Following the Thaler decision and pending any further possible appeal, the UK's position on AI inventors is at first sight very clear; under the Patents Act 1977 an AI system cannot be named as an inventor and the right to apply for an invention can only originate from a human inventor. However, the decision leaves two unanswered questions. The first is expressly recognised in the judgment. The second (related) question was not discussed in Thaler but is implicit in the provisions of the Patents Act regarding challenges to inventorship and entitlement. This Part 2 explores these questions and looks ahead to how the law may change in future.
Part 1 of this article summarised the decision in Stephen L Thaler v The Comptroller-General of Patents, Designs And Trade Marks  EWHC 2412 (Pat).
Question 1: Is the owner or controller of an AI system the inventor of any inventions created by the system?
Dr Thaler argued that DABUS was the inventor for his applications and disavowed any role in devising the inventions. However, at several points in the Thaler judgment Smith J suggests that his decision should not be taken as precluding an argument in future that the owner or controller of an AI system that "invents" something could themselves be said to be the inventor. At one point, Smith J even goes so far as to say that it would have been far easier for Dr Thaler to argue his case in this way. Does this mean that Dr Thaler could file applications for other inventions devised by DABUS and name himself as the inventor based on his ownership/control of DABUS?
Approaching this question from the perspective of section 7(3) (the inventor is the actual deviser of the invention) and Yeda (the actual deviser is the natural person who came up with the inventive concept) would lead to a detailed investigation of the interactions between DABUS and Dr Thaler which gave rise to the inventions. Smith J and the UKIPO assumed in Dr Thaler's favour (for the purposes of their decisions) that he was correct in his assertion that DABUS was fully responsible for identifying and solving the problems disclosed in the applications and recognising the novelty and salience of the solutions. They did not investigate further the actual role which Dr Thaler may have played in devising the inventive concepts (e.g. the extent to which his involvement in developing DABUS may have contributed).
Generalising from Dr Thaler's applications, it appears inconsistent with the approach to the definition of "inventor" endorsed by Smith J, to state that a person could be considered the inventor of any inventions devised by an AI system purely by virtue of owning that system. To take an extreme example, purchasing an off-the-shelf "invention machine" and pressing the "invent now" button, which results in the machine humming away for a while and then printing out a fully formed patent application which is promptly filed with the UKIPO, would not appear to qualify the owner of the machine as the person who came up with the inventive concept. At the other extreme, a person using an AI based text-to-speech program to dictate a patent application would clearly qualify as the inventor notwithstanding the use of an AI system in the process. There will of course be infinite shades of grey between these two extremes and, other than restating established principles, the Thaler decision does not provide any guidance on where to draw the line.
Question 2: What happens in practice if you name a human inventor for an invention devised by an AI system?
While Question 1 throws up complex issues regarding the boundary between human and AI contributions, as a practical matter section 7(4) of the Patents Act requires the UKIPO to make the presumption that the person who applies for the grant of a patent is the person entitled to do so. As Smith J pointed out, this presumption is rebuttable and the UKIPO is entitled to reject applications where the applicant’s belief as to inventorship or their entitlement to be granted a patent is incapable of justification. However, what if Dr Thaler's applications had been filed naming him both as the applicant and inventor and (as is usually the case) the documents filed with the applications said nothing about the process by which the inventions were devised? The UKIPO's role (as stated expressly in para 7.13 of the Manual of Patent Practice) is not to question assertions made by the applicant as to circumstances under which an invention has been devised. Assuming Dr Thaler had not himself highlighted the issue, the UKIPO would have processed the applications in the standard way.
The practical solution to the concern that inventions devised by an AI system cannot be patented under the Patents Act therefore appears to be always naming a human inventor for the application. However, assuming this did happen, would someone else be able to rebut the presumption under s7(4) and demonstrate that the human named as the inventor on the application was not in fact the inventor, and the person who applied for the patent was not entitled to do so? Both questions (inventorship and entitlement) are open to challenge by third parties. However, the rules governing challenges to inventorship differ in two important respects to the rules governing challenges to entitlement, leading to some unexpected consequences.
The first difference is the requirements regarding standing to bring the challenge. Under section 13(3), any person may challenge the naming of an inventor. However, under s8(1)/s37(1), standing to ask the Comptroller or court to determine questions regarding entitlement are limited to persons claiming an interest in the patent.
The second difference is the consequences of a successful challenge. A successful challenge to the patentee’s (or applicant’s) entitlement to be granted the patent will generally result in another party obtaining all or a part of the property in the patent (s8(2)/37(2)). The grant of a patent to a person who was not entitled to the grant is also a ground for revocation of the patent under s72(1)(c). As with other challenges to entitlement, standing to rely on this ground is limited to persons who themselves have a claim to be entitled to the patent. In contrast, the consequence of a successful challenge to those named as the inventor is the removal of that person (s13(3)). Importantly, the Patents Act does not specify the incorrect naming of an inventor as a ground for revocation of a granted patent.
How would this apply to the situation in which a human inventor is named but the invention was in fact devised by an AI system? Looking first at entitlement, there would be no person with standing to challenge the applicant/patentees entitlement as, adopting the reasoning from Thaler, no primary right to apply for the patent would have arisen (because only people can be inventors), and such a right could not have been transferred to any person by the AI system who actually devised the invention. In other words, the only entity able to challenge entitlement would be the AI system, but it would have no standing to do so as it lacks personhood. Once granted, the patent would not be susceptible to revocation under s72(1)(c), as no person would have standing to rely on this ground of revocation.
However, any person could ask the UKIPO, or a court, to remove the human named on the patent on the basis that the invention was devised by an AI system. There would of course be significant evidential challenges in a third party establishing that the named human inventor was not in fact the natural person who came up with the inventive concept. Assuming these could be overcome, the consequence of a successful challenge would be the removal of the human inventor. The UKIPO or a court would now find itself in a quandary. The human inventor must be removed but, applying the Thaler decision, the AI system which devised the invention cannot be named as the inventor in their place. To the extent this challenge took place during the application phase, the UKIPO may be willing to apply a creative interpretation of section 13 of the Patents Act to deny the application on the basis that it now has no named inventor. However, if the challenge arose once the patent had been granted, there do not currently appear to be any grounds under the Patents Act which would permit the court or the UKIPO to revoke the patent. The net result would be a patent which is registered and enforceable in the UK with no named inventors.
Question 2 is a hypothetical one, in the sense that it is premised on an AI system being solely responsible for devising the inventive concept of a patent. Some may therefore dismiss it as a question which has no immediate need of an answer, as they consider it impossible that an AI system could be solely responsible for devising an inventive concept, at least in the short to medium term. There may be some truth to this. However, there are also compelling reasons why it is relevant to consider the question now.
First, as demonstrated by the Thaler case, many people hold strong beliefs around the subject of AI inventions and inventors and are willing to file applications designed to test the existing framework; avoiding the question may not be an option.
Second, any changes to legislation will require significant consultation and debate. As many have pointed out with the regulation of AI in general, given the potential implications of AI technology, it is far better to start the discussion now than to be overtaken by events.
Third, issues relating to AI inventorship may come up indirectly in other disputes regarding ownership of inventions which have benefited from the use of AI technology. A common feature of many AI systems is that their creation and use involves the combination of efforts from a number of sources; those who develop the underlying learning algorithm, supply and select the training data to be used, set a particular objective for the system to achieve and review the results of its operation, may all be separate individuals or teams from different organisations. To the extent patent ownership is not addressed contractually, disputes may arise between these participants as to who is the actual deviser of an invention developed with the assistance of the AI system. The starting point to address this issue would be the Yeda approach of looking for the natural person(s) who came up with the inventive concept. However, the issue of AI inventorship may crop up here, as it may suit a party to run a case that if they are not entitled to the invention then it must have been devised by the AI (the implication being that the other party to the dispute should not be entitled to the patent).
The Thaler decision is not likely to be the end of the story regarding AI inventorship. Dr Thaler also filed patent applications for DABUS derived inventions in the US, EPO, Germany, Israel, Taiwan, India, Japan, China, South Korea, Australia, Canada and under the PCT. Other than the UK, the applications have thus far been rejected by at least the USPTO, the EPO and DPMA in Germany, with appeals currently pending in each case.
The issue of AI inventorship is also attracting significant attention from regulators. The UKIPO is currently running a call for views on the interface between patents and IP, which directly address points raised by the DABUS applications. WIPO is also engaged in an ongoing conversation on AI and IP, which includes AI inventorship.
Where will this lead? One concern is that existing rules could result in a class of inventions for which it is not possible to obtain a patent, because the inventive concept has been devised by an AI system. An equivalent issue exists in relation to copyright, where numerous examples already exist of AI systems creating artworks, songs and literature. However proposals to introduce a new right to protect works which would qualify for copyright protection but for their creation by an AI system (as opposed to a human author), have been met with strong resistance (see for example the discussion at the 2019 AIPPI London Congress reported here). It is not yet clear whether any proposals to permit AI inventors will meet the same resistance. The only thing we can be sure of is that the Thaler decision will be far from the last word on AI inventorship.