No protection for machine-led invention? | Fieldfisher
Skip to main content

No protection for machine-led invention?


United Kingdom

Can an AI be an inventor? Recent decisions of the High Court and the patent offices in the UK, EU, and US shed some light on this, but there may still be some problems to be resolved. In this article, Chris Eastham and Natasha Rao examine the decisions made by the various patent offices and courts in these countries in further detail and consider their implications for machine-led invention.


The concept of AI-assisted design—the use of artificial intelligence as a tool to compose music, create art, or explore novel scientific concepts—will not be new to readers. As the role of AI in the act of creation becomes more and more substantial, and begins to exceed the role played by humans, complex moral and legal questions arise about whether we should permit such inventions to be monopolised through patent protection.

These questions have recently been brought to a head by the 'DABUS' case, which concerned applications made by Dr Stephen Thaler to patent two inventions which he claims were made by a 'creative machine' he owns named DABUS (see our previous blog on this topic for further details). In these patent applications—made in the US, in Europe, and in the UK—Dr Thaler listed himself as the owner of the patents, and DABUS as the inventor.

US Decision

The United States Patent and Trademark Office (USPTO) recently published its decision on this case and ultimately concluded that an AI cannot be an inventor under US patent law as it is not a natural person. The USPTO identified various provisions in Title 35 of the United States Code that make reference to a natural person either by use of pronouns such as "whoever invents or discovers" or "himself or herself" or directly, by specifying, for example, that the inventor who executes an oath or declaration must be a person. In addition, the USPTO identified existing US case law indicating that corporations and states cannot be inventors because they are not natural persons, and that there must be evidence of conception of a definite idea in the mind of the inventor which is only possible in the context of a natural person.

European and UK decisions

The European Patent Office (EPO) and UK Intellectual Property Office (IPO) also both rejected Dr Thaler's applications on the basis that, under current European and UK law, the inventor identified in a patent has to be a natural person. Both the EPO and IPO further pointed out that, even if DABUS could be regarded as the inventor, Dr Thaler would not be entitled to claim ownership of those patents, because DABUS (as a non-legal or natural person) could not have transferred ownership of the patents to Dr Thaler in any case.

The IPO also rejected the applicant's argument that not allowing an AI to be an inventor for a patent would risk undermining the fundamental role of the patent system (namely, to encourage innovation through time-limited monopolies in exchange for public disclosure) because it would disincentivise the AI from disseminating information about its inventions. The IPO commented that, given that an AI is motivated to innovate through its design, there are other ways such information may be disseminated freely, such as through the internet. Further, allowing an AI to be an inventor was unlikely to affect whether information about its inventions would be publicly disseminated, given that that decision would always ultimately be down to the machine's owner or developers.

The IPO's decision has recently been upheld on appeal by the UK High Court, Mr Justice Smith summarising neatly that "the problem that [Dr Thaler] has is that there is nothing to be transferred to him and nobody capable of transferring it." The judge hinted that the court might be open to granting a patent for an AI-led invention where the owner or controller of the AI is themselves listed as the inventor; however, this argument had not been advanced by Dr Thaler because he considered that he would illegitimately be taking credit for an invention that was not his.

Wider consequences

Whilst these decisions clarify that an AI cannot be regarded as an inventor under patent law, in the UK and Europe they give rise to a situation that may come as a surprise to those involved in creating scientific inventions. Namely, that in circumstances where a machine has genuinely 'invented' something, patent protection will not be available. While the USPTO did not expressly deal with the question of who had 'created' the inventions, a similar conundrum exists in the US given the requirement for conception of a definite idea in the mind of the inventor as described in the US Manual of Patent Examining Procedure.

If an AI can easily be identified as merely the tool used in the development of a new invention, and the inventive step is taken by the individual using the AI, then that individual will be the inventor. The right to apply for a patent is then often passed to the person or company funding the work either by contract or by an employment relationship.

But what will happen if the line between computer-aided design and AI-generated inventions becomes blurred? If the AI is fully responsible for making the invention, the current legislative landscape precludes the invention from being patented. Note that the person who owns the AI has no right to apply for a patent merely by virtue of that ownership, unlike the situation in respect of copyright and design rights in the UK, where the 'author' or 'designer' of a computer-generated work is the person "by whom the arrangements necessary for the creation of the work are undertaken." The courts' differing attitude to patent ownership seems result from the fact that, unlike in the case of copyright or unregistered design rights, a patent does not come into existence automatically upon invention and must ultimately be applied for by a person) (see paragraph 49 of the recent High Court decision).

The IPO commented in its decision that the present patent system may not be suited to inventions by AI systems, and has encouraged further debate on the issue by way of a public consultation (responses to which are due by 30 November 2020).

If these inventions are not patentable, then a substantial commercial incentive to disseminate the information is removed. It is therefore possible that either the inventions will not be widely disseminated (and instead retained and used on a confidential basis), or that the use of AI to generate new inventions will be discouraged, potentially inhibiting innovation. On the other hand, unless the ability to use AI for machine-led invention is democratised, allowing patents to be granted for AI-generated inventions could lead to further concentration of market power in big tech firms and an increasing disparity between the AI 'haves' and 'have nots'.

To address these moral questions, it seems to us that there is a pressing need for legislators to consider whether there should be a legal framework in place to allow for the patentability of AI-developed inventions. Should there be a re-assessment of the definition of 'inventor', for example, such that it covers non-natural persons, and/or allows for the creator, developer or programmer of an AI to be considered the inventor of any inventions 'created' by that AI? Should machines be given a limited legal personality so that ownership of a patent (where a machine is listed as the inventor) automatically vests in the machine's owner?

We welcome any views on the above questions and strongly encourage any interested parties to take part in the consultation (the IPO's specific questions can be found here).

Sign up to our email digest

Click to subscribe or manage your email preferences.