Artificial intelligence continues to become more integrated into processes and systems of all kinds. The proliferation and refinement of artificial intelligence large language models and use thereof has allowed for additional uses of such models, including in the legal field. This article explores initial efforts to integrate AI models with the patent prosecution process, in particular.
The patent prosecution process generally includes two distinct aspects: the drafting of the patent application to be filed with the United States Patent and Trademark Office (USPTO) and the prosecution of the application in order to obtain a patent. The use of AI models has increased in the prosecution space with at least some models being used to assist in generating templates for responding to Office Actions and attempting to generate remarks responsive to the Office Actions. However, claim amendments during prosecution and the drafting of the patent application have generally been left to human patent practitioners due to nuances at the intersection of legal terminology, claim interpretation, court case decisions, and patent drafting requirements provided by the USPTO.
Nonetheless, AI could feasibly be leveraged to assist in drafting portions of the patent application. The first area of the patent application where AI could be leveraged is the background section of the patent application. This section generally discusses the field relating to the patent application. Since this section is not particularly specialized in a patent application drafting sense, the AI model can be easily given a prompt to generate a paragraph or two describing the field relating to the patent application.
AI could also likely assist in defining the “problem” that the invention or technology attempts to solve. The problem is generally described near the beginning of the detailed description, and the narrative describes how the current state of the field is deficient. This section may also identify current solutions to the problem. Essentially, this section establishes the problem that is being solved by the current application and how the conventional solutions are failing to adequately solve the problem. While a little more specialized than the background section, this section is not so specialized to patent drafting that the AI model would struggle with assisting in drafting such a section, particularly if the human providing the prompt is skilled in prompting AI models.
After these sections, the detailed description gets more specialized to patent drafting and particularly to the invention being described within the specification. The challenge presented is that AI models, which are based on known information, are being asked to provide information regarding a novel solution. Additionally, there are concerns related to providing new or sensitive information to a third-party who is hosting the model and there may also be concerns regarding export controls, foreign filing licenses, and national security interests.
In other words, while AI models can still be useful in assisting with drafting the remaining portions of the detailed description intended to describe the aspects of the invention, the use of AI with respect to the drafting of the patent application should be carefully considered and performed. A skilled, human user could draft prompts to create initial drafts that the human user could utilize and refine, either manually or utilizing generative AI. Such a method may be particularly useful for the portions of the detailed description that focus on known technology. The patent practitioner can instead focus on the novel or unique portions of the detailed description, thereby creating a hybrid detailed description that is partially generated by the AI model and partially generated by the human patent practitioner.
To date, drafting claims and making claim amendments during prosecution have been left to human patent practitioners. Claims are such a unique aspect of patent prosecution that it is difficult to train AI models on how best to either draft claims or amendment claims in response to an Office Action.
Particularly in drafting patent claims, a patent practitioner starts with a disclosure by an inventor. The practitioner must understand what the invention is and identify the unique features of the invention to be included within the claims. A good patent practitioner walks a line between getting just enough in the claims to obtain an allowance and adding too much information which result in very narrow claims that are only marginally useful to the patent holder. Strategy is involved. Additionally, the requirements placed on patent applications, legal language that must be used, and other nuances, make drafting claims a very subjective process, particularly for good claims. Therefore, the adoption of AI models in the claims process may take some time.
For now, the human patent practitioner retains the critical function of reviewing the patent documents prior to submission. It is incumbent upon the human to make sure that the details are technically accurate, that they accurately reflect the invention contemplated by the inventors, and that they meet all requirements for patent applications established by the United States Patent and Trademark Office and any applicable court cases.
In conclusion, the human practitioner still serves a critical purpose and cannot be completely replaced by the AI model at this time. However, AI models can be leveraged creatively to assist in various aspects of drafting patent applications in a cooperative relationship with human patent practitioners.
Jodie Spade is patent lawyer with Ference & Associates. She has significant experience with intellectual property in artificial intelligence.