DABUS, the inventor that wasn’t?
DABUS, the inventor that wasn’t?
Louis-Pierre Gravelle[*]
ROBIC, LLP
Lawyers, Patent and Trademark Agents
The now well-known “DABUS” case continues to make headlines, with the USPTO having now issued a legal decision by the Deputy Commissioner for Patent Examination Policy denying DABUS, the AI machine, status as an “inventor”.
This case stems from applications initially filed in the Fall of 2018 in Europe : EP 3 564 144 and EP 3 563 896, naming “DABUS” as the inventor of the inventions disclosed. US application 16/524,350 was filed on July 29, 2019, and is directed to “Devices and Methods for Attracting Enhanced Attention”, equivalent to EP 3 564 896. The application in the United States names “DABUS”, an AI machine, as the sole inventor.
At the formalities stage of review of the application by the USPTO, the Office issued a “Notice to file Missing Parts”, since the Application failed to name the inventor by her/his name. The owner of the application, Stephen Thaler, filed a petition for review of inventorship, stating that DABUS autonomously and independently created the claimed invention, and, probably more interestingly, DABUS was not specifically trained in the relevant area of the invention. The argument articulated a distinction between what is generally known as “narrow AI”, i.e. tools that are designed and taught to solve specific problems and “general AI”, where a collection of AI algorithms are designed with general knowledge and can adapt to solve problems based on their general “knowledge”. The word “knowledge” is in quotation, since many experts contend that a true “general AI” machines is decades away, if even possible. Here, Mr Thaler was most likely trying to steer the question away from identifying who trained a “narrow AI” machine, and therefore should be most likely be named as an inventor, to “the AI was trained on general knowledge, and based on a query made to the AI, the AI returned the invention that is the subject of the present application”.[†]
Unsurprisingly, and following similar decisions in the EPO and in the UK, the USPTO has taken the position that inventorship in the United States is limited to natural persons. The USPTO relied on statutes and decisions in non-patent contexts to support its interpretation that an “inventor” is a natural person. In the US Patent Act, 35 USC 100(f)-(g) use words like “individual” and “whoever”, which the USPTO stated implied natural persons. In addition, 35 USC 115(b) requires that the oath or declaration accompanying a patent applications be signed by an individual when that individual believes “himself or herself” to be an inventor.
This decision is consistent with the decisions in the EPO and the UK, where the offices did not recognize DABUS as being an inventor. In the EPO, the decision turned mainly on the rights that accompany inventorship under the patent system – who would have the right to enforce the patent, who would be able to transfer the rights. In other words, if we assume that an AI can be named as an inventor, and the AI refuses to sign say a Declaration of invention or an Assignment, how would one proceed to compel a signature? Identifying a disembodied entity as an inventor should logically procure that entity with the rights and obligations that befall on all human beings, i.e. legal personality. Clearly, the USPTO, and the case law in the United States, did not support such a conclusion.
Interestingly, in the UK, the UKIPO recognized DABUS as the “creator” of the invention, but could not find support in the legislation or case law to support the conclusion that an AI machine could also be named as an inventor.
It therefor appears clear that at least in the US and Europe, there is no support for naming an AI machine as an inventor on a patent application. While that question seems unambiguously settled, other questions remain, such as co-inventorship between humans and non-human entities, and the consequence of naming only humans as inventors, where it can be demonstrated that an AI machine was entirely or even partly responsible for the invention as claimed. Would the latter case under the current legislative framework, assimilate the AI machine to a “tool” operating under the direction and control of a human?
These issues and others are certain to stimulate much discussion and debate in the coming years, and may even trigger legislative changes to accommodate the coming reality that extremely powerful machines will play an important role in future technological, societal developments.
[*]Louis-Pierre Gravelle is a Lawyer, Engineer, Patent Agent and Partner for ROBIC, llp, a firm of lawyers, patent and trademark agents.
[†] The Decisions in the EPO, the UKIPO, and the USPTO, as well as the arguments presented by Mr Thaler use the expression “AI” as a “suitcase word”. In fact, AI is not a countable noun – AI is a scientific discipline, and that means that AI is a collection of concepts, problems and methods for solving them. It follows that an AI method is not really an “entity”; it is disembodied. See for example University of Helsinky’s course Elements of AI.