AI and IP: Copyright – the wider picture and practical considerations for businesses

What is the wider picture? Where might the position settle? What practical steps should businesses consider?

Introduction

Propelled by the supernovae of generative artificial intelligence, developments concerning the capabilities of artificial intelligence (AI) seem to be outpacing the current legal and regulatory frameworks in the UK and the rest of the world. Whilst the EU and beyond move to solidify the gaps surrounding uncertainty in existing copyright law more comprehensively, the UK government has championed a "pro-innovation" soft law, principles-based approach to facilitate the UK as an "AI superpower" in its recent UK AI White Paper launched in March 2023. However, will the exponential growth of AI outpace any regulatory code of practice, resulting in the intervention of legislation to tackle the concept of new or expanded intellectual property (IP) rights?

We explore the current position of copyright law, how it fairs in the light of innovation in AI and, given the current uncertainty in the UK, what this might mean for your business.

UK Copyright - the current position

A core question concerning copyright and generative AI is whether AI outputs are protected by copyright. The Copyright, Designs and Patents Act 1988 (CDPA) defines computer generated work as work "generated by a computer in circumstances that there is no human author of the work". Currently, the general consensus in the UK is that as we await the position of copyright law and AI to recalibrate, and simultaneously giving consideration to specific contractual arrangements with AI service providers, AI likely falls under the definition of computer generated work.

Under section 9(3) of the CDPA, the author for copyright ownership purposes of computer generated work is "the person by whom the arrangements necessary for the creation of the work are undertaken". It can be useful to visualise these separate requirements 'copyright ownership chain', which are often seen across other jurisdictions despite the UK uniquely holding provisions for works that do not have a human author. There are various pressure points on the copyright ownership chain caused by developments in AI.

Human Authorship

Whilst currently there exists a general consensus that AI likely falls into the definition of computer generated works, the current UK legislative drafting, notes that a "person" would be the author for copyright ownership purposes of computer generated works, and therefore, an AI system itself can hold copyright over AI-generated outputs. In considering whether an AI system can be the inventor under section 7(3) of the Patents Act 1977 in Thaler v Comptroller-General of Patents, Designs and Trademarks, the UK Supreme Court held that the scheme as a whole is on the footing the patents can only be granted to persons specifically. This indicates that the current reading of a "person" relates to a human, and not the capabilities of a human (or beyond) that an AI system could possess in the creation of AI-generated outputs.

In the US, a judgment issued on 18 August 2023, in Thaler v Perlmutter, by Judge Beryl A. Howell (Judge) affirmed the U.S. Copyright Office's (Office) decision that a work generated entirely by AI with no human input is not subject to copyright protection. This reaffirmed the Office's guidance issued in, March 2023, that copyright only protects material that is the product of human creativity. In Thaler v Perlmutter, the Judge found that "authorship" has been "synonymous with human creation over time" and "[h]uman authorship is a bedrock requirement of copyright". The Judge importantly noted that copyright law has "never stretched so far…as to protect works generated by new forms of technology operating absent any guiding human hand".

However, it appears the Office is finally gearing up to see whether the law on copyright is now indeed 'overstretched' by the capabilities of AI, particularly generative AI models that require limited human input in the creation of their outputs. On 30 August 2023, a few months after the Office held a webinar to direct applicants to disclose where a portion of copyright protectable work was created with the assistance of AI, the Office published that it is undertaking a study of the copyright law and policy issues raised by AI systems and whether legislative or regulatory steps in the area are warranted. The first general question posted for public comment by the Office notes that "generative AI systems have the ability to produce material that would be copyright protectable if it were created by a human author". Perhaps the future will see the Office take off its firm hand on its view on human authorship and consider the extent to which AI tools can distance the 'guiding human hand' from the creation of copyright protectable works.

In contrast to the US, the UK uniquely has a provision of computer generated works under the CDPA to allow for AI generated outputs to be copyright protectable, however, the same questions remain across most jurisdictions on (i) how much human authorship is required for a user of an AI system to qualify as the owner of AI-generated output; and (ii) how far away from the creative input can the human sit within the existing legal frameworks to maintain authorship?

Necessary Arrangements and Creation

Before the explosion of generative AI, there was far less consideration and contention of the requirement of "necessary arrangements" under section 9(3) of the CDPA. The use of AI as a tool, such as generative AI tools, to facilitate the creativity of human authors places greater uncertainty on all elements of the copyright ownership chain. For example, would inserting prompts into large language models, a type of AI that can mimic human intelligence, constitute "necessary arrangements" by the user? Or, does the chain of human authorship lead back to the developers of the AI system itself? Until English law deals with these uncertainties, care must be given to the specific contractual arrangements with AI service providers to determine which 'person' makes the necessary arrangements for the creation of the AI output. Some generative AI developments are assigning copyright protectable outputs to their users. However, users may also be liable for copyright infringement where the AI system is trained upon a third-party's pre-existing work and without the either the rightsholder's consent or being able to rely on a fair dealing exception to copyright infringement.

Copyright Infringement

Text and data mining (TDM) concerns the use of computational techniques to analyse large amounts of information to identify patterns, trends and other useful analytics, which can be used to train AI systems. For example, data mining systems copy works to extract and subsequently analyse the data they contain. Under section 29A of the CDPA, there is an exemption to copyright for TDM for non-commercial research. Unless covered under the CPDA's TDM 'non-commercial' exemption; or permitted under license, such copying of data constitutes copyright infringement in the UK.

Generative AI is trained on vast amounts of datasets, and as per the above, such datasets will likely contain work protected by third party copyright. Whilst it can be difficult for rights holders to prove their works are being used by the AI and that their works have been used in the creation of an AI generated input, the risk remains and must be accounted for. As the UK did not transpose the EU's Copyright Directive allowing the commercial use of TDM subject to an opt-out caveat, commercial AI data mining remains subject to copyright infringement or the obligations of licence fees. There has been some back and forth between the UK Intellectual Property Office (IPO) and the UK government on expanding the CPDA's TDM exemption. Propelled by discussions on generative AI and the need for clarity, the UK government has called on the IPO to produce a code of practice, to enable the development of AI via access to copyright protected works so that these may form the inputs for AI models, whilst also protecting the creative industries. The UK government flags in its response to Sir Patrick Vallance's Pro-Innovation Regulation of Technologies Review that an AI firm which commits to such code of practice, it can "expect to be able to have a reasonable licence offered by a rights holder in return". It remains to be seen whether this will come to fruition, and it seems the UK government does ultimately seem to accept that legislation may be the required end point.

However, the EU has already taken a more comprehensive legislative approach, even in the light of the Copyright Right Directive. The current draft of the EU AI Act requires generative AI providers to publicly disclose details of copyright protected works used in the training of generative AI models. This transparency would evidence the use of rights holders works to train generative AI models and takes (some) steps to answer the concerns of copyright infringement.

In response to these growing concerns on copyright infringement, on 7 September 2023, Microsoft announced the Microsoft Copilot Copyright Commitment where Microsoft will assume legal responsibility for their customers should they get sued for copyright infringement whilst using the company’s Copilot AI services, provided the customer uses the guardrails and content filters integrated into the AI products. Whilst the new commitment does not change Microsoft's position that it does not claim any intellectual property rights in the outputs of its Copilot AI services, it provides confidence to users of generative AI services, as opposed to the relevant rightsholders, to continue such use without the overhanging threat of the legal and financial consequences of copyright infringement. This move may signal a shift in the standard user terms and conditions of various AI service providers.

How to capture AI's ever-evolving capabilities?

The Members in the European Parliament's amendments adopted by the European Parliament as the current position on the draft on 14 June 2023. This proposal finally saw the introduction of various definitions under the EU AI Act.

The key definitions consisted of:

• Article 3 (1d) – "General Purpose AI System" meaning an AI system that can be used in and adapted to a wide range of applications for which it was not intentionally and specifically designed
• Article 3(1c) – "Foundation Model" meaning an AI model that is trained on broad data at scale, is designed for generality of output, and can be adapted to a wide range of distinctive tasks
• Article 28(b) - Obligations of the provider of a foundation model – This article sees "Generative AI" described as a subset of providers of foundation models used in AI systems specifically intended to generate with varying levels of autonomy, content such as complex text, images, audio or video". Such systems and providers who specialise a foundation model into a generative AI system will need to, for example:
  • Comply with the transparency obligations outlined in Article 52(1); and
  • Train, and where applicable, design and develop the foundation model in such a way as to ensure adequate safeguards against the generation of content in breach of Union law in line with generally-acknowledged state of the art, and without prejudice to fundamental rights including the freedom of expression.

The development of this, to an extent, tier based system acknowledges that there are overlaps and divergences in the types of AI systems being developed and that in turn, different legal obligations may need to be considered for these evolving capabilities of AI.

For example, under the draft EU AI Act, generative AI is specifically subject to additional transparency obligations as a way to address concerns of copyright infringement (as discussed above). Under Article 28b(4)(c), the ‘provider’ of a foundation model used in or specialised into a generative AI system must ‘document and make publicly available a sufficiently detailed summary of the use of training data protected under copyright law’. This must be done before the model is placed on the market or put into service in the EU (Article 28b(1)), something which will be welcomed by rightsholders, although it is not yet clear what constitutes a "sufficiently detailed summary". Equally the EU AI Act does not impose any express obligations on an AI deployer to protect the rights of third parties rightsholders, rather an implication can be made when reading the recitals in conjunction with the express obligations concerning transparency, governance and technical documentation. This maybe an argument a competent authority and the European Court of Justice may seek to rely upon to provide a mechanism affording protection to IP rightsholders.

The drafting is not comprehensive, for example in defining differing AI , it does not capture the evolution of large language models capable of generating code from text prompts explicitly in its definition of generative AI, but it does provide some guidance on previously unanswered questions. In the UK, whilst Lord Young of Graffham said the inclusion of computer generated works at the inception of the CDPA was “the first copyright legislation anywhere in the world which attempts to deal specifically with the advent of artificial intelligence”, this was almost four decades ago; the questions posed by the exponential growth of AI may now mean that the UK's legal framework also needs to be adjusted in order to capture potential developments in the AI systems and models. This should be in broad strokes, as not to stifle innovation but to provide greater certainty.

Where might the UK's position settle?

In 2022 following consultation, the UK government responded, as at the time the use of AI was in its early stages, there was no evidence that copyright protection for computer generated works is harmful and that the existing laws adequately captured the issue of needing to establish authorship in cases of alleged copyright infringement. By 2023, the UK government called upon existing regulators in the UK AI White Paper, to develop tailored, context-specific approaches to AI (including developing pragmatic codes of practice), providing a less regulated environment to promote the development of AI, unlike the proposals in the EU and the more rigid nature of the US. The IPO invited industry representatives from the technology, creative and research sectors to form a working group with meetings having commenced on 5 June 2023, guided by a target of producing a code of practice by the summer (if one can be agreed or adopted).

However, in its second report on connected technology, the Culture, Media and Sport Committee flagged that opening up proposals such as moving towards an exception to allow commercial TDM text risks reducing the arts and cultural production to mere ‘inputs’ in AI development and "shows a clear lack of understanding of the needs of the UK’s creative industries". This signals that there is a stark difference between the UK government's pro-innovation approach and the view of various industries on the level of IP protection required, and in turn, the creation of new, or expanding the definitions of existing, IP rights.

Perhaps, whilst the UK has promoted its 'flexible' existing frameworks in the past in relation to new technologies, AI signals the need for the creation of a new legal framework governing AI and IP rights, especially in a globalised digital economy. Attempts for a degree of harmonization across borders already seems to be on the horizon, as seen by a draft of the voluntary US-EU AI Code of Conduct expected to be published imminently, therefore it may not be long before such a code becomes the de facto harmonized standard, as was seen with the EU's approach to data protection. The UK will need to act with some rigour on how to move its AI agenda forwards to remain a hub for AI innovation.

Considerations for businesses

The uncertainty posed by the current legal and regulatory framework means businesses looking to create, or use an AI system, should consider steps which it can take to minimise a claim for copyright infringement. Below, we set out some practical considerations:

• Identifying the copyright protected works which are going to be used and determining whether such works are still within copyright (not expired) and if so, trying to seek a licence from the rightsholder(s).
• Consider if any other IP rights or restrictions are applicable to such works.
• Carefully consider how the generative AI system is to be trained; if this is to include use of copyright protected works, then you should consider whether the extracted data and the training models themselves would be considered copies under the CDPA.
• As mentioned above, some jurisdictions such as the EU and US permit commercial TDM and therefore would such software development activities be better placed to occur (including being hosted) in another jurisdiction and therefore mitigate the risk of a claim for copyright infringement arising.
• Seeking which mechanisms can be introduced to ensure human oversight still plays a key factor in the selection of the inputs used so as to reduce the chance of infringing outputs being generated.

For an overview of IP considerations common to the implementation of AI based solutions, re-visit the first in our AI and IP: Back to Basics series here.

We will be holding further events this year on AI, covering IP considerations for your business. If you would like to register your interest, please click here.

Authors: Sunny Kumar, Partner and Saba Nasrolahi, Trainee Solicitor