January 22, 2025
The GIST Editors' notes
This text has been reviewed in line with Science X's editorial course of and insurance policies. Editors have highlighted the next attributes whereas making certain the content material's credibility:
fact-checked
peer-reviewed publication
trusted supply
proofread
AI-generated patents: Balancing innovation with disclosure accuracy

In a paper printed in Nature Biotechnology, three Stanford College-affiliated co-authors, together with Stanford Legislation College Professor Lisa Larrimore Ouellette, deal with a problem of rising significance on the intersection of regulation and expertise: How the usage of synthetic intelligence to codify and talk data could affect the usefulness of the technical literature for subsequent researchers and innovators. Specifically, they studied how the rising use of AI within the patent-drafting course of is exacerbating present challenges round patent regulation's disclosure necessities.
Patent regulation requires that inventors not solely describe their innovations, however accomplish that with sufficient readability that somebody expert within the related subject can replicate the work. This disclosure requirement helps disseminate data about new innovations, and is a vital a part of what society will get in return for granting inventors unique rights to their innovations.
Disclosure additionally ensures {that a} patent's scope is proscribed to what the inventor contributed, in line with the authors of "How Will AI Have an effect on Patent Disclosures?" Nevertheless, implementing the disclosure requirements has traditionally been inconsistent, they write. Patent examiners, working underneath tight time constraints and with generally restricted coaching, usually prioritize evaluating novelty and obviousness over assessing disclosure high quality.
"AI will amplify errors in each improperly granting poorly disclosed patents and improperly denying patents on the premise of earlier AI-generated references that don’t really disclose these innovations," write co-authors Lisa Ouellette, the Deane F. Johnson Professor of Legislation; Victoria Fang, JD '24, a former United States Patent and Trademark patent examiner who collaborated on the examine whereas a scholar at Stanford Legislation; and Nicholas T. Ouellette, a Stanford College professor of civil and environmental engineering.
However the rise of AI patent drafting additionally supplies a possibility to reform the patent examination course of and promote innovation, in line with the authors.
"The disruption to the patent drafting and evaluate course of created by AI provides us a possibility to enhance disclosure throughout the patent system," mentioned Lisa Ouellette, the writer of a number one patent regulation casebook and a big physique of scholarship that focuses on empirical and theoretical issues in mental property and innovation regulation.
"Rising enforcement of present disclosure guidelines and heightening necessities to restrict granting patents to those that have carried out actual ingenious work would higher equip patent workplaces to do their work and would enhance the patent system general."
The USPTO lately issued steerage requiring human evaluate of AI-drafted patent purposes, "however this is not going to be sufficient," say the paper's authors, who undertook a examine, described of their article, to judge whether or not present AI instruments are able to producing believable patent specs which might be unlikely to be rejected by a patent examiner for failure to satisfy the disclosure necessities.
The authors' suggestions embody elevating disclosure requirements to require inventors to implement a minimum of some model of their invention earlier than submitting a patent software (or meet stricter disclosure necessities for unproven applied sciences). In addition they name for improved coaching of patent examiners, together with offering them with examples of insufficient disclosures and emphasizing the significance of technical accuracy.
Placing AI instruments to the check
To discover the capabilities of AI in drafting patent disclosures, the researchers designed a examine evaluating two patent-specific AI programs (Edge and Vaero) and the general-purpose ChatGPT-4o. They examined the instruments with 4 applied sciences associated to fluid dynamics: two well-documented, present applied sciences doubtless included within the AI's coaching knowledge, and two hypothetical applied sciences.
For the prevailing applied sciences, the AI instruments produced patent specs that typically included adequate element for a talented researcher to duplicate the innovations. Nevertheless, the outputs additionally contained inaccuracies or deceptive statements alongside appropriate info. This means that whereas AI can help in describing established applied sciences, its outputs require cautious human oversight to make sure reliability.
When the researchers turned to hypothetical applied sciences, asking the AI platforms find out how to implement unproven ideas, not one of the instruments generated workable options or correct technical descriptions for these unsolved issues. However somewhat than acknowledging their shortcomings, the AI platforms created seemingly credible, however in the end incorrect or speculative, specs.
"Apparently, the AI-generated specs for each present and hypothetical applied sciences have been usually adequate to fulfill the present degree of scrutiny supplied by patent examiners even when they didn’t meet the scientific requirements," mentioned co-author Fang. "This factors to a mismatch between how patent regulation requirements are utilized and the sensible utility of patent disclosures."
"The outputs of AI instruments can look convincing at first look, however can even have some fairly critical flaws. With out cautious evaluate, they danger diluting the worth of patent disclosures as a supply of technical data," mentioned co-author Nicholas Ouellette. "And if examiners incorrectly grant patents that don't really clarify find out how to implement an invention, these patents stand as a authorized barrier to researchers who later determine find out how to make the invention work.
Towards a greater system
"The issue isn't with these AI patent-drafting instruments themselves," Lisa Ouellette mentioned. "In actual fact, they present monumental promise for decreasing the prices of patenting and inequalities in who has entry to the patent system."
As a substitute, the authors assume AI instruments reveal extra basic issues with patent regulation's disclosure requirements and the way these requirements are enforced. They suggest a number of methods to enhance disclosure high quality and mitigate the dangers of AI-generated patents, together with:
- Requiring inventors to implement a minimum of some model of their invention earlier than submitting a patent or to satisfy stricter disclosure necessities for unproven applied sciences;
- Offering patent examiners with higher coaching and assets to judge disclosures successfully;
- Bringing in exterior specialists to evaluate disclosures and improve the rigor of patent examinations, significantly for cutting-edge or extremely technical innovations;
- Utilizing AI itself to bettering patent examination by figuring out probably problematic disclosures for nearer scrutiny; and
- Extending the window for difficult a patent's disclosure past the present nine-month restrict to offer a security web for addressing points missed throughout preliminary examination.
"By refining disclosure requirements, reconsidering examination apply, and utilizing expertise properly, we are able to make sure that patents are granted just for what the inventor really invented and disclosed," Lisa Ouellette mentioned.
Extra info: Lisa Larrimore Ouellette et al, How will AI have an effect on patent disclosures?, Nature Biotechnology (2025). DOI: 10.1038/s41587-024-02515-6
Journal info: Nature Biotechnology Supplied by Stanford College Quotation: AI-generated patents: Balancing innovation with disclosure accuracy (2025, January 22) retrieved 22 January 2025 from https://techxplore.com/information/2025-01-ai-generated-patents-disclosure-accuracy.html This doc is topic to copyright. Other than any honest dealing for the aim of personal examine or analysis, no half could also be reproduced with out the written permission. The content material is supplied for info functions solely.
Discover additional
Thinker finds glitch in worldwide patent legal guidelines 0 shares
Feedback to editors