George Mason University Antonin Scalia Law School

Professor Tabrez Ebrahim on Artificial Intelligence Inventions

The following post comes from Associate Professor of Law Tabrez Ebrahim of California Western School of Law in San Diego, California.

a pair of glasses, an apple, and a stack of booksBy Tabrez Ebrahim

Artificial intelligence (AI) is a major concern to the United States Patent and Trademark Office (USPTO), for patent theory and policy, and for society. The USPTO requested comments from stakeholders about AI and released a report titled “Public Views on Artificial Intelligence and Intellectual Property Policy.” Patent law scholars have written about AI’s impact on inventorship and non-obviousness, and they have acknowledged that the patent system is vital for the development and use of AI. However, there is prevailing gap and understudied phenomenon of AI on patent disclosure. The Center for Protection for Intellectual Property (CPIP) supported my research in this vein through the Thomas Edison Innovation Fellowship.

In my new paper, Artificial Intelligence Inventions & Patent Disclosure, I claim that AI fundamentally challenges disclosure in patent law, which has not kept up with rapid advancements in AI, and I seek to invigorate the goals that patent law’s disclosure function is thought to serve for society. In so doing, I assess the role that AI plays in the inventive process, how AI can produce AI-generated output (that can be claimed in a patent application), and why it should matter for patent policy and for society. I introduce a taxonomy comprising AI-based tools and AI-generated output that I map with social-policy-related considerations, theoretical justifications and normative reasoning concerning disclosure for the use of AI in the inventive process, and proposals for enhancing disclosure and the impact on patent protection and trade secrecy.

AI refers to mathematical and statistical inference techniques that identify correlations within datasets to imitate decision making. An AI-based invention can be either: (1) an invention that is produced by AI; (2) an invention that applies AI to other fields; (3) an invention that embodies an advancement in the field of AI; or (4) some combination of the aforementioned. I focus on the first of these concerning the use of AI (what I term an “AI-based tool”) to produce output to be claimed as an invention in a patent application (what I term “AI-generated output”).

The use of AI in patent applications presents capabilities that were not envisioned for the U.S. patent system and allows for inventions based on AI-generated output that appear as if they were invented by a human. Inventors may not disclose the use of AI to the USPTO, but even if they were to do so, the lack of transparency and difficulty in replication with the use of AI presents challenges to the U.S. patent system and for the USPTO.

As a result of the use of AI-based tools in the inventive process, inventions may be fictitious or imaginary, but appear as if they had been created by humans (such as in the physical world) and still meet the enablement and written descriptions requirements. These inventions may be considered as being either imaginary, never-achieved, or unworkable to the inventor, but may appear as if they were created, tested, or made workable to reasonable onlookers or to patent examiners.

The current standard for disclosure in patent law is the same for an invention produced by the use of AI as any invention generated by a human being without the use of AI. However, the use of AI in the inventive process should necessitate a reevaluation of patent law’s disclosure function because: (1) AI can produce a volume of such fictitious or imaginary patent applications (that meet enablement and written descriptions requirements) that would stress the USPTO and the patent system; and (2) advanced AI in the form of deep learning, which is not well understood (due to hidden layers with weights that evolve) may insufficiently describe the making and using of the invention (even with disclosure of diagrams showing a representation of the utilized AI).

Such AI capabilities challenge the current purposes of patent law, and require assessing and answering the following questions for societal reasons: Should patent law embrace the unreal fictitious and imaginary AI-generated output, and if so how can the unreal be detected in patent examination from disclosure of that created by a human? Should inventors be required to disclose the use of AI in the inventive process, and should it matter for society?

Patents are conditioned on inventors describing their inventions, and patent law’s enablement doctrine focuses on the particular result of the invention process. While patent doctrine focuses on the end state and not the tool used in the process of inventing, in contrast, I argue that the use of AI in inventing profoundly and fundamentally challenges disclosure theory in patent law.

AI transforms inventing for two reasons that address the aforementioned reasons for reevaluation of patent law’s disclosure function: (1) The use of an AI-based tool in the invention process can make it appear as if the AI-generated output was produced by a human, when in fact, it was not so; and (2) even if an inventor disclosed the use of an AI-based tool, others may not be able to make or use the invention since the AI-based tool’s operation may not be transparent and replicable. These complexities require enhancing the disclosure requirement, and in so doing, present patent and trade secret considerations for society and for inventors.

The USPTO cannot reasonably expect patent examiners to confirm whether the patent application is for an invention that is fictitious or unexplainable in an era of increasing use of AI-based tools in the inventive process, and heightened disclosure provides a better verification mechanism for society. I argue that enhanced patent disclosure for AI has an important role to play in equilibrating an appropriate level of quid pro quo.

While there are trade-offs to explaining how the applied AI-based tools develop AI-generated output, I argue for: (1) a range of incentive options for enhanced AI patent disclosure, and (2) establishing a data deposit requirement as an alternative disclosure. My article’s theoretical contributions define a framework for subsequent empirical verification of whether an inventor will opt for trade secrecy or patent protection when there is use of AI-based tools in the inventive process, and if so, for which aspects of the invention.

There are a plethora of issues that the patent system and the USPTO should consider as inventors continue to use AI, and consideration should be given to disclosure as AI technology develops and is used even more in the inventive process.