Connecting decision makers to a dynamic network of information, people and ideas, Bloomberg quickly and accurately delivers business and financial information, news and insight around the world.
Connecting decision makers to a dynamic network of information, people and ideas, Bloomberg quickly and accurately delivers business and financial information, news and insight around the world. English To Thai Translation Service
The US Patent and Trademark Office has rolled out artificial intelligence tools to aid patent examiners as they assess the stream of new submissions from inventors seeking protections for their creations.
The patent office’s Artificial Intelligence and Emerging Technologies Partnership met Wednesday and saw the AI-based features that examiners are actively using for prior art searches during patent application reviews, a critical step in the patent evaluation process. Stakeholders from Google, universities, and law firms presented at the fourth installment of the partnership series aimed at collaborating on the intersection of IP and AI.
The PTO began innovation to incorporate AI into the patent evaluation process as recently as 2020. Since then, they’ve introduced new tools—equipped with bureaucratic monikers—twice in the last three years to their legion of examiners. Their effort is to use new technology to speed up and improve the quality of their application examinations.
“The problem is that examiners have to go through an enormous amount of knowledge in a very short amount of time,” Jonathan Horner, a supervisory patent information technology specialist at the PTO, said at the meetingin Alexandria, Va. Examiners are charged with reviewing decades of technological history when reviewing applications, he said.
Mostly engineers and scientists, these government officials receive more than 200 hours of training within their first year on the job, Nicholas Jensen, PTO academy supervisory patent examiner trainer, said during a speech at the forum.
The AI-backed tools now provide another way to ensure examiners are able to efficiently compare the invention at hand to a comprehensive database of existing patents. These new features allow “examiners to review the same number of documents as they normally would, but not take away those chunks of documents that could be potentially relevant art just because they may have used the wrong keyword or missed a synonym” in their search, Horner said.
The office’s use of AI raised at least one question at a conference earlier this month about whether patents reviewed with AI tools would be as likely to be challenged at the Patent Trial and Appeal Board. “Do you think that would influence the PTAB at all?” IPWatchdog founder Gene Quinn asked PTO Director Kathi Vidal at an event attended by more than 200 attorneys.
Vidal didn’t answer the question, but said the PTO is “thinking about all those kinds of things.”
The PTO’s technology center began briefing the Patent Public Advisory Committee subcommittee on AI prototypes in 2020.
After making a prototype of an AI-powered patent search tool available to 600 examiners, the PTO took steps to release the “More Like This Document” feature to the full examiners corps in fiscal 2021, according to PPAC’s 2021 annual report. The feature “uses AI to retrieve similar documents based on examiner selection” and works with both foreign and US patent documents that could include potentially relevant prior art.
In September 2022, the PTO deployed the new AI-based “Similarity Search” feature in its Patents End-to-End search suite to assist examiners conducting searches. The tool “receives examiner-selected application information, including the specification, as input and uses trained AI models to output a list of domestic and foreign patent documents that are similar to the patent application being searched,” according to PPAC’s 2022 annual report.
PPAC has said these AI initiatives “directly impact patent quality and the efficiency of the USPTO” and are critical to ensure that it issues quality patents that can stand up to scrutiny when challenged.”
As of November, 8,500 examiners had been trained on the next-generation search tool, according to the committee. Examiners have conducted more than 1.3 million searches using AI search tools, Vidal’s said in a blog post published Friday. The agency hasn’t made available data detailing the performance of the AI tools since their adoption.
The patent office also has an auto-classification tool that utilizes AI to ensure applications are assigned to “the examiner best positioned to examine the application over the art.” Examiners work in different technology areas, so classification helps get applications in the hands of examiners with relevant skills and backgrounds.
The number of applications involving AI continues to grow, with patents containing AI making up more than 50% of technologies examined in 2020, according to Vidal’s post. The patent office plans to partner with Carnegie Mellon University to roll out a 21-course curriculum on AI tailored to the needs of patent examiners, Vidal said.
The office is now considering making its AI search tools publicly available, according to Vidal’s blog post.
Horner emphasized that applicants and the public should trust examiners’ use of AI-generated tools, calling them an “assistive tool to the examiner” and saying the examiner retains control over deciding whether a piece of prior art is relevant to patentability. “We are in no way looking for AI to perform a patent search,” he said.
The patent office works with industry partners—including some who were at this week’s event—to ensure that its models are sufficient to train the AI tools for adoption, Horner said. The patent office accounts for potential bias in its tools by not submitting applicants names’ into its models, he said, adding “that’s just a very basic example but we do much more beyond that.”
The tools also work regardless of the publication language, including Chinese, Korean, Japanese, German, French, and English. Horner said this has led to an uptake in the amount of foreign prior art used in rejections and in the ability to show potential for allowance.
The input data for these tools also include unpublished applications, Horner said, which is a “big deal” to examiners because they previously had to do everything relying on published patent applications. This allows examiners to compare active applications to other pending efforts. “Being able to leverage an unpublished application and use machine learning and AI on it really helps the examiner get a leg up in their search,” he said.
As the tools get smarter, petitioners will be challenged to come up with prior technology that hasn’t been in front of an examiner, Venable LLP partner Elizabeth Manno said at the event. Uncited prior art often supports a challenge to a patent’s validity at the PTAB, so less of it could make petitions more difficult.
To contact the reporter on this story: Annelise Gilbert at agilbert1@bloombergindustry.com
To contact the editors responsible for this story: James Arkin at jarkin@bloombergindustry.com; Kartikay Mehrotra at kmehrotra@bloombergindustry.com
AI-powered legal analytics, workflow tools and premium legal & business news.
Vietnamese To English Translation Service Log in to keep reading or access research tools.