The US Patent and Trademark Office has put an AI tool called Compass to work checking the formal requirements of patent applications involving engineered or modified DNA sequences, acting commissioner for patents Barry Schindler said at the Google Public Sector Summit on 20 October, Nextgov reported. Schindler said agents could help shorten a patent review process that can take five years to about one year by freeing examiners to speak with applicants.
Key points
- Compass checks formal requirements for submissions involving engineered or modified DNA sequences.
- Schindler cited a backlog of 700,000 applications and a workforce of around 9,000 patent examiners.
- He wants agents to help newer examiners draw on experienced colleagues’ knowledge, while keeping staff engaged in the underlying work.
- The office also has an internal chatbot, Scout, and is seeking a new chief AI officer.
Compass takes on DNA sequence checks
Schindler described a three-stage patent process comprising pre-examination, examination and post-examination. For an application covering a man-made or altered DNA sequence, an examiner must first check the submitted sequence before considering whether the invention qualifies for a patent. Compass performs what Schindler called a formality check on those submissions. The task had often involved repeated written exchanges between the office and the applicant.
That allocation of work matters to Schindler because some of the examiners handling these applications are PhD biochemists. His case for Compass is that they can spend less time on the preliminary check and more on the scientific and legal assessment. He also wants them to have more time to speak directly with applicants. Schindler described examination as a negotiation in which an exchange by phone could replace some of the correspondence that occupies examiners’ time.
Schindler tied the potential reduction in review time to those conversations, rather than to an agent deciding whether an invention deserves protection. He said examiners welcome the prospect of calling applicants but struggle to fit calls around their other work. His estimate of a process lasting about one year depends on agents taking on enough of that work for conversations to happen. It is a proposed gain in review time, not a time already achieved by Compass.
Schindler’s 700,000-application backlog
The office has a backlog of 700,000 applications and around 9,000 patent examiners, Schindler said. New examiners begin receiving cases almost immediately, he said, because of that workload. Applicants can therefore encounter markedly different levels of experience: Schindler contrasted an examiner with 20 years on the job with one who has two. He wants agents to help pass knowledge from experienced examiners to newer colleagues and make reviews more consistent.
His account joins two uses of agents that place different demands on the people using them. Compass takes a defined preliminary task off an examiner’s desk. Passing on experienced examiners’ knowledge would also affect how less experienced staff approach their cases. Schindler said even small uses of agents could be worthwhile and urged the office to begin with the problem it wanted to solve, rather than apply a single approach across its work.
USPTO weighs examiner skills alongside Scout
Schindler also warned against letting assistance become a substitute for learning. He referred to research suggesting that heavy reliance on AI can impair cognitive skills. An examiner who uses agents without understanding and continuing to learn the fundamentals “can get dumb”, he said, according to Nextgov. Schindler framed the design problem as helping staff without allowing the agent to take over their work.
Compass is part of a wider set of internal AI tools. USPTO employees use an enterprise chatbot called Scout for research, analysis, drafting and summarising, ExecutiveGov reported. Those activities differ from the formality check Schindler described for DNA sequence applications, but both put AI into tasks performed by office staff rather than into a service offered directly to patent applicants.
The office has also opened a search for a chief AI officer to lead its adoption of AI and machine learning across the agency. Jerry Ma, its first chief AI officer, has stepped down.
