Generative artificial intelligence (AI) is quickly becoming a major force in the world of technology, with its ability to create new content and even replicate the styles of individual humans, including as a tool used by inventors in the creation of patent-eligible technology. As AI continues to advance, it is raising a host of new legal issues—in particular, intellectual property issues. From questions of ownership and authorship to the use of protected IP in generative AI systems, the legal landscape will need to evolve to keep pace with the rapid development of generative AI.
What Is Generative AI?
Generative AI is a type of artificial intelligence that can create new content based on a set of input training data. This content can take the form of images, text, and audio, and can even be designed to mimic the style of a specific person or group. A popular method by which generative AI operates is by training two neural networks to work against each other, with one network generating content and the other evaluating it to determine how closely it resembles the input training data. As the AI model trains, the generating network becomes increasingly sophisticated and produces a diverse range of outputs. Examples of generative AI systems include Dall‧E (an AI image generator) and ChatGPT (an AI text generator), both projects created by Silicon Valley startup OpenAI. And with generative AI projects continuing to attract capital, the field (and thicket of legal issues) will only grow.
AI Systems Have No Right to Patent Inventorship
In 2022, the Federal Circuit held in Thaler v. Vidal that AI does not have the right to be the inventor of a patent. The Federal Circuit reasoned that the term "inventor" under the United States Patent Act is construed to require a human inventor.
In May 2020, the USPTO denied patent U.S. Patent Application No. 16/524,350 submitted by computer scientist Stephen Thaler for failure to "identify each inventor by his or her legal name." Thaler named an AI system, Device for the Autonomous Bootstrapping of Unified Science (DABUS), as the sole inventor. Thaler claimed that he did not contribute to the conception of the claimed invention and petitioned the USPTO director to vacate the notice of incomplete application in order to allow for the identification of an AI system as the sole inventor. The USPTO ultimately denied the petitions on the ground that "a machine does not qualify as an inventor." The USPTO looked to the plain reading of the Patent Act, noting that it refers to inventors as natural persons. After a subsequent summary judgment in favor of the USPTO in Virginia district court, the Federal Circuit took up the question.
In Thaler, the Federal Circuit found that under the plain meaning of the Patent Act, "individuals" and "inventors" are unambiguously natural persons. While the Patent Act does not define the term "individual," the Federal Circuit pointed to the Supreme Court decision in Mohamad v. Palestinian Auth., explaining "[a]s a noun, 'individual' ordinarily means a human being, a person." The court explained further that there is no indication Congress intended a different reading of the term "individual." In view of this decision, the law is clear (for now) that AI is not capable of being an inventor under the Patent Act because AI is not a "natural person" as required by the Patent Act.
The Inventive AI Assistant
The Patent Act defines "inventor" as "the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention." Similarly, the Patent Act defines "joint inventor" and "coinventor" as "any 1 of the individuals who invented or discovered the subject matter of a joint invention." Therefore, based on the decision in Thaler and the Patent Act, if an inventor is to be recognized by law, 1) they must be a natural person and 2) they must have invented or discovered the subject matter of the invention.
While AI cannot be an inventor because it is not human, can AI assist in the invention or discovery of the subject matter of an invention? Thaler did not expressly address the issue of AI-assisted invention. Thaler explained that the case did not present "the question of whether inventions made by human beings with the assistance of AI are eligible for patent protection."
Thus, it appears AI may be able to assist a human inventor but will not be listed as an inventor on any subsequent patent. However, this still requires that at least one human rise to the level of a coinventor under the law. An inventor is a person who contributes to the conception of an invention described and claimed in a patent application. "Conception" is the "formation in the mind of the inventor, of a definite and permanent idea of the complete and operative invention." "An idea is sufficiently definite and permanent when only ordinary skill would be necessary to reduce the invention to practice, without extensive research or experimentation." In general, a coinventor must have contributed in some significant manner to the conception or the reduction to practice of the invention, as opposed to a contribution that is not insignificant in quality when measured against the dimension of the full invention. Also, the coinventor must have done more than merely explain to the actual inventor well-known concepts and/or the current state of the art.
This will likely pose problems for co-inventorship in the case of AI-assisted inventions. Using Thaler as a guide, a human inventor must contribute to the invention in a legally significant way (i.e., rise to the level of a coinventor under the law) in order for a patent to be granted under the Patent Act. A human inventor working with generative AI provides contributions to the conception of the invention at two stages: 1) the input of training data into the generative AI system and 2) the handling of the subsequent output.
For a human inventor to provide contributions at the input stage, they must do more than explain well-known concepts and/or the current state of the art. For a human inventor to provide contributions at the output stage, they must show they used the output to conceive of and reduce to practice the final invention. Any evidence of conception of the invention by a human inventor at either stage should be documented.
 This is only one example of how a generative AI model may be trained. At the time of writing of this article, popular models include diffusion models, generative adversarial networks (GANs), and generative large language models.
 Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022).
 See 35 U.S.C. §§ 101, 115.
 35 U.S.C. § 100(f).
 35 U.S.C. § 100(g).
 Thaler, 43 F.4th at 1213.
 Hybritech, Inc. v. Monoclonal Antibodies, Inc., 802 F.2d 1367, 1376 (Fed. Cir. 1986).
 Bd. of Educ. v. Am. Bioscience, 333 F.3d 1330 (Fed. Cir. 2003) (quoting Ethicon Inc. v. Surgical Corp., 135 F.3d 1456, 1460 (Fed. Cir. 1998) (internal citations omitted).
 See Pannu v. Iolab Corp., 155 F.3d 1344 (Fed. Cir. 1998).