As chief information officer, Jamie Holcombe leads USPTO’s AI processes that benefit inventors and patent examiners
Jamie Holcombe was a 25-year-old serving in the U.S. Army when he was asked to lead a communication platoon of American soldiers. They were male and female. They ranged in age from 19 to 43. They came from diverse backgrounds in culture and language.
George Washington also knew the daunting challenges of young leadership. When he was a 25-year-old, he was colonel of the Virginia militia.
Although Holcombe would never compare his duties to Washington’s, he strived to lead in the same way as America’s first president.
“George Washington’s leadership style suited me: Collaborate and survey everyone’s input before deciding clearly with boundaries—not micromanaging!–and executing ferociously,” he says.
Holcombe, chief information officer at the USPTO, is the principal advisor to the agency on the design, development, and management of its information systems and technology. “We operate and maintain the information infrastructure that manages inventions and ideas.”
Since coming to the USPTO from the private sector in 2019, one of his major accomplishments has been overseeing AI processes that benefit inventors and patent examiners. These have dramatically improved efficiencies while lowering costs.
When searching for prior art, inventors now have a public tool beyond the standard search engine. In late 2019, the enriched citation application programming interface (API) debuted to provide patent offices and the public with greater insight into the patent evaluation process, using data extraction.
The enriched citation API gives quick access to an organized and prioritized citation list that helps users identify relevant prior art. This ultimately improves the accuracy and consistency of the patent examination.
Empowered by state-of-the-art machine learning, AI, and natural language processing (NLP) algorithms, the enriched citation API analyzes the structure and content of the approximately 2,000 application responses (office actions) received daily. Patent applications are assigned classification codes from a list of over 250,000 terms, including the subject matter of the invention.
The API uses sophisticated information extraction and entity extraction algorithms to accurately locate:
- Statutes used by examiners,
- Claims rejected based on prior art,
- Particular prior art references cited, and
- Specific relevant sections in the cited prior art references used.
The algorithm routes the application to the correct examiner, then assigns the relevant codes for the purpose of assigning work to each.
“Previously, our staff had to craft enriched citations manually, and needed training on proper citation formats,” Holcombe said. “Such a process could not ‘scale’ properly to keep up with the volume of data. Think about that old ‘Lucy and the Chocolate Factory’ episode.”
Another new innovation leverages machine learning, or ML—another subset of AI—to deliver an auto-classification tool for classifying patent documents using the Cooperative Patent Classification system. This potentially reduced classification costs to the USPTO by more than 80 percent.
New AI/ML algorithms are “trained” with usage to classify patent and non-patent documents with CPC symbols and C* symbols in hours, at one-tenth the cost. The service also incorporates user feedback to verify and validate the accuracy of results.
This auto-classification tool, called AutoClass, results in a smarter routing system that has helped the USPTO realize savings of $1 million-$2 million per year via reduced spending on outside contractors.
Holcombe says the USPTO has seven emerging technology initiatives underway that can drive new operating efficiencies. “Our top goals now include giant stabilization and modernization initiatives to support the growth of our business as we provide technology that empowers our employees and stakeholders,” he says.
In assessing the results of his department’s AI/ML efforts, Holcombe lists four takeaways linked to inventors, employees, and the agency as a whole:
- Start with a use case tied to ROI for the business. Act now and be bold!
- Ensure the results you are providing are useful and constructive (to users, not you).
- Remember, AI/ML supplements human intelligence. It does not replace it!
- Confirm with experts that results are reasonable before using a feedback loop.
Holcombe, by the way, was awarded the Douglas MacArthur Leadership Award and Most Outstanding Officer of the Year before leaving the U.S. Army. His leadership accomplishments at the USPTO have just begun.