“Limits of Using Artificial Intelligence and GPT-3 in Patent Prosecution”, 2022 ():
[cf. 2022] GPT-3 may have a dramatic impact on the methods practitioners use to draft patent specifications and, more importantly, patent claims. GPT-3 may be able to provide practitioners with ready-made context-consistent language that can greatly enlarge the scope of a patent claim. Given the proper prompt, GPT-3 may also be able to generate claims and specifications for claims and translate “legalese” into understandable natural language. Although GPT-3 has been shown to have limitations when generating creative writing styles, these limitations may not be as problematic when generating patent claims and specifications.3 This is because patent drafting is constrained by a unique set of rules, canons, and language that has already been litigated.4
GPT-3 and other AI technologies have the potential to revolutionize patent prosecution. However, there are many pitfalls that patent prosecutors and litigators should recognize when dealing with this new technology.5 Additionally, there are many limits to what AI can and cannot do when it comes to patent-claim and specification drafting.6 In this Article, we argue that traditional patent doctrines such as “enablement” and “specific utility” should be bolstered to act as gatekeepers to limit the claims generated by AI. Additionally, we argue that United States patent law may ultimately be forced to move from a peripheral claiming system to a more central claiming-based system to deal with AI-generated claims. Furthermore, practitioners will have to think carefully about how to fulfill their professional obligations to supervise GPT-3 style technology, especially in light of its tendency to generate biased content.7 Finally, if the technology is only available to large private firms, it may further exacerbate the access to justice gap.
…Working examples may play a more important role especially if GPT-3 has the ability to draft and use “prophetic examples” when drafting patent applications.” Working examples are examples that have actually been performed in a laboratory or in a real-world setting.” In contrast, prophetic examples are examples based on experiments that are never performed.” Prophetic examples are used in patent applications to illustrate the potential and hoped-for uses of a patented invention.” Currently, these prophetic examples are sanctioned by the USPTO and can help establish specific utility for the invention, even if those uses are not enabled.” The problem with the use of GPT-3 is that the AI tool can create an almost unlimited number of convincing prophetic examples.” This problem is exacerbated because prophetic examples can be drafted in a way that is difficult to distinguish from actual working examples.
[Prophetic examples always struck me as absurdly abusive, so if generating them in bulk kills them, then good.]