The U.S. Patent and Trademark Office (USPTO) issued new guidance on Feb. 12, 2024, regarding the use of artificial intelligence (AI) while developing new inventions. The U.S. Court of Appeals for the Federal Circuit previously held that an AI system cannot be named as an inventor on a patent, but its opinion left open the question of whether persons can be named as inventors when they use an AI system during the inventive process. President Joe Biden's Executive Order calling for guidance on this issue compelled the USPTO to act. Now, both examiners and stakeholders have some clarity regarding AI-assisted inventions.
New Policy, Old Standard
USPTO Director Kathi Vidal said in issuing the guidance, "[t]he right balance must be struck between awarding patent protection to promote human ingenuity and investment for AI-assisted inventions while not unnecessarily locking up innovation for future developments."
To that end, the guidance first provides that AI-assisted inventions are not categorically unpatentable. No sections of the Patent Act support the position that a person using a tool, including an AI tool, cannot be a named inventor. Additionally, while courts have said that conception is performed in the mind and the patent system is designed to encourage human ingenuity, neither of these principles preclude AI-assisted inventorship. See Pfaff v. Wells Elecs. Inc., 525 U.S. 55, 60 (1998); Graham v. John Deere Co., 383 U.S. 1, 9 (1966).
Second, the existing standard for determining inventorship of persons should be extended to inventorship in the emergent, AI-assisted context. Under the existing test, an inventor must "(1) contribute in some significant manner to the conception or reduction to practice of the invention, (2) make a contribution to the claimed invention that is not insignificant in quality, when that contribution is measured against the dimension of the full invention, and (3) do more than merely explain to the real inventors well-known concepts and/or the current state of the art[.]" (Pannu factors) See Pannu v. Iolab Corp., 155 F.3d 1344, 1351 (Fed. Cir. 1998). In simplified form, the existing standard for inventorship is that each person must make a significant contribution to the invention.
The key inquiry then for persons who use an AI system is whether their contribution was significant enough, not whether the contribution of the AI system would rise to the same level of inventorship if those contributions were made by a person. Extending this thinking to patent claims, a natural person must significantly contribute to each and every utility, design or plant patent claim before those claims should issue.
Guiding Principles for Applying the Standard
Applying the Pannu factors to determine whether a person's contribution is significant is a highly fact-dependent exercise. Nevertheless, the USPTO gives a few guiding principles to assist that process:
- A person's use of an AI system in creating an invention does not negate the person's contributions as an inventor.
- Because an idea does not rise to the level of conception, a person who only presents a problem to an AI system may not be a proper inventor of an invention identified from the output of the AI system. However, a significant contribution could be shown in the way the person constructs the prompt in view of a specific problem to elicit a particular solution from the AI system.
- Because reduction to practice alone is not sufficient to demonstrate inventorship, a natural person who merely recognizes and appreciates the output of an AI system is not necessarily an inventor. However, a significant contribution could be made to the output of an AI system to create an invention (e.g., a person who conducts a successful experiment using the AI system's output could demonstrate that they provided a significant contribution).
- A person who develops an essential building block from which the invention is derived may be considered to have provided a significant contribution. In some situations, a significant contribution can be made by the person who designs, builds or trains an AI system in view of a specific problem to elicit a particular solution such that they are an inventor.
- Maintaining "intellectual domination" over an AI system (e.g., owning or overseeing it) does not, on its own, make a person an inventor of any inventions created through the use of the AI system.
Illustrative Examples
Principles 1 and 5 above draw seemingly clear lines, while principles 2 through 4 are prone to interpretation. The USPTO helpfully provides reference points in two hypothetical examples where AI systems play different roles in the inventive process and discusses how the Pannu factors should be analyzed in those scenarios.
Example 1, titled Transaxle for Remote Control Car, poses scenarios in which engineers, Ruth and Morgan, are tasked with developing new remote control cars in time for the holiday rush, so they employ the help of a generative AI system called Puerto5. Ruth and Morgan provide a prompt stating, "Create an original design for a transaxle for a model car, including a schematic and description of the transaxle."
Scenario 1.1: Ruth and Morgan claim the exact transaxle output by Puerto5. They are not inventors as per principles 2 and 3.
Scenario 1.2: Mogan builds the transaxle output by Puerto5 using steel, a common material in the industry, and claims this construction. Morgan is not an inventor as per principle 3.
Scenario 1.3: Ruth and Morgan further prompt Puerto5 to provide an alternative design. They conduct experimentation on the output, discovering that a dimension needs to be elongated, certain components need to be in precise locations for the transaxle to be operable, and a new design for fasteners improves on conventional fasteners. They claim a transaxle with the modifications from their experimentation. Ruth and Morgan are inventors as per principles 1 and 3.
Scenario 1.4: Morgan prompts Puerto5 with the new design from Scenario 1.3 and asks for manufacturing suggestions. Puerto5 suggests that a component can be milled out of aluminum, and Morgan claims this invention in a dependent claim. Morgan and Ruth and inventors as per principle 1.
Scenario 1.5: Maverick is the AI engineer who designed Puerto5 and trained it on a diverse set of data using standard techniques. He was unaware of Puerto5's future application to remote control transaxles. Maverick is not an inventor as per principle 5.
Example 2, titled Developing a Therapeutic Compound for Treating Cancer, poses scenarios in which a professor, Marisa, is researching the development of a drug to treat prostate cancer. Wanting to identify lead drug compounds that selectively target a specific protein, Marisa consults with an AI expert, Raghu, and instructs him to use a deep neural network (DNN)-based prediction model called Drug Target Interaction Predictor (DTIP) to find viable candidates amongst a large dataset of compounds. Lauren, the lead data scientist, trained DTIP on diverse sets of compounds and targets from previous experiments. Marisa selects the output compounds that indicate potential for high efficacy.
Scenario 2.1: Marisa and Naz, a postdoctoral fellow, identify potential structural modifications to one of the selected compounds, CID_1, through experimentation. Naz prepares intermediates to CID_1 and finds that CID_1-int is more stable than the others. Marisa then finds that CID_1-mod, the compound synthesized from CID_1-int, exhibits increased efficacy. Claim 1 covers the method of identifying and synthesizing a lead compound comprising feeding a DNN model and modifying the output, and Claim 2 covers the structural formula of CID_1-mod. Marisa and Naz are inventors for both inventions, while Raghu and Lauren are inventors for neither as per principles 2, 3 and 5.
Scenario 2.2: Marisa hopes to find a compound with good binding affinity, as well as other properties. Raghu develops new a generative AI system, Molecular Optimizer (MO), for this specific purpose. Raghu and Marisa train MO in an iterative process to predict potential structural modifications to input compounds that optimize the desired properties. Once fine-tuning is complete, Raghu inputs the selected compounds from DTIP and MO outputs modified compounds, including MID_1. Marisa determines that MID_1 is the most viable candidate and claims the structural formula. Raghu and Marisa are inventors as per principles 1 and 4.
Call for Feedback
The USPTO wants to hear from practitioners and is accepting comments through the 90-day comment period. Feedback may supplement the USPTO's guidance in the future.
