Gradualism—The status quo will solve over time as this evolves into a relevant issue, buying time to parse through unanswered questions—Rushing it now will upend the patent system.

The lag between technological developments in artificial intelligence (AI) and the law that governs it is growing as the pace of AI adoption increases. In patent law, open questions have evolved from whether AI is itself patentable to questions surrounding inventorship where AI is involved in the development of an invention. Inventorship is a key inquiry in patent law, as an inventor owns the patent unless and until the inventor assigns the patent to another entity.[11]

AI, an invention itself, is now being used to develop new patentable inventions, notably in the pharmaceutical industry, where it is being used both to discover new drugs and repurpose existing drugs.[12] Two main questions about inventorship arise when AI is involved in the invention process: is AI itself an inventor under current patent law? And who is the proper inventor if not AI itself?

To date, the first question has also been addressed by several patent offices around the world, with varied outcomes. The EPO, IPO, and USPTO have all declined to extend inventorship to AI. Common across these rejections is the requirement that inventors be natural persons. All of these decisions have been issued in response to Thaler’s DABUS AI system.

The EPO refused two such applications, finding that an inventor designated in the application has to be a human being, not a machine.[13] In its decision, the EPO noted that there appears to be an internationally applicable standard that the understanding of the term inventor refers to a natural person, and that numerous courts have issued decisions consistent with this understanding.[14] These decisions were appealed, and the EPO Board of Appeal issued two preliminary communications on June 21, 2021 agreeing with the EPO’s decisions, stating that, under Article 60 of the European Patent Convention (EPC), an inventor on a patent application must have “legal capacity,” meaning the “ability, according to a source of law, to be the subject of rights and duties,” as governed by national law.[15]

Similarly, the UK IPO rejected the applicant’s argument that DABUS was the properly named inventor as an “inventor and person . . . are one and the same, namely a natural person – a human and not an AI machine.”[16] This decision was upheld on appeal.[17]

In contrast to the EPO and UK IPO decisions, the Federal Court of Australia held in July 2021 that DABUS could be an inventor under Australia’s Patents Act. The court reasoned that under the Australian act an inventor is an agent, which could include artificial intelligence systems, and the conclusion was consistent with “the reality in terms of many otherwise patentable inventions where it cannot sensibly be said that a human is the inventor,” and nothing in the act dictated otherwise.[18]

Similarly, South Africa’s Companies and Intellectual Property Commission granted a patent titled “FOOD CONTAINER AND DEVICES AND METHODS FOR ATTRACTING ENHANCED ATTENTION” which lists as an inventor “DABUS,” noting “[t]he invention was autonomously generated by an artificial intelligence.” [19] The Australian and South African decisions are likely to be appealed.

So who gets the credit?

While it appears to be fairly established that AI cannot itself be listed as a named inventor under the current statutory scheme in the United States, the question of who, if anyone, can take credit for AI’s inventions remains unanswered.[20] In rejecting DABUS as an inventor, the USPTO noted that it had not made any determination concerning who or what created the invention claimed in the application. This seems to allow room for an individual to take the credit.

In some instances, AI has allegedly generated inventions that resulted in patents, but the AI’s developer’s listed themselves as the sole inventors.[21] For example, the Creativity Machine, which Thaler had disclosed in 1994, generated novel ideas through the use of a neural network; it was itself the subject of a patent.[22] Thaler was listed as the inventor of a subsequent patent,[23] the subject matter of which he claims was invented by the Creativity Machine.[24] The Creativity Machine’s role was not disclosed to the PTO.

Similarly, John Koza’s Invention Machine was allegedly responsible for inventing the subject matter of US Patent No. 6,847,851, which was not disclosed to the USPTO.[25] Ironically, Thaler’s complaint in his pending District Court case cautions that “future patent applicants may attempt to circumvent the new standard by inaccurately listing a natural person who does not meet inventorship criteria.”[26] Per his complaint, Thaler claimed that he “is in the business of developing advanced AI systems that are capable of generating patentable output in the absence of a person who otherwise meets inventorship criteria.”[27] He also noted that if he listed himself as inventor – which he had done before and remained silent on in his complaint – “he would have two issued patents or be well on his way to that outcome” which “would be an undesirable outcome because it would involve misrepresentations and individuals claiming credit for work they did not perform.”[28]

Thaler’s silence on his “inventorship” for his previous AI-developed patents may be attributed to the severe civil and criminal penalties he cites in his complaint for fraudulently listing an inventor: it is grounds for rejection under 35 U.S.C. Sections 101 and 35 U.S.C. 115, and deliberately misreporting inventorship may result in criminal punishment under 18 U.S.C. Section 1001.[29]

In the United States, “[t]he threshold question in determining inventorship is who conceived the invention.”[30] “Unless a person contributes to the conception of the invention, he is not an inventor. … Insofar as defining an inventor is concerned, reduction to practice, per se, is irrelevant.”[31] Though “it is not essential for the inventor to be personally involved in carrying out process steps…where implementation of those steps does not require the exercise of inventive skill.”[32]

The type of AI may be determinative of inventorship. The USPTO’s 2020 AI report notes that many comments addressed the fact that there is no universally recognized definition of AI.[33] Currently available narrow AI systems are designed to perform well-defined tasks, in contrast to artificial general intelligence that has the capacity to learn and perform tasks more like the human mind. Narrow AI systems still require input and training.[34] Who, then, is the proper inventor where narrow AI is involved? Is it the computer programmer, the owner of the data sets that “teach” AI, or the lab scientist looking for the new pharmaceutical compound or drug target? Patent offices and courts have not yet addressed this question.

A parallel example

A parallel can be drawn between the use of narrow AI and therapeutic antibody discovery. In AI drug discovery, AI helps identify compounds, but human input is still required to test the compounds in vitro.[35] Similarly, in therapeutic antibody discovery, animals or cells are infected with an antigen to induce an immune response and generate antibodies, which are then tested and optimized by research scientists.[36] There is no question that antibodies have routinely been patented, and the research scientists who discover them are the named inventors on those patents. There has been no attempt to date to confer inventorship on an animal or cell line involved in the antibody discovery process.

A similar approach could be followed in AI drug discovery, whereby the scientists involved in the final steps of the process are the inventors. While others are involved in developing and training the AI, they do not have a direct role in developing the end drug. “Conception is the formation in the mind of the inventor, of a definite and permanent idea of the complete and operative invention, as it is hereafter to be applied in practice.”[37] Those who develop and train the AI do not have an idea as to what the complete and operative drug will be. And those involved in the process of testing compounds identified by AI may not have a hand in the development and training of the AI involved. Inventorship is dependent on the claimed invention. Several inventions may arise, from the AI itself to the drugs identified.

As AI becomes more common, and as general AI becomes a reality, perhaps patent offices around the world will confer inventorship status on AI. For now, natural persons will need to take credit for AI-generated and AI-assisted inventions or risk a finding that their application does not comply with 35 U.S.C. § 115(a).

An open question for Congress

Just who those natural persons should be is an open question remaining for the Congress. Allowing AI to be named as an inventor is not without risk to the litigation process and ought to be carefully considered should Congress wish to expand the scope of inventorship.

For example, when applying for a patent, inventors have a duty to disclose to the USPTO all information known to that individual to be material to patentability. Who would be liable for inequitable conduct if, during prosecution, disclosures were omitted? Unlike a human inventor who can be deposed in connection to a proceeding involving the patent, how would the patent’s future challengers be able to discern what art was available to the AI system at the time of the invention? Perhaps artificial general intelligence will advance to a point that these concerns will be mitigated by the time Congress addresses AI inventorship. Until then, the courts and PTO will have to continue to define the contours of inventorship under the existing statutory scheme.