27 November 2020

AI inventors? Why should we care?

In the Autumn/Winter 2019 issue of Inside IP, we reported on the filing of a test case which argued it should be possible to name an Artificial Intelligence (AI) as the inventor on a patent application. Since then, some of the world’s biggest patent offices have decided that current law does not allow this, but the issue of AI inventors is far from settled.

In 2018, several patent applications were filed in the name of Dr Stephen Thaler for inventions purportedly invented by AI named DABUS (Device for the Autonomous Bootstrapping of Unified Sentience). On the grounds that DABUS independently conceived of the inventions and that no human inventor could be identified, the applications list DABUS as the sole inventor.

Formal inventorship requirements vary across different jurisdictions, but the USPTO, the EPO and the UKIPO have all rejected the possibility that DABUS can be named as an inventor on the patent applications, reaching the common consensus that, under current legislation, an inventor must be a ‘natural person’ for the purposes of a patent application.

The EPO and USPTO decisions are still under appeal, but a UK High Court judgment in late September dismissed the corresponding UK appeal, affirming the decision of the UKIPO that a person – meaning a ‘natural person’ and not merely a legal person – must be identified as the inventor. That is to say, even if DABUS could be considered to have created the inventions, it could not be considered to have been an inventor for the purposes of the UK Patents Act because the statutory requirement that the inventor is the “actual devisor of the invention” implies there is a human ‘someone’ to do the devising.

The question of whether an AI can be an inventor is not merely academic, but part of a broader point about AI inventions and their commercial reality. Being an inventor conveys certain legal rights and is integral to the concept of patent ownership and entitlement. Under the UK Patents Act, the right to grant of a patent belongs “primarily to the inventor”, subject to any other agreement made by the inventor (such as an assignment or employment contract). It is through these legal agreements with inventors that companies can derive the right to own patents. An AI such as DABUS lacks legal personality and is not able to own property or to enter into legal agreements to assign property rights. Furthermore, as the UKIPO decision notes, “even if I were to accept that DABUS meets the requirements of the Act to be considered an inventor, derivation of right through ownership of the inventor does not meet the requirements of section 7(2)” of the UK Patents Act (emphasis added).

Accordingly, even if DABUS could be listed as the inventor of the patent applications in question, Dr Thaler has no rights in them and is not entitled to grant of a patent for the invention.

There is a clear tension in this current position. On the one hand, the UKIPO is prepared to assume that an AI can invent something with no human input, such that no natural person can be considered the actual devisor; on the other hand, the UKIPO has decided that a patent application cannot be obtained if the AI is listed as the inventor and that the owner of any such AI inventor is not entitled to apply for a patent for the invention. At the same time, there are legal implications for filing a patent application to which one is not entitled, as well as for naming the wrong inventor; in the United States in particular, “identifying a natural person, who did not invent or discover the subject matter of the invention, as the inventor in a patent application would be in conflict with the patent statutes” and could lead to the patent being held unenforceable.

As these test cases and resulting patent office decisions have highlighted, there is currently a gap in the law, the result of which is that the existing patent system does not cater for the protection of inventions generated solely by AI. While the UK High Court judgment hints that interpretation might allow to close this gap within the framework of existing legislation, the question of how AI inventions should be treated is still an open one.

It is also a question which has garnered increasing amounts of attention in the last year or so, with the UK government opening a consultation on the matter in September and a third WIPO session on Intellectual Property and AI scheduled for November. The level of interest is understandable when we consider the history of patents, the public policy rationale and the ‘patent bargain’, which grants a 20 year patent monopoly to an inventor for two reasons: (i) to have inventors share what they learn, so others coming after them can build on their ideas; and (ii) to reward investment in research. As we have seen, the starting place for the grant of a patent has so far been the human inventor, the ‘natural person’, albeit a company can then own and exploit the patent.

However, as the DABUS cases show, the questions we need to be asking now are: does it matter if there is no human inventor? What does it mean for investment and transparent sharing of learning if a company cannot obtain a patent because the inventor was an AI?

These questions may not seem that pressing at the moment. These test cases are the first patent applications to name an AI inventor and so far there has been no outcry that businesses are failing to obtain patents because of arguments concerning AI inventors. However, the DABUS applications illustrate the direction that this technology is heading in and as AI capabilities increase we may reach a tipping point where significant numbers of inventions can no longer be protected by patents.

We suspect that an inability for a company to see a return on its investment in research is likely to change the business interest in these questions quickly. If this happens, policy makers will have to consider the patent bargain and grapple with questions such as: If patents for inventions generated by AI are not available, will businesses stop publicly sharing their knowledge and/or struggle to obtain investment, and is that a problem? Does the cost of development using AI warrant patent protection or are inventions generated by AI based on a comparatively cheap form of research that shouldn’t be afforded the same protection as other inventions? Is financial investment in research worthy of policy protection at all or is it the human endeavour which we seek to reward? In particular, policy makers will have to decide what the purpose of the patent system is and whether, on balance, there is a sufficient policy reason to change it.

We cannot know what future policy debates will conclude. In the meantime, the default position is that patent protection requires there to be a human devisor and companies working in this field should keep a close watch on the origins of inventions made using AI. If the issues and questions above seem relevant to the innovations driving your business, then contact your usual Venner Shipley advisor, because this is the debate for you.