Director's Forum: A Blog from USPTO's Leadership

« Spotlight on Commerc... | Blog homepage | USPTO celebrates... »
Thursday Mar 18, 2021

Artificial intelligence tools at the USPTO

Blog by Drew Hirshfeld, performing the functions and duties of the Under Secretary of Commerce for Intellectual Property and Director of the USPTO

Artificial intelligence image

Among the most important technological developments has been the advent of artificial intelligence (AI), a transformative technology that promises tremendous societal and economic benefits. USPTO operations can be one of its beneficiaries. The integration of AI technologies into next generation tools offers an exciting opportunity to enhance the quality and efficiency of patent and trademark examination.

To incorporate AI into our examination tools and processes at the USPTO, we’ve undertaken a comprehensive development strategy including extensive market research and rigorous testing of a wide range of proof-of-concepts to identify the best solutions. Our objective is not just to deploy smarter technology, but to build a smarter organization by coupling the strengths of our workforce with the strengths of AI. This is the beginning of a whole new trajectory in how we leverage technology to transform patent and trademark operations for the better.

We are incorporating AI tools into two critical areas of patent examination: search and classification.

Performing a complete prior art search is a critically important component of the patent examination process and the USPTO’s mission to issue reliable patent rights. However, the exponential growth of prior art and tremendous pace of technological innovation make it increasingly more difficult to quickly discover the most relevant prior art. To meet this challenge, we have developed an AI-based prototype search system that helps to identify relevant documents and provides suggestions for additional areas to search. In addition to providing world-class patent AI models, the system is designed to learn from the world’s greatest patent searchers, our USPTO examiners. The system is configured to automatically capture feedback data from our examiners to yield additional enhancements over time. We are also developing features to help examiners interpret results generated by the AI models to provide transparency into the system. A beta version of this new AI tool was released to a subset of examiners in March 2020. Assessments conducted to date yielded promising results, and steps are being taken to incorporate AI into our next generation search tool for examiners.

We also developed an auto-classification tool that leverages machine learning to classify patent documents using the Cooperative Patent Classification (CPC) system. The system can suggest CPC symbols, and includes the ability to identify claimed subject matter for additional refinement of the suggested CPC symbols similar to our AI search system. The auto-classification system also includes indicators that provide users with insight into the reasoning of the AI, by linking suggested CPC symbols to specific portions of the document. Enhanced feedback mechanisms designed into the system integrate with our existing classification processes to support training the AI. Based on an analysis of system performance, the USPTO implemented auto-classification in December 2020 to automatically identify claimed subject matter with CPC for internal operations. As a result, the agency is realizing reductions in procurement expenditures for acquiring CPC data. Additionally, we are continuing to develop further capabilities to support a broader range of patent classification requirements at the USPTO.

These successes are demonstrating the value of applying AI to improve the agency operations and strengthen the IP system. To continue building from these successes, our Patents team has expanded their investigations to explore potential new opportunities to leverage AI. For example, research is now underway on AI-based image search capabilities which could open up whole new ways to retrieve prior art. This could be particularly useful for searching patent applications where examiners rely heavily on images for making patentability determinations, such as design patent applications.

On the Trademarks side, we recently completed market research in AI capabilities for image comparison and for checking the acceptability of identification of goods and services against the entries in the Trademarks ID Manual. The USPTO team developed AI prototypes to compare trademark images, to suggest the correct assignment of mark image design codes, and to determine the potential acceptability of the identifications of goods and services. A beta test of these prototypes through a common user interface with approximately 10 stakeholders began in November 2020 and continues, with a larger beta possible later this year. In addition, the USPTO has tested solutions for false specimen detection capabilities using a software program, which was integrated on December 1, 2020 into the agency’s efforts to identify digitally manipulated specimens of use or mock-ups of web pages. Finally, a prototype of an AI based chatbot for answering frequently asked questions via the USPTO website could be ready for beta testing later this year.

Overall we have achieved some remarkable milestones and made great strides toward integrating AI into the USPTO’s day-to-day functions. Stay tuned for more exciting updates from the USPTO on AI in the near future.

Comments:

Are you considering allowing registered agents and attorneys to access the search aspect of that proposed system?

Posted by Bruce A. Lev on March 18, 2021 at 02:16 PM EDT #

It would be nice to be able to use the same search tools as the Examiners do, once the Public Search Facility re-opens. Maybe the search system would be available remotely, even as a paid subscription service like the EPO offers. Being able to recreate the Examiner's search to see what the Examiner pulled up in certain search strings would be of great value. One question though. If one relies on AI to a large extent, how do you keep it from learning the wrong things or "unlearn" something? Text searching is very context specific and also depends at times on the quality of English from translations of applications of foreign origin. Just because something has the "right words" does not mean it is the "right reference." And sometimes, the "right words" aren't used at all by Applicant.

Posted by Fritz M. Fleming on March 19, 2021 at 09:05 AM EDT #

It would be helpful if machine learning is also incorporated into USPTO’s publicly-accessible advanced patent search engine, rather than being just an exclusive feature of the patent search engine used by the examiners. Also, the USPTO patent full-text and image database interface hasn’t seen a makeover for decades now. It is long overdue for an update to at least give it a much more modern look that’s more in-line with the rest of the USPTO’s website. As it stands, it looks like a remnant from the past that seems to be a tacit acknowledgement by the USPTO that most people probably don’t use it as much as they use other patent search databases like Google Patents and thus not worth updating. Still, having another publicly accessible database that can be used for patent search verification purposes (e.g., to verify Google Patents search results) would be very useful to many people, especially to those who can’t afford to pay for subscription-only patent search databases. Regarding the USPTO’s upgraded auto-classification system’s including “indicators that provide users with insight into the reasoning of the AI, by linking suggested CPC symbols to specific portions of the document,” is USPTO suggesting that it’s using machine learning combined with some kind of machine reasoning, neural logic reasoning, ontology reasoning, logic based symbolic, diagrammatic, or graphical reasoning? It’s unclear because USPTO merely mentioned providing “insight into the reasoning,” rather than providing some kind of reasoning per se. In any case, this would be an extremely useful feature that could significantly improve the accuracy, precision, and consistency of CPC code assignments across all fields of invention. If one uses Google Patents (which has benefited significantly from being owned by arguably the top AI company in the world) as a benchmark, then machine learning as a whole still has ways to go, at least when it comes to doing patent searches. Google Patent’s reliability and consistency can sometimes be easily tripped, for no obvious reason, when one makes even very minor changes in a previous keyword string. For example, adding even just one generic term to a set of multiple specific keywords can sometimes cause Google Patents to churn out significantly more irrelevant results than if the generic term were excluded. But it’s clear that adding machine learning capability into USPTO’s patent search examination is just one of the many necessary steps USPTO needs to take to stay abreast of the times.

Posted by Vincent Violago on March 19, 2021 at 10:56 AM EDT #

It’s clear that adding machine learning capability into USPTO’s patent search examination is just one of the many necessary steps the USPTO needs to take to stay abreast of the times. It would be helpful if machine learning is also incorporated into USPTO’s publicly-accessible advanced patent search engine, rather than being just an exclusive feature of the patent search engine used by the examiners. Also, the USPTO patent full-text and image database interface hasn’t seen a makeover for decades now. It is long overdue for an update to at least give it a much more modern look that’s more in-line with the rest of the USPTO’s website. Having another publicly-accessible database that can be used for patent search verification purposes (e.g., to verify Google Patents search results) would be very useful to many people, especially to those who can’t afford to pay for subscription-only patent search databases.

Posted by Vincent Violago on March 19, 2021 at 11:04 AM EDT #

Regarding the USPTO’s upgraded auto-classification system’s including “indicators that provide users with insight into the reasoning of the AI, by linking suggested CPC symbols to specific portions of the document,” is USPTO suggesting that it’s using machine learning combined with some kind of machine reasoning, neural logic reasoning, ontology reasoning, logic based symbolic, diagrammatic, or graphical reasoning? It’s unclear because USPTO merely mentioned providing “insight into the reasoning,” rather than providing some kind of reasoning per se. In any case, this would be an extremely useful feature that could significantly improve the accuracy, precision, and consistency of CPC code assignments across all fields of invention.

Posted by Vincent Violago on March 19, 2021 at 11:04 AM EDT #

Excellent initiative to use AI for patents search and classification etc. As business owner i can understand lot of people use deceptive practices to steal your identity your brand name .. finding smart ways to protect and promote brand creation is great. I hope AE will make difference

Posted by Kashif Mahmood on March 20, 2021 at 07:37 PM EDT #

Will the public in every country have access to this technology in the future?

Posted by cara menanam bawang merah on March 21, 2021 at 01:55 AM EDT #

Post a Comment:
  • HTML Syntax: NOT allowed