Bar Association of Erie County Bulletin
May/June 2024
Alexander C. Kaczmarek
The past three articles have focused on the substantive aspects of generative AI on copyright law and the right of publicity.
Just as generative AI continues to grow and evolve, the use cases for generative AI also continue to expand. Marketing professionals and software developers, for example, took to early all-purpose iterations of generative AI models for ease of obtaining simple text-based outputs. Now, programs have been tuned specifically for generative blog and advertisement copy, thumbnails, and website landing pages for marketing professionals and for code generation for software developers. Just as these industries have seen generative AI programs emerge to fit their needs, the legal field has seen specifically tuned generative AI programs emerge to meet (or at least attempt to meet) our needs. As Cyber Law begins a dive into the practical considerations of the use of generative AI programs, this first column focuses on the impact, challenges, and future potential of generative AI patent drafting and prior art search programs.
Generative AI patent drafting and prior art search programs claim to be specifically tuned generative AI models designed and trained to streamline the drafting of patent applications and responses to office actions. Based on training sets of global patents, scientific journals, and other technical documents, these models analyze input prompts describing the invention against existing technical information and prior art, to generate patent applications. Specifically, many programs tout an ability to perform prior art searches and provide similarity scores, generate patent applications (including full generation of the background, description, and claims based on a simple description of the invention as an input), refine technical language for ease of understanding, perform predictive analytics on potential challenges during the examination phase and likelihood of approval, and draft responses to office actions.
Turning to generative AI itself, an input prompt asking “describe the benefits of generative AI patent drafting programs in two paragraphs, to be included in an article for a legal audience” returns the following output:
Generative AI patent drafting programs offer a streamlined and efficient approach to creating comprehensive patent applications. They automate the often laborious task of drafting, leveraging extensive data sets to rapidly generate precise and clear claims, specifications, and diagrams. This efficiency reduces the time and costs associated with preparing patent documents. Additionally, they can identify and include relevant prior art references, thereby enhancing the defensibility of the claims. The use of these programs allows patent professionals to focus on higher-level strategic considerations, such as refining the inventive concept and crafting unique arguments for patentability.
Moreover, generative AI enhances the quality and consistency of patent applications. The technology’s ability to analyze vast amounts of legal precedent and technical literature ensures a well-structured, standardized output that aligns with best practices in patent drafting. It minimizes errors and inconsistencies, offering a higher degree of predictability in obtaining patent grants. The programs can also tailor applications to meet specific jurisdictional requirements, improving global patent prosecution efficiency. For law firms and corporate legal departments, generative AI represents a significant advancement, helping practitioners to handle larger portfolios and improve client satisfaction by offering faster, more accurate, and cost-effective patent drafting services.
Appreciating these benefits, and in response to President Biden’s Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence, the United States Patent and Trademark Office (“USPTO”) recently issued guidance on the use of AI in practicing before the USPTO.[1] The guidance covers the existing rules and policies and how each applies to the use of AI tools in matters before the USPTO.
Taking each rule and policy in turn, the guidance discusses (1) the duty of candor and good faith (including the duty to disclose all information known to be material to patentability); (2) signature requirements and corresponding certifications; (3) USPTO electronic systems policies; (4) confidentiality of information; (5) duties owed to clients; and (6) foreign filing licenses and export regulations. These rules and policies apply universally to all patent practitioners, regardless of whether AI tools were used in preparing submissions.
The duty of candor and good faith requires disclosing all information material to patentability and prohibits any form of fraud or intentional misconduct. This duty is enforced by the signature requirements and corresponding certifications required for nearly all forms of correspondence with the USPTO. Practically, this duty is also enforced by the USPTO electronic systems policies requiring individual, non-transferable or sharable, accounts for each practitioner making a submission before the USPTO. Together, none of these rules or policies prohibits the use of generative AI patent drafting or prior art search programs, and do not imply a general obligation to disclose the use of such programs,[2] but do require oversight and review by the submission’s signatory. One example where disclosure would be required is where generative AI, in interpreting the description of the invention, generates one or more claims that do not have a significant contribution by a human inventor. This could occur where the generative AI patent drafting program introduces alternative embodiments that the inventor did not conceive. Examples where modifications or limitations are required include verification and editing of specifications and drawings prepared by a generative AI patent drafting program for accuracy and proper form; and manual review and removal of cumulative or irrelevant matter in an Information Disclosure Statement prepared and compiled by an AI prior art search program.
The next three listed rules and policies on confidentiality, the duties owned to clients, and foreign filing licenses and export regulations deal with the programs used and the inputs made into such programs. Where a generative AI patent drafting program gains a license to use inputs for training, a client’s proprietary information may filter into outputs provided by the program to other third party users. Not only would this be a potential breach of a practitioner’s ethical duty to maintain confidentiality of information, but such disclosure and reuse of the existence of and enabling information on the invention could potentially be considered a non-confidential public disclosure triggering the grace period for filing a patent application in the United States and barring the filing of an application in some other countries. If a generative AI patent drafting program’s servers are located outside of the United States, the entry of invention information may be considered an export. Even where servers are located in the United States, use of that generative AI program by non-U.S. persons may be deemed an export.
The guidance is not intended to be exhaustive on or unnecessarily restrictive of the use of generative AI patent drafting or prior art search programs, but is instead an affirmative statement that the USPTO does not intend to limit the growth and development of such programs and that the USPTO rules and policies already in place are well equipped to address many of the concerns raised by such programs. As further evidence of this acceptance, the USPTO itself has implemented the use of AI prior art search tools for patent examiners.[3]
Generative AI patent drafting and prior art search programs represent a significant advancement toward legal services focused generative AI programs. As specially tuned generative AI programs trained on expansive sets of global patents, scientific journals, and other technical documents, these programs promise improved efficiency, accuracy, and creativity while reducing costs and streamlining the patent drafting process. However, in the next Cyber Law column, we will dig deeper into these promises through a discussion of challenges and deficiencies caused by legal complexity, data quality, and contextual understanding, and strategies for effective implementation of generative AI patent drafting and prior art search programs.
[1] See Guidance on Use of Artificial Intelligence-Based Tools in Practice Before the United States Patent and Trademark Office, 89 Fed. Reg. 25609 (Apr. 11, 2024); see also Exec. Order No. 14110, 88 Fed. Reg. 75191 (Nov. 1, 2023).
[2] The duty to disclose the use of such programs is implicated when the use rises to the level of materiality under 37 CFR 1.56(b).
[3] See Guidance on Use of Artificial Intelligence-Based Tools in Practice Before the United States Patent and Trademark Office, 89 Fed. Reg. 25609, 25610 (Apr. 11, 2024)