Patenting AI: the nature of inventorship and the mechanics of ownership

On 21 December 2021, the EPO Boards of Appeal released its latest decision on the DABUS cases relating to the identification of AI as the inventor of a patentable creation. Adopting a similar approach to that taken by the UK Court of Appeal in September 2021, the EPO decided that: “AI cannot be named as inventor on patent applications… under the European Patent Convention (EPC) an inventor designated in a patent application must be a human being”.

Background

Since Dr Thaler first applied for a patent naming his AI creativity machine (also known as an ideation machine), DABUS, as the inventor in 2018, the use of AI to generate inventions, and the ownership of those inventions, has been much discussed, but without any clear answers. Under the UK Patents Act 1977, the first owner of a patentable invention is, in lieu of other arrangements, the inventor, while the inventor is the “actual devisor” of the invention (Section 7(2) of the Patents Act 1977).

In naming DABUS as the inventor and himself as the owner (as is the case in respect of both the UK patents and the European patents in the recent EPO decision), Thaler prompted questions concerning not only the nature of an inventor, but also the mechanics of ownership of inventions.

Creativity machines as devisors of inventions

“In a human we would have called it creative, genius. It was not. It was brute processing power and clever algorithms.” (Richard Susskind talking about DeepMind’s AlphaGo at a talk for ILBF on 25 November 2021)

This quotation gets to the heart of the debate about AI as an inventor. Like other ideation machines, DABUS works to generate ideas or possible solutions to problems. While people are likely to be involved in selecting and training a specific AI system, the results are generated by the AI. These may include an inventive answer to a problem (an invention), and, as such, the ideation machine may be the “actual devisor” of the invention. How it arrives at that invention, however, is less a eureka moment and more the result of data processing carried out by the AI.

The process of the AI raises several questions related to inventorship, including where the patentable invention lies in the process (is it in the results or in the creation of the AI system?), and if the result is patentable, who the inventor – and therefore, the owner of the invention – may be. Here we will assume that results generated by AI systems may be patentable inventions – though, no doubt, the debate about the place of the invention in the process will continue to develop as use of these AI systems matures.

Developments at the UKIPO

The UK Intellectual Property Office (UKIPO) recently closed a consultation seeking views on whether the law in the United Kingdom should be changed to better accommodate AI-devised inventions. In the consultation, the UKIPO asked respondents to choose from four possible approaches:

• Option 0 – no change to the current law;
• Option 1 – ‘inventor’ expanded to include humans responsible for an AI system that devises inventions;
• Option 2 – allow patent applications to identify AI as inventor. This may be achieved by either amending legislation to allow AI to be named as the inventor; or to remove the requirement to name an inventor if the invention is devised by AI; and
• Option 3 – protect AI-devised inventions through a new type of protection.

Invention and ownership

Options 0, 1 and 2 all raise potential issues for ownership of patents protecting AI-devised inventions. For the purposes of this article, we will focus on those raised by options 1 and 2.

A human responsible for the AI system

Identifying a human who is responsible for the output of an AI system is emotionally pleasing, because it enables humans to remain centre stage. More pragmatically, it also means that we can borrow from existing legal frameworks to identify inventors and ownership mechanisms. For example: UK copyright protection for computer-generated works or copyright afforded to film producers.

Copyright protection for computer-generated works 

UK copyright law contains a provision allowing the “person by whom arrangements necessary for the creation of [a computer-generated work were] undertaken” to be named as the author of the work. The author in copyright holds a similar position to the inventor in a patent, with the starting point being that the author is the first owner of the copyright work.

However, there are problems with this approach under copyright. A copyright work, for example, must be original and, while originality has a lower standard than inventiveness under patent law, there remains a question as to whether a computer can generate something that is truly original. The problems with this provision were highlighted by the UKIPO in its consultation, when it requested opinions about whether this provision should be removed from UK copyright law. Seeking to replicate in patent law a mechanism that has proved problematic in copyright appears to risk creating new areas of uncertainty, rather than bringing clarity to the law on AI-devised inventions.

Authorship for film producers

The producer of a film is an author of the copyright work (ie, the film), despite not necessarily having day-to-day involvement with the script, direction or acting. English law affords the producer this authorship right, because the producer is “the person by whom the arrangements necessary for the making of the […] film are undertaken” (s.178 Copyright Designs Patent Act 1988).

In both of these cases, the idea that the person who makes possible the generation of the new work, whether computer-generated or a film, should be rewarded with authorship is repeated. The human(s) behind an AI system may be akin to a film producer (or software programmer), in that they are the individual who programmes and/or provides training information for the AI system, thereby putting into place arrangements that enable the AI system to produce the resulting invention.

AI as the inventor – to name or not to name?

At first sight, allowing the AI system to be named as an inventor appears to be the most straightforward solution because it would remove the requirement for there to be a human inventor, but would otherwise retain the current system of patent filing. In contrast, removing the requirement to name the inventor only in the case of AI-devised inventions would create a false sense of ease – the idea that the process of patent filing is the same, but the inventor is not named.

However, things are not so straightforward when considering ownership. As the inventor is the first owner of the patent, does that then means that the AI is the first owner of the patent?

This potential outcome is fraught with uncertainty. An AI is neither a natural nor a legal person (like a company), so cannot own anything. If the AI is to be named as the inventor, then we need additional changes to legislation to allocate ownership of AI-devised inventions.

Wrestling with ownership

If the inventor will be the human responsible for arranging the AI, assuming that the inventor is an employee, the owner of the patent application under UK patent laws will be their employer. But who will that employer be?

There are already companies that specialise in developing AI for ideation purposes. These have developed AI systems, which they make available for use by other companies (eg, in the field of drug discovery). If the AI delivers an innovative, potentially patentable, solution that incorporates information from a healthcare company and uses the AI framework developed by the software company, who should own that solution (and any patent that results from it)? Should it be the software company as the employer of the programmers, and/or the healthcare company whose staff provided the information and used the AI solution?

The law, as it stands, does not provide a straightforward answer to this problem and highlights the possibility that such an invention may, by default, be co-owned.

Contracting for clarity

The uncertainty in relation to the potential ownership of AI-devised inventions may be regulated by contract between the parties (as is common in research collaboration agreements where there is often a split between ownership of improvements, depending on the role played by each party, and the use of either party’s pre-existing (background) intellectual property). Without careful attention and focus on the intellectual provisions in contracts governing the use of an AI system, it is possible that the developer of the AI, whether intentionally or otherwise, will either own or co-own any resulting AI-devised inventions.

Difficulties can arise in the protection and commercialisation of co-owned inventions, and there is no reason to think that co-ownership would be any easier to manage in relation to AI-devised inventions. As a result, we would advise both developers of AI systems and their users to carefully consider the potential outcomes from each project to identify intellectual property that may be generated, and allocate ownership rights from the outset.