AI and IP: Back to Basics and Business Impact
17 August 2023
17 August 2023
What ownership can be held over the outputs of AI processes? What might this mean for your business?
Artificial intelligence ("AI") can be used in a broad range of applications, each of which may raise specific intellectual property ("IP") issues.
When utilising AI-based tools, or collaborating with partners or purchasing AI services from third parties, it is important, both from a risk perspective and to ensure value creation, to understand whether IP will subsist in materials used at each stage of the process, who will own them, and whether any restrictions will be placed on the exploitation of the results of AI.
This article summarises the IP considerations common to the implementation of AI-based solutions in the context of the ever-increasing capabilities of AI and the shifting legal and regulatory landscape.
IP can subsist in numerous aspects of AI processes, including the following:
Several IP rights may cover each of the above categories of material, and the approach to the subsistence and ownership of IP relating to those categories differs significantly between countries.
Popular AI tools including both machine learning AI systems as well as generative AI powered on large language models ("LLMs"), such as OpenAI's ChatGPT, are trained on vast amounts of data. Aside from data protection and privacy considerations , the right to use a given dataset may be affected by third party IP rights and by contractual restrictions.
As in the EU, databases may be protected as a whole by a separate database right which subsists in a database where there has been a substantial investment in obtaining, verifying and presenting the contents of the database. Database rights protect against the extraction and/or reutilisation of the whole, or a substantial part of, a database and the repeated and systematic extraction of insubstantial parts of the contents of a database. Copyright (which protects against unauthorised reproduction of copyrighted materials) may also subsist in a database. As such, both raw collections of unlabelled and labelled data may be protected.
Outside the EU and UK, the IP protection afforded to databases is typically more limited, and generally reliant on the copyright or confidentiality/trade secrets protection. In many jurisdictions, the aggregation and collection of information, even if the underlying information is available in the public domain, may be protected as a trade secret.
The process of labelling data and parsing a dataset for use in AI processes will typically increase the likelihood of a database being eligible for protection under the law of confidential information/trade secrets, as long as the dataset is not otherwise available in the public domain in that precise form.
While not IP rights per se, access to datasets and underlying data items may have been granted subject to contractual restrictions which affect the use and ownership of IP in databases and works derived from datasets. When engaging with third parties in relation to any AI-based solution, specifically defining the ownership position and the rights to use in relation to any inputs and modified datasets at the outset of the relationship will help avoid disputes further down the line.
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 UK's Copyright, Designs and Patents Act 1988 ("CDPA"), there is an exemption to copyright for TDM for non-commercial research. Therefore, unless covered by the CPDA TDM exemption or permitted under license, such copying of data constitutes copyright infringement.
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 license fees. There has been some back and forth between the UK Intellectual Property Office ("IPO") and the UK government on expanding the CPDA TDM exemption. Propelled by discussions on generative AI and the need for clarity, the UK government has called on the IPO to produce of a code of practice, to enable the development of AI through access to copyright protected works as input to AI models, whilst also protecting the creative industries.
The core of any AI process is typically an algorithm or a collection of algorithms. In many cases, the high-level details of the algorithms are widely known in both industry and academia, leaving little scope for making IP claims to wide-ranging methodologies. However, subtle differences in the implementation of existing models and the tuning of weighted parameters can make dramatic differences to the performance of an AI-based process or service. As such, variations on well-known, established algorithms may themselves give rise to valuable new IP.
Sets of weighted parameters may also be protected by several forms of IP. Database rights may subsist; copyright and confidentiality protection may also provide additional protection.
In a collaboration, it is likely to be relatively easy to identify which party is providing the algorithms and who (as between the collaboration partners) owns the IP; but the ownership of weighted parameters may be less clear unless clarified under the applicable collaboration contract(s).
Trade secrets may protect the algorithms themselves, or the particular way in which an algorithm is implemented or ‘tuned’. The need to maintain confidentiality may influence the way in which AI software is utilised or deployed to customers or clients.
Patent protection is widely available for AI-based inventions and significant volumes of AI patent applications are now made every year. In Europe, the fact that mathematical and business methods and programs for computers are excluded from patentability ‘as such’ has not prevented the European Patent Office ("EPO") from granting large numbers of patents for AI inventions - the vast majority of which can only be implemented in software. Many patent offices, including the EPO, have issued guidance and convened conferences regarding their approach to the examination of AI-related patent applications.
Although large numbers of AI-related patents have already been granted, relatively few have been tested in court. As the capabilities of AI develops, the discussion also turns to whether an AI system can be marked as the inventor in a patent application. It is arguable in many countries that a human inventor is required before an entitlement to patent protection arises.
This year, the Supreme Court heard the arguments of Thaler v Comptroller-General of Patents, Designs and Trademarks, concerning the UK IPO refusing to accept patent applications where an AI system was argued as the inventor, on the grounds that this did not satisfy section 13(2) of the Patents Act 1977 which clearly requires an applicant to identify "the person or persons whom he believes to be the inventor or inventors'. The Comptroller-General argued that accepting AI as an inventor would require fundamental policy considerations from the UK government (as it would do for industry and the IP community at large), as unlike the position under UK copyright law, there are no provisions in the UK patent legislation by which inventions devised by computers are deemed to have been devised by any identifiable human inventor.
Further disputes may arise relating to entitlement to, and/or subsistence or inventorship of, patents for inventions devised by AI. The discussion may focus on whether an AI-based application was being used by a human inventor as a ‘tool’, where the inventive concept was in fact devised by the human, or whether the invention had, in substance, been devised by AI working without human intervention. Nevertheless, despite the risks of revocation or invalidation, patents for AI-related inventions, and portfolios of AI patents in particular, may give organisations significant competitive advantages – both from an offensive and defensive perspective.
The actual implementation of the algorithms in software code may be protected by copyright. Copyright may protect both the executable form of the program and the source code. The scope of copyright protection in software has been narrowed over time, and in most jurisdictions it is now difficult to protect algorithms and methods using copyright; copyright will still protect the specific implementation of an algorithm in code form. However, the implementation of the same algorithm, in new independently-created code, may not constitute an infringement of copyright subsisting in the original code.
Under the CDPA , computer generated work is defined as work that is "generated by a computer in circumstances that there is no human author of the work". The author for copyright ownership purposes of computer generated work is "the person by whom the arrangements necessary for the creation of the work were undertaken" as under section 9(3) CDPA. There remains uncertainty on whether AI would fall within the definition of computer generated work, and even if so, circumstances would dictate whether the originality requirements are met.
There is still scope for dispute as to how such an individual is to be identified. For example, a purchaser of an AI-based software tool who uses the tool for their own purposes may be the most obvious ‘author’; however the provider of AI-based software as a service may have competing claims to works created by users of its service, and may be able to argue that the provider, rather than the user, undertook the ‘necessary arrangements’.
Whilst the general consensus seems to be that, despite the unsettled position in the UK, AI-generated works likely fall within the definition of computer generated works , questions arise on whether the ease of inserting prompts into LLMs amounts to "necessary arrangements". The answer likely varies on a case to case basis and the level of contribution and creativity involved in relation to the input to an AI tool.
Regardless of where discussions may suggest the answers lie, care must be given to the contractual relationships governing specific AI use cases.
The UK AI White Paper published in March 2023 ("White Paper") outlines the UK government's proposed "proportionate and pro-innovation" framework for regulating AI in the UK. To ensure the framework is effective, the UK government called upon its existing regulators, to develop tailored, context-specific approaches to AI (including developing pragmatic codes of practice), and confirmed its commitment to take forward a key recommendation by Sir Patrick Vallance to establish a regulatory sandbox for AI. This would be welcomed by most businesses so that they can continue to enjoy a relatively less regulated environment, as well as providing opportunities for collaboration and enabling experimentation. However, in its response to Patrick Vallance's report, the UK government also notes that whilst it believes in a pragmatic code of practice led by regulators (which aims to provide confidence to innovators and investors), this may be followed up with legislation. This is in stark contrast to the UK government's response to the UK IPO's first consultation on AI and IP in 2022, where it was concluded that no changes in copyright law were planned for the protection of computer generated works, as well as the human requirements for inventorship under UK patent law, given that the use of AI was in its early stages. This suggests that technological advances in AI may now outstretch the UK's current legal and regulatory frameworks governing where IP may subsist, fuelled by the supernovae of developments in generative AI.
Even in other jurisdictions where there appears to be clear requirements for a ‘human authorship’, the developments in the capabilities of AI raises the interesting question of where the legal boundaries may lie. The discussion may evolve to focus more heavily on where such human intervention lies within the chain of creation and the use of AI as supportive tools in the creation process. Similarly, whilst the laws relating to entitlement to patents and ‘inventorship’ vary significantly between jurisdiction, in most cases, the question of ownership of inventions in which AI played a role is likely to focus on the identification of an identifiable human inventor and the contribution made by that individual. If copyright or patentable inventions do reside in the outputs of AI-based processes, in the context of a collaboration, there may be disputes as to which of the collaboration partners contributed the necessary elements giving rise to the IP rights. As such, ownership of IP in outputs should also be addressed in any collaboration contracts.
Relatively few businesses currently create their own proprietary AI systems. AI systems are often built using a combination of third party elements; such as an open source algorithm and infrastructure (provided by IBM or AWS for example). Investors should obtain or conduct a mapping exercise of the various IP elements to the AI system (including in relation to its hosting and maintenance).
If the core elements to the AI system are provided by a third party, investors will want certainty that the third party AI provider will continue to provide these elements in the future and, if this may not be the case, consider the consequences of the third party failing to do so. Investors should also review any underlying third party terms to ensure the target complies with the same.
When investing in, or acquiring, a provider of AI-based services or a specific AI-based software product, the due diligence will typically focus on:
Contributor: Saba Nasrolahi, Trainee Solicitor
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.