Inside IP Firm Sterne Kessler’s Development of Its AI Patent Drafting Tool

In March, intellectual property firm Sterne, Kessler, Goldstein & Fox announced the launch of Patent Assist AI, a proprietary generative artificial intelligence tool designed to generate the description section of patent applications.

Sterne Kessler came to the decision to develop its own tool roughly one year ago, after conducting a blind test of commercially available patent drafting tools against outputs generated by publicly available large language models (LLMs). When the publicly available models outperformed some of the vendors under consideration, the firm began to consider building something internally.

“Our first inclination was, ‘OK, these [commercially available] tools aren’t necessarily outperforming just using public LLMs,’” said Daniel Block, a director in the firm’s electronics practice group. “Maybe this is something we can do.”

Sterne Kessler also wanted a tool that specifically focused on the description section of the
patent application, whereas most commercially available AI patent drafting tools generated complete patent applications based on the disclosures submitted by inventors.

Tools that generated complete patent applications raised certain risks, such as an AI system being credited as the inventor of a patent application, which could make the patent harder to assert in the future. They also would have deprived the firm of the chance to take advantage of the skills and expertise of its staff.

“The real value of a patent is in the claims,” said Michael Specht, director and chair of the
electronics practice group. “That’s the real high value input from the attorneys, the time keepers: coming up with the claims and the figures.”

“If we can do that and leverage our attorneys really well, and feed that input into the tool, that can save us time, it can make us more efficient,” he added.

Patent Assist AI fits into the firm’s traditional patent application workflow. Inventors submit
invention disclosure forms to the firm, and follow up with an invention disclosure call with one of the firm’s attorneys. The attorney then drafts the claims, which detail the exclusive rights the patent would establish, and produce diagrams.

Patent Assist AI subsequently uses the form submitted by the inventor as well as the claims
and diagrams drawn up by the attorney to generate the detailed description. The resulting output reflects the underlying work done by the attorney, and can be edited with their supervision.

“For us … what the tool has to generate is something that’s easily editable into the final
form,” Block said. “What we’re trying to do is fit this into the pipeline of what our attorneys do on a day-to-day basis.”

How It Was Built

The tool is built on private instances of publicly available LLMs, and is hosted on the firm’s
servers, with elements on AWS servers. The tool was trained on detailed hypothetical patent applications created by the firm to avoid the use of client data, and is designed so that client data can not be used to train the underlying models or influence subsequent uses.

Block and Specht told Legaltech News that Sterne Kessler’s focus on a specific practice area allowed it to develop a tool tailored to a discrete use case relevant across the firm. Its IP expertise, and the skill sets required for that work, also made it easier to develop internally, achieve buyin and speed adoption.

“There was a lot of intellectual curiosity,” Specht said. “We hire people that are very technical, and this was a way to excite them and get people involved with this, as opposed to fearing it.”

“We had to have a number of discussions about all these things internally,” Block added. “The fact that … our managing partner has an electrical engineering background made those discussions a little bit easier. Our general counsel has an electrical engineering background, that made a lot of these discussions a lot easier.”

Patent Assist AI was developed by Sterne Kessler Labs, the firm’s internal R&D wing, which
is staffed by the firm’s existing workforce part-time and tasked with sourcing and developing tools across the patent pipeline.

Block, who leads the Labs’ AI efforts, mentioned prior art searching as a further area apt
for leveraging AI, while Specht noted that the firm also hopes to use AI tools in the investigatory phase of its pipeline.