Creation vs. Conception: Can an AI Machine be an Inventor?

October 28, 2020 – Article
Computational Creativity, a new subfield of Artificial Intelligence (AI), is the study of software that mimics human creativity. AI researchers in this field are designing software to paint pictures, write poems, and compose music. In science and engineering, some believe that the output of certain sophisticated AI software can represent a new invention conceived of, not by the author, user, or trainer of the software, but by the software itself, and that the software or the “AI machine” should be listed as the inventor on a patent application for the new invention. However, the question of whether an AI machine can create or generate an invention is different from the question of whether the AI machine can be considered an inventor, as that term is defined under U.S. patent law, and can properly be listed as the inventor on a U.S. patent application.

The AI-machine-as-inventor proposition was recently tested by Dr. Stephen Thayer, who filed patent applications in the USPTO, the UKIPO, and the EPO for inventions alleged invented by a neural-network-based AI machine entitled “DABUS.” In each case, the inventor was listed as “DABUS” on the application, with Dr. Thayer listed as the applicant and assignee. Each Office objected to the listed inventor information for the same reason: DABUS is not a natural person. Notably, in rejecting DABUS as the inventor, the UKIPO accepted the assertion that DABUS nevertheless “created” the invention.


According to Dr. Thaler, DABUS includes “…a first artificial neural network, made up of a series of smaller neural networks, that has been trained with general information from various knowledge domains. This first network generates novel ideas in response to self-perturbations of connection weights between neurons and component neural nets therein. A second ‘critic’ artificial neural network monitors the first neural network for new ideas, and identifies those ideas that are sufficiently novel compared to the machine’s pre-existing knowledge base. The critic net also generates an effective response that in turn injects/retracts perturbations to selectively form and ripen ideas having the most novelty, utility, or value.” Ryan Abbott, (2020).

Further, Dr. Thaler argues that DABUS was not created to solve any particular problem, but only received training in general knowledge in a field. Moreover, Dr. Thaler contends that DABUS itself “…proceeded to independently conceive of the invention, and to identify it as novel and salient.” Id. Thus, Dr. Thaler seeks to distinguish DABUS from other AI machines in which the designer of the AI machine was largely responsible for a discovered invention “…by virtue of having exhibited inventive skill in developing a program to solve a particular problem, by skillfully selecting data to provide to a machine, or by identifying the output of a machine as inventive.” Id.

In a first one of the U.S. patent applications in question (“Neural Flame,” Application No. 16,524,350), DABUS is asserted to have invented a warning light for attracting enhanced attention, including a controllable light source configured to pulse based on an input signal of a lacunar pulse train having specific fractal properties. A second application (“Fractal Container,” Application No. 16/524,532) is directed to a food or beverage container having a generally cylindrical wall with a fractal profile of convex and concave fractal elements.

For each application, DABUS was listed as the inventor. An assignment was also filed assigning all rights in the invention from DABUS as the inventor to Dr. Thaler as the assignee. The assignment was executed by Dr. Thaler as the legal representative of DABUS.

USPTO Decision

The USPTO objected to DABUS being listed as the inventor in both applications, and on April 22, 2020, published its decision in the Neural Flame application. In particular, the USPTO issued a Notice to File Missing Parts, objecting to the Application Data Sheet (ADS) listing the inventor’s given name as “DABUS” and the family name as “Invention generated by artificial intelligence.”

Dr. Thaler filed two petitions requesting that the Notice be vacated, arguing that inventorship should not be restricted to natural persons because a machine that would meet the inventorship criteria if it were a natural person should also qualify as an inventor, and it is actually unlawful for the owner of an AI machine, rather than the AI machine itself, to be listed as inventor if the owner did not in fact create the invention. Further, Dr. Thaler also cited several public policy motivations for allowing AI machines as inventors.

Regarding the first argument, Dr. Thaler acknowledges that some Federal Circuit decisions and certain patent statutes imply that an inventor must be an “individual” or a “natural person.” However, he argues that Congress simply assumed that only a person could invent, and overlooked the possibility that a machine could be an inventor. Moreover, he argues that the court decisions were directed at preventing inventorship by a state or corporation, not a machine.

Regarding the second argument, Dr. Thaler argues that DABUS must be listed as the inventor because DABUS conceived of the invention, and to list someone else who did not contribute to the conception of the invention (e.g., the inventor of DABUS) would be illegal under 18 USC 1001. Further, Dr. Thaler argues that if DABUS cannot be listed as the inventor, no person can be listed such that the invention will enter the public domain once disclosed, which is an undesirable policy result.

Both of Dr. Thaler’s petitions were denied. In the latter decision, the USPTO cites to the MPEP, various U.S. statutes (35 USC § 100, § 101, and § 115), and several Federal Circuit opinions to conclude that “…conception…must be performed by a natural person.” USPTO Decision on Petition, April 22, 2020,page 5.

In particular, the USPTO noted that 35 U.S.C. § 100(f) states that “[t]he term ‘inventor means the individual…who invented or discovered the subject matter of the invention,” while 35 U.S.C. § 101 states that ‘[w]hoever invents…” Moreover, 35 U.S.C. § 115(a) requires that “…each individual who is the inventor…of a claimed invention in an application for patent shall execute an oath or declaration…,” and 35 U.S.C. § 115(b) requires that the oath or declaration “…shall contain statements that…such individual believes himself or herself to be the original inventor….”.

Thus, the USPTO concluded that interpreting “inventor” broadly to encompass an AI machine “…would contradict the plain reading of the patent statutes that refer to persons or individuals.” USPTO Decision on Petition, page 4.

While acknowledging that they were directed to states or corporation, the USPTO further cited to several Federal Circuit cases that concluded that conception is a mental act and that “…inventors must be natural persons.” Univ. of Utah v. Max-Planck-Gesellschaft zur Forderung de Wissenschaften e. V., 734 F.3d 1315 at 1323 (Fed. Cir. 2013). Finally, the USPTO cited to the MPEP, which defines conception as “the complete performance of the mental part of the inventive act” and “the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention as it is thereafter to be applied in practice.” See MPEP §2138.04 (emphasis added). The USPTO noted that the terms “mental” and “mind” indicate that conception must be performed by a natural person.

Other Legal Issues

Although not specifically addressed by the USPTO, the DABUS patent applications raise other legal issues, some of which were mentioned in the UKIPO and EPO decisions.

First, if DABUS is the inventor but is not a legal person and cannot own property, how can DABUS assign rights to intellectual property? In other words, it is not clear how Dr. Thaler as assignee obtained the rights to the invention from DABUS if DABUS is the inventor. In the UKIPO case, Dr. Thaler’s team acknowledged that DABUS is not a legal entity, but argued that the rights to the invention were acquired by virtue of ownership of the inventor. However, the UKIPO rejected this argument as not being covered under existing UK law.

Further, if DABUS is the inventor, the “applicant” can be any person or legal entity (1) to whom the inventor has assigned, (2) to whom the inventor is under an obligation to assign, or (3) who otherwise shows sufficient proprietary interest in the matter. See 37 C.F.R § 1.42. Thus, if DABUS is listed as the inventor, but is admittedly not a person who can satisfy (1) or (2) above, Dr. Thaler must show a sufficient proprietary interest in the matter to be the applicant, which requires submitting a petition showing that a sufficient proprietary interest in the patent application.

Policy Considerations

Dr. Thaler argues that allowing patents on the output of an AI machine incentivizes the development of such machines and promotes innovation, while “[f]ailure to permit patent protection for the output of autonomously inventive machines threatens to undermine the patent system by failing to encourage the production of socially valuable inventions.” See Ryan Abbott, Moreover, he argues that not listing DABUS as the inventor would be immoral because doing so would allow others to take unearned credit for an invention, which devalues and diminishes the role of inventorship in general. Further, if no patent applicant can be filed because no individual can be legally or morally listed as the inventor, innovation using AI machines is discouraged.

However, the USPTO decision would not appear to generally discourage patent applications on inventions created or generated by AI machines. Even if an AI machine is allowed to be listed as the inventor, the invention would still be owned and potentially monetized by the assignee. Thus, it would appear that the incentives for researchers to develop invention-producing AI machines would not be affected either way by whether a natural person or an AI machine is listed as the inventor on the patent application. Moreover, Dr. Thaler’s legal and moral arguments in favor of computer inventorship are predicated on the assumption that the AI machine did in fact create or invent the claimed subject matter, which is not a settled issue and was not addressed by the USPTO.

U.S. District Court Case

On August 6, 2020, Dr. Thaler filed a complaint against the USPTO in the Eastern District of Virginia (Stephen Thaler v. Andrei Iancu and the USPTO) asking for declarative and injunctive relief that would set aside the USPTO Decision. The complaint asserts that “requiring a natural person to be listed as an inventor as a condition of patentability is contrary to law,” and that the USPTO actions were arbitrary, capricious, and an abuse of discretion, unsupported by substantial evidence.

In the complaint, Dr. Thaler presents arguments similar to those set forth in the USPTO petitions, but additionally argues that the mental aspect of conception should not be a barrier to AI inventorship: “[w]hether or not AI can ‘conceive’ of an invention, ‘think,’ or have something analogous to consciousness should be irrelevant with regards to the existence of patent rights.” Complaint, Case 1:20-cv-00903-LMB-TCB, page 12, item 57.

In particular, Dr. Thaler argues that a “conception” requirement that requires an AI machine to effectively mimic human thought would serve no purpose (other than to exclude AI machines), and that designing a completely digitized version of the human brain that could mimic human thought would likely be inefficient and difficult for the USPTO to verify. Further, citing to analysis by Alan Turing, one of the founders of computer science, regarding the question of whether computers can think, Dr. Thaler argues that a better question is whether an individual could tell the difference between responses from a computer and from a person, i.e., whether machines could perform in the same manner as “thinking” entities. Finally, Dr. Thaler argues against a conception test for inventorship because, in his view, the patent system should be indifferent as to how an invention comes about, noting that the “flash of genius” requirement was abolished by the 1952 Patent Act.

Thus, Dr. Thaler believes that the mental aspect of conception required by the USPTO for human inventors should not be required of AI machines since (1) conception is merely a by-product of an assumption by Congress that only people can invent, (2) even if AI machines could mimic consciousness, it would be hard for the USPTO to verify, and (3) a better standard of judging “consciousness” or “thinking” for AI machines is to evaluate their output compared to humans.

Going forward, it will be interesting to see if the courts will find Dr. Thaler’s arguments more persuasive than the USPTO did, and whether Congress will consider changes to U.S. patent law to allow for AI inventorship.