Clifford Chance

"Artificial intelligence" is a broad term used to describe a range of computer and software functionality. There is no single definition of AI. It is often misused to describe automated processes, but most commonly it is used to describe different levels of machine learning.

Can AI be an inventor?

If an AI system generated an invention, is it protectable under patent law? Could and should that AI system be considered the inventor for the purposes of a patent application? These are not just philosophical questions. In fact, it was reported this week that a team of academics have informed the patent offices in the UK, Europe, and the US that a patent application they filed was for an invention actually devised by an AI system called Dabus, developed by AI pioneer Stephen Thaler.

Patents are a "bargain" with the state. In exchange for disclosing an invention, the state grants a time-limited monopoly over that invention.  Only an inventor, or a person deriving its right from the inventor, is entitled to the grant of a patent. Under the Patents Act 1977, the "inventor" is defined as the "actual deviser" of the invention. There is no explicit provision in the Patents Act 1977 or European Patent Convention that states that an inventor must be a human being, but under current English law only a natural person has been considered capable of being an inventor.

This was the same conclusion reached by a recent study published by the European Patent Office in conjunction with researchers at Queen Mary University. In a multi-jurisdictional analysis, which covered countries including the United Kingdom, a number of continental European countries and the United States of America, it found that the contribution required in order to be considered as an inventor, must be "creative or intelligent in its essence" and that this ultimately required human intervention. No jurisdiction in the survey provided for an AI system "itself" to be considered an inventor or joint inventor.

The study concluded not only that AI systems cannot be considered inventors in any of the jurisdictions surveyed, but that there are no convincing reasons to change this. At present, from a legal perspective at least, AI systems are viewed as a sophisticated tool which aid invention but which cannot be considered to be the deviser of the invention. An invention independently generated by an AI system cannot, therefore, be entitled to a patent.  This raises the question of whether the use of AI systems as part of the inventive process could actually prevent entitlement to a patent, if it were not possible to establish a human inventor?

The technical reality might be that the AI system did, in fact, "devise" the invention in question, even if the legal analysis cannot accommodate this conclusion.  Under the Patents Act 1977, entitlement to a patent ultimately flows from being the actual deviser (inventor), either to the inventor itself (or joint inventors themselves) or someone deriving entitlement from the inventor(s) by agreement or statute (e.g. an employer of the "devising" employee). If there is no human involvement, and therefore no inventor for the purposes of the Patents Act 1977, does that then mean that there is no entitlement to the invention? If a human could not be identified as the (co-)inventor then it would appear to be the case.

The Patents Act 1977 is unequivocal that "no other person" may be granted a patent. There are no provisions in the Patents Act 1977 equivalent to those in the Copyright Designs and Patents Act 1988 regarding the ownership of computer-generated copyright works (discussed further in "AI and IP: Licensing (UK law)". Arguably in the scenario where the AI system is technically responsible for the invention, there is a vacuum of entitlement, since no human can make a valid claim and the AI system has no entitlement to pass to a natural or legal person. There are limited options to revoke a patent for an invention created by an AI system if the applicant instead claims there was a human inventor, but there are potential criminal sanctions for falsification of the registry.

Even if a human can be identified as an inventor, the use of AI systems poses a further risk to patentability in general. If it could be shown that the human involvement was merely to create the conditions for the AI system to undertake a task which resulted in an invention, there is a risk that the "invention" itself may not be patentable for want of an inventive step. We look at this question in more detail in "AI and Patents – "the person skilled in the art"

This article started with the question could an AI system be an inventor for the purposes of a patent application. Under the current state of patent law it seems not. The justification for this is a requirement for creativity or intelligence at the heart of an invention which, whilst not explicit, underpins the patent system. AI systems might support the process of human invention but are not (currently) deemed capable of anything more. It might be worth at this point looking up US Patent No. 5,852,815 in the name of Stephen Thaler – the same man behind the Dabus AI system which made the news this week. Only after this patent was granted did Thaler reveal that the patented invention had been created solely by an artificial neural network he designed known as the "Creativity Machine". And this patent was granted over 20 years ago.

As the gap between the technical reality of the capability of AI systems and legal analysis of inventorship increases, maintaining the legal analysis may become increasingly difficult. One solution the authors suggest may be to introduce entitlement provisions akin to those for copyright to deal with scenarios where the AI system is rightly considered responsible for an invention. However, this then raises another concern, which is whether those with the most powerful AI systems become the dominant force in patentable innovation and whether that will undermine the value and purpose of the patent system.