Inventorship Determinations May Get Muddier: Should Naming an Inventor in a Patent Application be Required for AI-Developed Inventions?

July 2020

This article was published on page 8 of the August 2020 edition of IEEE’s Digital Reflector.

By: Greg K. Gerstenzang

Under current US law, there is no provision for determining inventorship or ownership to otherwise inventive solutions developed by artificial intelligence (AI) systems.  AI systems are used in industry and academia to develop solutions to problems in areas such as drug discovery, materials science, and nanotechnology.  The inability to determine inventorship of inventive solutions developed by AI systems in these and other fields makes it nearly impossible to apply for patents or properly assign ownership to these solutions.  Without guidance, this inability to apply for patents for AI-developed inventions or assign ownership may, potentially, stifle innovation and impair one of the objects of the patent system itself – to disseminate knowledge regarding new inventions to the public by describing these inventions in patents.  To incentivize this type of innovation, a solution is needed that allows for AI-developed inventions to be protected, enforced, and assigned.

Under US case law “[t]he threshold question in determining inventorship is who conceived of the invention.”  One who reduces a conceived invention to practice or who performs optimization of a conceived invention, without contributing to the conception of the invention itself, is not considered an inventor.  For example, if a chemist invented a method of synthesizing a novel chemical compound and instructed a lab assistant to run experiments to determine an optimal temperature at which the method should be performed, the chemist, but not the lab assistant, would be considered the inventor of the method.

For inventions that are developed utilizing an AI system, who, if anyone, is the inventor?  Where a person conceives of an inventive solution to a problem and then instructs an AI system to optimize certain features of the inventive solution, the answer is clearer.  The performance of routine optimization by the AI system will not call into question whether the AI system is an inventor.  Things get murkier, however, if a person presents a problem to be solved by an AI system, and the system develops an otherwise inventive solution that the person had not conceived of.  Now, who is the inventor of the solution?  The person did not conceive of the solution, so in accordance with current patent law, the person is not the inventor.  But what if the person programmed the algorithm that the AI system utilized to arrive at the solution?  Again, according to US law, if the person did not conceive of the particular solution arrived at by the AI system, he/she is not the inventor. The question of who may be properly named as the inventor of an otherwise inventive solution developed by an AI system is not just for academics.  To apply for a patent on an invention, one must list the inventor(s).

Declaring an AI system as an inventor because it develops an otherwise inventive solution is inconsistent with current patent statutes and case law.  The patent statutes utilize pronouns (e.g. “Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter … may obtain a patent therefor”) that strongly suggest, but do not explicitly state, that an inventor must be a natural person.  In 2019 a group of patent attorneys known as the “Artificial Inventor Project,” applied for a patent listing an AI system as the inventor and listed the developer of the AI system, Stephen L. Thaler, as the applicant.  The United States Patent and Trademark Office (USPTO) rejected the application for not listing a human inventor, finding that the language of the patent statutes should be interpreted to limit inventors to natural persons, and citing court precedent holding that only natural persons can be “inventors.”

Therefore, if neither the person who instructs an AI system or an AI system itself can be considered the inventor of a solution, could anyone apply for a patent on the solution?  According to current patent law, it would seem the answer is “no.”  This may impact innovation, because the promise by the USPTO, and indeed the Constitution, to grant the right to exclude others from practicing a patented invention provides an incentive for individuals and companies to fund research and development for new inventions. Without the possibility of being able to file patent applications on inventions created by an AI system, there is little protection for the perceived inventors, e.g. the companies or persons who own the AI.

A patent provides its named inventor(s) with rights to the invention defined by the claims of the patent.  However, there is nothing in the law, statute or case law, that holds that an AI system — any other object that may itself be considered property, can own property.  If an AI system cannot “own” any intellectual property rights to an inventive solution it produces, and the operator of the system cannot claim to be the inventor of the solution, then there is no reasonable solution for patenting the solution. Without the patent, there is no defined owner of the intellectual property, e.g. the solution.

Perhaps the concept of “inventor” may need to be expanded to include AI systems operating under the control or direction of a human or to include the human(s) operating the AI systems, regardless of the extent to which they conceived of the inventive solution.  Or perhaps the requirement to name a particular inventor in a patent application may be eliminated, and patents may be applied for naming only applicants or assignees. Presumably, AI systems will continue to advance and will be used more extensively in developing solutions to problems that people may have difficulty otherwise solving. For example, developing or identifying promising drug candidates for different diseases, or new algorithms to identify valuable stocks. The possibilities of what AI might find are limitless.  Therefore, finding the answer to the questions: “how to address the requirements for identification of an inventor in a patent application?” and, “how to assign ownership of AI-developed inventions?” will become imperative as the present absence of possibility to file for and obtain a patent will likely correlate with the reduction of new inventions being shared and protected.