Guidance on using AI to Invent

Due to the quick rise of artificial intelligence (AI), most (if not all) of the laws relating to the US patent system were not written with AI in mind. Court precedent and the United States Patent and Trademark Office (USPTO) guidelines have taken the place of written law to guide the USPTO in matters relating to AI. A notable example of this is Thaler v. Vidal, a Federal Court of Appeals case that determined whether AI can be listed as an inventor on a patent application. In this case, Dr. Stephen Thaler created an AI program that he listed as the only inventor on two US patent applications. The USPTO rejected these applications for lack of a proper inventor. Dr. Thaler appealed to the district court and the federal circuit, where it was held that the patent applications were properly rejected for listing an AI program as the inventor.

In the Thaler case, the federal circuit held that an inventor must be a natural person. This conclusion restricts AI from being an inventor on a patent application but leaves open questions as to whether AI can be used to invent. These open questions remained since August of 2022, when Thaler was decided, until now. The USPTO recently issued new guidance on inventions created with the assistance of AI, where inventions must have a “significant contribution” from a natural person. Notably, the patentability of an invention “focus[es] on human contributions, as patents function to incentivize and reward human ingenuity.” The guidance recognizes that this determination is “difficult to ascertain” without a bright-line test.

The Definition of “Significant Contribution”

To understand the USPTO’s guidance, it is essential to know some background information on how AI programs work. AI programs can function differently depending on their purpose and whether they are generalized or specialized. The program created by Dr. Thaler may be considered a specialized AI program that was explicitly created to invent. However, ChatGPT and other popular AI programs are generalized, using large databases of knowledge to provide answers to a user’s inputs. In these programs, a user provides a prompt for the AI to answer and can offer further prompts to modify the answer provided by the AI.

In the new guidance, the USPTO provides a few generalized examples of what does and does not amount to a significant contribution. What is not a “significant contribution” is when a user “only presents a problem to an AI system,” and the AI provides the solution. An example of this would be a user asking ChatGPT to design pin fins on a heat sink to meet certain criteria. A “significant contribution,” however, can be shown “by the way the person constructs the prompt in view of a specific problem to elicit a particular solution from the AI system.” To differentiate over merely presenting a problem to the AI, the user could add specific configurations for the pin fins, additional criteria or limits to be met, or other further details in the prompt. It is difficult to predict how much detail is necessary in the prompt to meet the requirements of “significant contribution” at this time, but the more detail, the better.

The USPTO provides further (general) examples of what could be considered a significant contribution, including when a user provides an “essential building block” to the AI, or someone who “designs, builds, or trains an AI system in view of a specific problem to elicit a particular solution could be an inventor.” Each of the examples leaves much to the imagination as to what a “significant contribution” is, especially when the latter example of designing, building, or training an AI system is in stark contrast to the results of Thaler v. Vidal.

Using AI to Invent, in Practice

Whether you are an in-house attorney concerned about compliance or an inventor yourself, it can be difficult to determine how much AI involvement is allowed in the invention. It appears that the USPTO is also unclear on how much involvement is required. Aside from the generalized examples provided by the USPTO, the new guidance appears to directly conflict with 35 USC § 101 that permits a patent for “[w]hoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof . . ..” Prior to the guidance, it could be interpreted that merely reading an AI output of the invention would constitute discovery of the invention and therefore be patentable under 35 USC § 101.

The new guidance appears to dismiss the possibility of a person discovering an AI’s invention (and thereby being able to patent the invention), which creates a gray area regarding whether certain inventions made by AI can be patented at all. For example, if a person uses a general AI program to invent but does not significantly contribute to the invention, the invention cannot be patented. This appears to contradict the USPTO’s acceptance of an invention being patentable by multiple independent people. Accordingly, if a person makes an invention with an AI (where the person does not significantly contribute), can that person still “invent,” or conceive, that invention without the AI afterward? In other words, can the invention be invented in hindsight? The guidance does not appear to contradict this scenario.

Since the guidance, laws, and precedence on AI are new and lack certain details, it is difficult to determine exactly how the USPTO and courts will enforce the limits on AI-assisted inventorship. However, there are a few steps that can be taken at this time to (attempt to) avoid issues with AI-assisted inventorship. First, each invention should be independently verified by a person. Verification may include reinventing the invention produced by the AI, which may be done by the person who prompted the AI or an independent person. Second, every patent application filed should have a human inventor tied to it. This may be the person who used the AI to discover the invention or the independent person who verified the invention. Third, inventors must be sure not to use public AI programs like ChatGPT that do not keep data private. The prompts and answers provided to and from these programs are considered public disclosures and would disqualify any inventions produced therefrom.

How the Patent System May Develop With AI

The USPTO and patent laws, in general, will need to change as AI further develops. Particularly, AI presents many challenges to the patent system in view of the current rules and new USPTO guidance. For example, AI cannot be listed as an inventor but can create prior art. As public AI programs like ChatGPT do not keep data private, prompts and generated answers are considered public disclosures (i.e., prior art). AI has the potential to create inventions at a much faster rate and on a larger scale than human inventors and, therefore, presents a threat of boxing human inventors out of the patent system. Another example includes the potential for transforming the standard of skill in the art. AI allows one of skill in the art to expand expertise and depth of knowledge, so it is conceivable that one with a bachelor’s degree with AI assistance may be performing at a level of someone with a Master’s degree or Ph.D. These and additional examples show that AI has the potential to drastically affect the patent system, where the laws and the USPTO will need to adapt accordingly.

Related Services:

Intellectual Property

About the Author:

Alex W. Holtshouser is an Associate in Dickinson Wright’s Troy office, where he focuses his practice on mechanical and electro-mechanical devices, including automotive, biomedical, and aviation technologies. He can be reached at 248-433-7574 or, and his biography can be accessed here.


The post Patenting with Artificial Inventors appeared first on IP Blog.