary work which describes a system or invention does not entitle the author to claim protection for his system or invention as such,’ the same equally applies to text and data mining which are more concerned with accessing the information disclosed in—rather than taking the expression of—protected works.16 Introducing an open-ended exception (ii), however, is a matter of legislation. In 2011, Professor Hargreaves was specifically asked to investigate the benefits of fair use and how it could be implemented in the UK.17 At the time, the Hargreaves Review concluded that the introduction of fair use into UK law would likely be inconsistent with the EU copyright framework. Instead, the Review recommended the adoption of several closed exceptions stemming from the InfoSoc Directive, including text and data mining. Following Brexit, now may be the time for the UK to rethink fair use. Innovation is determined by a wide variety of economic, cultural, political and social factors, and in the field of copyright, fair use has been a successful legal mechanism in promoting it. In the USA, fair use has allowed the emergence of indexing and search technology, the Google Books project, and, more recently, the copying of code from the Java API into the Android operating system.18 The recent jurisprudence of the US Supreme Court may suggest that most AI-related uses of copyright works are likely to fall within fair use. Would these types of innovation and other potential applications of AI be equally accommodated by rule-based, purpose-limited exceptions such as copying for text and data analysis? We do not think so. The interests of rightholders are of course legitimate. However, the proposed Option 1 of developing a licensing environment that would provide lawful access to the underlying data within countless copyright works seems unrealistic. Requiring rights clearance for TDM and other AI uses of protected materials would increase transaction costs significantly, raising entry barriers for small and medium-sized enterprises (SMEs), in particular market entrants. Big tech corporations would likely retain access to enormous, high-quality, exponentially growing amounts of data, while others ‘may find it economically attractive to train their algorithms on “cheaper”, which often means older, less accurate or biased, data’.19 Based on the evidence currently available (or the lack thereof), we argue that no reform is necessary in this area (Option 0). Significantly, there is no compelling economic evidence or policy for AI to be formally recognized as ‘inventor’. Unlike (real) human inventors, AI does not have a moral claim to inventorship, neither do we anticipate any disputes relating to entitlement to grant to arise from a purported ‘AI inventor’. We share the view (and frustration) of other legal scholars20 and practitioners21 that the AI inventorship debate is seriously overblown and, indeed, seems to be detracting from other, more significant issues in the field. The existing patent framework is fully capable to accommodate technological developments in AI, just as it has been done with biotechnology.22 Furthermore, any reforms which the government may understand to be required should be implemented at the international level, which may not seem achievable or even realistic at this juncture. Formal recognition of this putative inventorship would have to be mirrored across most patent systems; otherwise, applications claiming UK priority may be found incompatible, raising significant barriers to and associated costs with international prosecution. The European Patent Office (EPO), for example, has recently confirmed on appeal the rejection of the DABUS applications EP 18 275 163 and EP 18 275 174. While the decision has yet to be made publicly available, the EPO made it clear that ‘only a human inventor could be an inventor’ and ‘a machine could not transfer any rights to the applicant.’23 Unless harmonization is sought—and hopefully achieved—via international law, interventions at the national level will only risk inconsistency. Rather, the current UK position, following the Court of Appeal’s judgement in Thaler v Comptroller General of Patents,24 is one of relative legal certainty—we do know that AI cannot be named as inventor and, empirically, nothing suggests there is a pressing need for this to be changed. Hence, the reform proposals 1 and 2 under consideration would run counter to the tradition of UK patent law which has been largely developed by judicial practice striving for consistency with EPO decisions.25 In a rapid developing field such as AI, ex post regulation through minor doctrinal adjustments within the discretion of courts and patent offices should be the norm. In the past, more significant policy issues such as the patentability of second medical use inventions have been addressed this way under the European Patent Convention.26 Legislative intervention of the kind being proposed is unwarranted, running the risk of increasing transaction costs associated with patent protection without any tangible benefit. Particularly, there is no conclusive evidence that AI systems can effectively invent autonomously.27 Indeed, the previous call for views on AI concluded that ‘there appeared to be near complete agreement that AI systems are not, or not yet, independent agents seeking patent rights without human intervention’.28 In response to that consultation, IBM stated that ‘AI with the ability to invent without the assistance of a human is a considerable way off … We believe that AI will remain tools that assist humans, rather than invent independently and autonomously, for a considerable time’.29 It is therefore not surprising that some have questioned the ability and legitimacy of the so-called DABUS system, which is not sufficiently explained in any of the patent applications referencing it. Put this way, one might speculate whether the Thaler litigation amounts to anything other than a publicity stunt.30 The introduction of a new, sui generis right to protect AI-devised inventions referred to as Option 2 would also be ill-advised. This would significantly increase costs associated with determining the content of this law, including matters of prosecution and enforcement, which may have a differential impact on small and medium-sized enterprises (SMEs) in the field of technology. There is no guarantee that this new form of protection would develop in the same way as or even build on the existing patent jurisprudence, for example. Recent experience with database rights and supplementary protection certificates both illustrate the difficulty in determining, let alone predicting, how the relevant statutory provisions will be interpreted and applied. By and large, patent applications for AI-related inventions—particularly those featuring deep learning and neural networks—are expected to increase over the next years.31 Even if AI reaches the stage of developing inventions with minimal or no human intervention and those outputs prove to be unpatentable on such grounds, there is no economic evidence indicating this would be detrimental to the investment in and the development of AI technology. As a practitioner has suggested, ‘the main commercial players in the AI field, such as Google DeepMind, continue to navigate the patent system without apparent concern about the issue of AI inventorship’.32

Topics & Concepts