Can AI Improve Patent Quality?

Congress aimed to encourage innovation by incentivizing inventors with patents, but an overburdened USPTO continues to grant low-quality patents. Although there may be a role for artificial intelligence to improve the patent prosecution process, there are some challenges outside of the USPTO’s control that limit the technology’s potential.

Attribution

JPxG on Dall-E 3

The purpose of patent law is to promote innovation for the benefit of society. Congress created a system to reward great ideas that improve the way we all live, but this system is not optimal. Reports say that the United States Patent and Trademark Office (USPTO) currently receives over 500,000 patent applications a year. Out of the patents filed, only about half of these are eligible to be patented. Still, every application must be examined as mandated by 35 U.S.C. § 131. The examination process involves searching all prior art, reviewing applications to ensure they meet formal requirements, and writing office actions communicating findings on patentability to applicants and/or other relevant parties. The review of prior art is what usually leads to patent quality issues. Examiners often fail to flag the deficiency of weak patents, specifically ones that are either obvious or lack novelty. Although litigation serves as a safety net for low-quality patents that have survived the prosecution process, it is an overly expensive method that should not be the primary means to deal with this issue.

Patent examiners are letting low-quality patent applications go through because they are overwhelmed with an immense workload and because their performance is measured by the number of applications they examine. IPWatchdog reports that the USPTO is currently dealing with a backlog of over 826,000 unexamined applications. Although the agency is working to increase its staff, with 969 patent examiners hired in fiscal year 2024 the rate of new hires is not high enough to counteract the flood of annual patent applications, let alone the ones in the backlog. Many scholars suggest that a better way to attack issues of patent quality is to improve the procedures and standards examiners use to grant them. A new system is needed to remove ineligible patent applications at a quicker rate than can be done by individual examiners.

For a patent to be granted, the invention must be novel, non-obvious, and patentable subject matter that is sufficiently documented and filed under a timely patent application. Given the rise of artificial intelligence (AI), many including the USPTO itself have wondered whether AI could help in improving the patent system, particularly with regard to the prior art search process. Imagine a process that filtered for the best quality patents before they arrived at the desk of a patent examiner. AI could prove useful because of its unmatched speed and efficiency, its revolutionary search capabilities, its abilities to automate repetitive tasks, map results contextually, and overcome lingual and jurisdictional barriers. At the very least there may be a role for AI to flag certain applications that fail obvious requirements for eligibility. Such a system would alleviate some of the burden on examiners to determine the validity of patents.

It is important to note, however, that there are some obstacles to using AI in patent prosecution. Given the USPTO’s current financial limitations, optimally integrating AI into the examination process might not be possible. There is also a potential for AI prior art searches to overload patent examiners with more information than they can handle, much of it being of undetermined quality. Meanwhile, some scholars argue that the problem of patent quality is even deeper than the incapacity of the USPTO, casting further doubt on AI as a solution. They argue that the issue of low patent quality really stems from systemic incentives to file weak patent applications. Patentees are encouraged to be as vague as possible when describing their inventions so they can expand the scope of their patents after they are granted. The dominance of low-quality patents in the patent system also motivates patentees to acquire a large portfolio of low-quality patents over a small portfolio of higher-quality ones. In such a system, administrative fixes alone may not be able to resolve the entire problem.

AI has made a major impact in recent years with many industries rushing to integrate it into their systems. Although it may have a future role in patent prosecution, both AI and the USPTO will have to overcome several hurdles before its use can become a reality. Even then, its impact on reducing weak patents will be limited by factors outside of the USPTO’s control. Separate work will have to be done to change the systemic incentives to file low-quality patent applications and until then, there is little that AI can do to help.