Patent Cl(AI)ms: Using AI in Claim Interpretation

On May 28, 2024, the Eleventh Circuit Court of Appeals issued its decision in Snell v. United Specialty Ins. Co., No. 22-12581, 2024 WL 2717700, at *11 (11th Cir. May 28, 2024). On its face, it is a relatively run-of-the-mill decision concerning an insurer’s duty to defend and contract interpretation, but in the concurring opinion, Judge Newsom mused on “whether and how AI-powered large language models [“LLMs”] like OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude might—might—inform” the interpretation of legal texts. Id. at *11.

Of course, Judge Newsom is not alone in his musings. Jurists and legal professionals alike have been exploring the utility of LLMs in all areas of the law. If LLMS can be used in interpretation of contract terms, it may be possible to use LLMs in patent claim construction and interpretation. As with contract terms, patent claims are often interpreted using their plain and ordinary meaning unless the specification says otherwise. That said, there are two issues unique to patent claim construction that raises unique challenges.

First, claim terms are given their plain and ordinary meaning “to a person of ordinary skill in the art [“POSITA”] in question at the time of the invention”. Phillips v. AWH Corp, 415 F.3d 1303, 1313 (Fed. Cir. 2005). While temporal considerations are not unique to claim construction (contract terms are also generally given the meaning as of the effective date of the agreement), the application of a POSITA standard is unique. POSITA may vary greatly depending on the subject matter of the patent because a POSITA is often defined by art specific experiences, educations, and qualifications. To the extent LLMs are trained on data aggregated from the general public, it appears unlikely that LLMs would have the ability or capacity to inform through the lens of a POSITA.

Second, by nature, claim terms, and the words and phrases incorporated therein, are often new and novel. LLMs, by nature, are trained on existing words and phrases. While new words and phrases may be interpreted through the lens of existing words and phrases, LLMs’ “understanding” of new words and phrases could easily deviate from the intended meaning of language set forth in a claim, particularly when applied in the context of new technologies.

Despite these issues, LLMs may have utility in claim construction. As the Federal Circuit wrote, “[i]n some cases, the ordinary meaning of claim language as understood by a person of skill in the art may be readily apparent even to lay judges, and claim construction in such cases involves little more than the application of widely accepted meaning of commonly understood words … In such circumstances general purpose dictionaries may be helpful.” Phillips 415 F.3d at 1314 (Fed. Cir. 2005). If general purpose dictionaries can be helpful, perhaps LLMs can be helpful as well. As with the application of LLMs in all areas today, law and otherwise, it may be worth exploring LLMs as an additional tool to traditional methods of determining the plain and ordinary meaning of a patent claim.