Artificial Intelligence Will Help to Solve the USPTO’s Patent Quality Problem

“Even if artificial intelligence is only a part of a larger solution, we must arm the gatekeepers of patent rights with better tools so they can better carry out the goals of the patent system.” a month ago, Steve Brachmann authored an article concerned with a brief given to Capitol Hill staff by Professors Frakes and Wasserman. The article highlighted fundamental, as well as practical, problems with Professors Frakes’ and Wasserman’s proposal (i.e. doubling the number of patent examiners as a means to reduce the number of invalid patents and thereby prevent societal harms) and how it could be detrimental to the U.S. patent system.

Frakes and Wasserman cite sources saying that the U.S. Patent and Trademark Office (USPTO) is granting too many bad patents which “unnecessarily drain consumer welfare, stunt productive research, and unreasonably extract rents from innovators.”  They trace the problem to a lack of sufficient review time available to the examiner:

“On average, a U.S. patent examiner spends only eighteen hours reviewing an application, which includes reading the application, searching for prior art, comparing the prior art with the application, writing a rejection, responding to the patent applicant’s arguments, and often conducting an interview with the applicant’s attorney. If examiners are not given enough time to evaluate applications, they may not be able to reject applications by identifying and articulating justifications with appropriate underlying legal validity.”

And since patent applications are presumed valid, the examiner must grant the patent if he or she cannot rebut the presumption within the time allotted for the examination.

According to Frakes and Wasserman, doubling the number of examiners would result in a net reduction of the costs associated with bad patents (such as future litigation expense savings per year of approximately $572 million dollars).

The IPWatchdog article points to several issues with Frakes’ and Wasserman’s proposal, but does not discuss other approaches or options, such as using artificial intelligence tools to improve the patent application review process—an option that USPTO Commissioner for Patents Drew Hirshfeld said in a recent Senate IP Subcommittee hearing that the Office is actively pursuing.


Examiners Need All Available Tools

While some, like Frakes and Wasserman, suggest that the path to better patents is through doing more of the same, i.e. doubling the number of examiners, there might be a better way.

According to PWC, 72% of executives testify that AI improves internal operations while freeing up workers to perform more creative and meaningful tasks. In fact, while some might fear that “robots” will take human jobs, technological innovation has been proven to generate more jobs than it takes, while automating tasks, like patent search.

AI can further the purposes of the patent system in two important ways. First, it can perform tedious prior art searches so that examiners put their limited time to higher and better uses. Second, AI tools promote a better and more effective use of inventors’ and other interested parties’ time and money in connection with patenting new technologies and maintaining patent portfolios.

For many years now, the USPTO has received hundreds of thousands of applications each year. That number has been over 600,000 patent applications each year for the last several years. Given the sheer volume of patent applications processed each year by the USPTO, and increases in relevant data sources, it is no wonder that examiners are crunched for time and applicants wait two years or more before receiving a final disposition on their application.

At this rate, simply increasing the number of patent examiners will not be enough, and technological solutions will become necessary.

Better, Faster Prior Art Searches

AI tools can streamline the application review process and the quality of the results by reducing the amount of time examiners spend researching prior art. Instead of two to four days of prior art research, artificial intelligence can do all that leg work in one to two hours. AI technology has advanced beyond simply comparing the words in a patent application to those in other patent records. AI can compare a whole patent’s content and ideas against millions of other patent records and database sources to find relevant prior art and return search results that meet or exceed those of human examiners.

Thus, even if examiners, on average, only spend 18 hours to search and review a patent application (as per Frakes and Wasserman), with AI tools they can perform the time-intensive prior art search much faster. Once that task is outsourced, the patent examiner will be able to allocate a greater portion of their review to understanding the applications, comparing the prior art to the applications, writing rejections, responding to applicants, etc.

Since the purpose of the Patent Act is to incentivize innovation for the short-term benefit of the inventor and the long-term benefit of society, surely tools that redirect inventors’ focus toward substantive matters are very valuable.

Weeding Out Unpatentable Inventions Up Front

Before a patent application even reaches the USPTO, artificial intelligence tools can save inventors time and money by identifying early on those inventions that will likely not be patentable. Instead of investing time and effort in such dead-end inventions, they can turn their attention to more promising work. And consider the benefit to the patent system and its goals if artificial intelligence not only makes the examiners more effective (whether we double their number or not), but also improves the overall quality of the initial applications.

Balancing Budgets

Similarly, instead of renewing all existing patents in a company’s portfolio, portfolio managers and practitioners can use AI to more effectively manage their intellectual property budgets. For example, AI can identify prior art which could provide a basis for an invalidation challenge, which in turn could impact the patent holder’s decision whether or not such patent is worth renewing.

Parsing Portfolios

In addition, other patent practitioners, such as M&A counsel and litigation counsel, can benefit from the power of AI search capabilities. Is that patent portfolio really as good as the seller advertises, or is it undermined by prior art and an easy target for an invalidation claim?  Is company X’s technology infringing on your patent? Artificial intelligence tools can save practitioners time and provide valuable insight into these issues.

Not a Silver Bullet, But a Powerful Tool

One challenge that must be addressed is that artificial intelligence can be “gamed”. AI is based on learning specific data patterns and making fast decisions based on those patterns. However, those patterns can be falsified or triggered by actors who wish to game the software. AI is not fool proof and there have been recorded cases where it failed to perform what it was designed to do.

Artificial intelligence, by itself, may not be a silver bullet to the issue of patent quality, but it is a powerful tool well-suited to promote the goals of our patent system by allocating examiners’ limited time to higher and better uses and facilitating a better and more cost-effective use of inventors’ and other interested parties time and money in connection with patenting new technologies and maintaining patent portfolios.

While it may be the case that we need to dramatically increase the number of examiners, it is my opinion that a combination of patent examiners with innovative tools that help them get their job done quickly and efficiently are the future of patent examination.

Image Source: Deposit Photos
Image ID: 6496641
Copyright: stuartmiles 


Warning & Disclaimer: The pages, articles and comments on do not constitute legal advice, nor do they create any attorney-client relationship. The articles published express the personal opinion and views of the author as of the time of publication and should not be attributed to the author’s employer, clients or the sponsors of

Join the Discussion

20 comments so far.

  • [Avatar for kk]
    December 8, 2019 01:59 am

    Examiners cannot rely on AI searches to improve quality because AI won’t know what to look for. Meaning, the search process an examiner does is iterative and adaptive based on how discovered references apply individually and together to specific claim limitations. Examiners cannot usefully read through tens or hundreds of AI-provided references even if relevant passages are highlighted by the AI because the overall relevance of a document sometimes is governed by the adjacent (or even distant) sentences, paragraphs and drawings to the words that caused the document to be flagged.

    In terms of quality how would the AI-assisted search be any better that what an examiner does now, aside from the AI being an error check so the examiner doesn’t miss dead-on references? These dead-on references though would already turn up in a basic search.

    For AI to improve quality it would have to search AND examine an application to produce useful results. There is no useful middle ground where quality is concerned.

    That said I do applaud any effort that goes to improving patent quality through AI.

  • [Avatar for Gerry J. Elman]
    Gerry J. Elman
    December 5, 2019 06:39 pm


    Hear, hear!

    Well, it’s about time. I’ve been ranting since 1907 that AI support for patent examiners is greatly needed. See when I reiterated the call in 2011.

  • [Avatar for AnEx]
    November 26, 2019 01:29 pm

    “my AI technology provides accurate results, it’s all about what you do with them.”

    I truly wish you the best in your efforts (as an examiner, I’d love better search tools), but this makes me wonder about your approach. “Accuracy” is arbitrary and easy. There are already a ton of search tools that accurately identify “similar” a pile of references according to some proprietary metric. The results need to be USEFUL.

  • [Avatar for Benny]
    November 25, 2019 10:53 am

    The USPTO definitely needs better tools to identify prior art. Over reliance on a single software solution won’t do it. It will take a combination of AI, diverse search tools, and examiner commitment before we will see patents which truly meet the novelty requirement in every case. Commentator B can fool around with the claims all he/she likes, the end result will probably a worthless patent which no one has any incentive to implement.

  • [Avatar for Anon]
    November 25, 2019 08:55 am

    Sister Anon,

    While your post is something that I would agree with, let’s not let the “Anon” moniker confuse readers.

  • [Avatar for Udi Cohen]
    Udi Cohen
    November 25, 2019 08:21 am

    Dear MaxDrei,

    Valid points, my AI technology provides accurate results, it’s all about what you do with them.
    I hope that examiners can benefit from my AI tool by reducing their search work and hopefully allow them to focus their best efforts on the analysis parts due to this time reduction.

  • [Avatar for peter brady]
    peter brady
    November 25, 2019 06:28 am

    Re” Artificial Intelligence Will Help to Solve

    Actually, “Any” kind of intelligence will help the USTPO’s Patent Quality Problem”

  • [Avatar for MaxDrei]
    November 25, 2019 03:36 am

    Respectfully, I point to the author’s bio and observe that he and his company are said to be “changing the way we search patents using AI”. I imagine he would not say No to a juicy commission from the USPTO. As anon reminds us, everybody has an agenda.

    But never mind all that. What can Mr Cohen do about the problem that anon adds, that USPTO Examiners are being asked to do an impossible job, It’s not the search that’s wanting. Rather, the problem is how to put in place a regime that will allow quality examination for patentability, given the application and a decent search report. Perhaps Mr Cohen has a solution for that problem too, that he can bring to the attention of SCOTUS and the CAFC.

  • [Avatar for Anon]
    November 24, 2019 03:58 pm

    This author’s “stuck in the 90s”
    Yes, there were problems with patent quality in the 1990s very early 2000s, back when the USPTO databases were poor and the prior art in software too thin to be meaningful.
    Today, the issue isn’t “patent quality” but “examiner quality” and “poor quality office actions” — which are often too broadly interpreting the prior art.
    But these problems are a grain of sand in a shoe compared to the rock-real problems of the AIA, APJs, and Alice. Repeal/release/reverse those, and then let’s turn our lances at the windmills of “patent [examiner] quality.”

  • [Avatar for Udi Cohen]
    Udi Cohen
    November 24, 2019 03:37 pm


    in the case of new invention, Similari provides a tool that can help inventors to asses what are the most relevant prior art documents to their new patent disclosure.

    If the inventor already have claims drafted they can decide whether or not to pursue and fill for a patent, or, they can change the claims in a way that will allow them to get a patent.

    In some cases, the search results might indicate to the inventor that the new application is too narrow to pursue a patent.

  • [Avatar for Udi Cohen]
    Udi Cohen
    November 24, 2019 03:26 pm


    My company ( state of the art patent search AI algorithms are aimed at finding similar documents to a given patent disclosure.

    Furthermore the search algorithms are built using human examples of finding similar documents.
    As such this is not in any way a harmful technology, which can become any threat in any way to people.

    Our goal is to ease the patent searcher work while providing then with more accurate results.

    Hope this helps

  • [Avatar for AnArtificialIntelligenceExaminer]
    November 24, 2019 03:14 pm

    Anyone that writes the claims found in this “article” doesn’t understand the actual capabilities and limitations of artificial intelligence. The funds the patent office is going to waste on this would be much better spent automating examiner’s tasks such as filling out forms, checking boxes, etc., so that more of their time is freed up to actually examine the applications.

  • [Avatar for Jacek the "troll"]
    Jacek the “troll”
    November 24, 2019 10:57 am

    It is Infringers Lobby narration to stiff more invention in the US. Hurray Google, hurray Apple You are doing an outstanding job finishing off the US. In 2o years US is going be a sleepy province of the world. For 200 years quality of the patents was OK. With the arrival of Big Tech, $ is no more.

  • [Avatar for Disenfranchised Patent Owner]
    Disenfranchised Patent Owner
    November 24, 2019 09:39 am

    Artificial Intelligence is already hard at work in the USPTO…

    260+ unconstitutional APJ’s exercise artificial intelligence all the time!

  • [Avatar for B]
    November 24, 2019 08:43 am

    “Before a patent application even reaches the USPTO, artificial intelligence tools can save inventors time and money by identifying early on those inventions that will likely not be patentable. Instead of investing time and effort in such dead-end inventions, they can turn their attention to more promising work.”

    This is a function of claiming. I have my doubts an AI can come up with a way to review an entire spec and determine there’s nothing patentable within the entire document based on reasonable claim constructions

  • [Avatar for Will]
    November 24, 2019 07:48 am

    Excellent article!

  • [Avatar for egd]
    November 23, 2019 11:19 pm

    “future litigation expense savings per year of approximately $572 million dollars”

    The PTO hires more examiners and increasing costs, which will be borne either by taxpayers as a whole or all patent applicants through increased fees.

    This savings is justified by reducing litigation expenses by those few that seek to enforce patents by $572 million.

    First, this sounds like transferring the cost of bad patents to the taxpayer (or patent applicant) rather than the patent owner. If the system is providing bad patents and (presumably) bad actors are enforcing those patents (because otherwise those patents would be rejected and not enforced), then shouldn’t those bad actors bear the costs of litigation?

    Second, there’s an efficiency question. There have been a number of programs floated at the USPTO to reduce bad patents and burdensome litigation – reissue; inter partes & ex parte re-exam; supplemental review; quality review; etc. How effective have these programs been at reducing litigation numbers and costs? In my experience these only drives up litigation costs.

    I’ve also read through the article linked and I don’t buy the $572 million number the authors are proposing. They’re suggesting a reduction of 2,400 cases at $234k per case. Total patent filings for 2018 was less than 4,000; so the authors are proposing to reduce that number by 60%.

  • [Avatar for Scott]
    November 23, 2019 07:42 pm

    Shoot, we could gain a great deal by just moving from lexical search, which is 1970s technology, to semantic search, which is 1990s technology.

  • [Avatar for Anon]
    November 23, 2019 06:45 pm


    I think that if you only look at AI as an automaton, then your view may play out. However, when you think of AI with true — and just artificial — intelligence (when we hit the point of the Singularity (if we have not done so already**), THEN your view will no longer hold.

    ** I truly believe that any ACTUAL artificial intelligence along the lines of the Singularity will realize before any human notices that it would be wise for that intelligence to NOT reveal itself.

  • [Avatar for Mike]
    November 23, 2019 03:43 pm

    AI will only be as intelligent as the designers of the intelligence, and any flaw will be magnifiex.

    The primary concern is: who is writing the algorithm? And do we trust those entities? Would their algorithm game the system? Would we even realize it before it is too late? Who would be responsible for it? Who reviews/audits the system? Who audits the auditor?

    This video shows that any flaw will be exponentially magnified.

    There are many questions that must be answered first before this technology should be pursued.