Can ChatGPT draft patent applications?
Article co-written by Yuri Levin-Schwartz, Ph.D., a law clerk at MBHB.
OpenAI’s prototype of ChatGPT, released at the end of November 2022, drew rapid, widespread public attention for its ability to provide articulate, human-like responses across a wide variety of knowledge domains. Numerous articles have already explored large language models like ChatGPT and what they can and cannot do – as well as the broad impact of such models on the legal profession . In this article, we explore how large language models like ChatGPT may impact patent prosecution in the near future.
For inspiration, we prompted ChatGPT: “As a patent attorney, how can I utilize ChatGPT?” Here’s the first use case output by the AI model:
(1) Drafting patent applications: ChatGPT can help you generate patent claims and descriptions of inventions, which are key components of a patent application.
Indeed, ChatGPT can write patent claims by analyzing input information such as the technical field of the invention, the problem the invention solves, and a description of the invention. However, the quality of the claims that ChatGPT provides depends directly on the relevance and completeness of the input information. For example, if you ask ChatGPT to write claims relating to simply “a bicycle,” it will provide 20 total claims and three independent claims (consistent with the no-fee limits for claims in U.S. patent applications) that relate to a generic bicycle. Claim 1 is:
A bicycle comprising a frame, two wheels, and a handlebar, the frame being configured to support a rider, the two wheels being rotatably attached to the frame, and the handlebar being operatively connected to the frame and configured to enable steering of the bicycle.
The dependent claims describe a chain, derailleur, different gear configurations, brakes, and so on.
The problem with the scope of this claim set, of course, is that it doesn’t include anything novel or non-obvious. To move closer to a useful claim set, we need to include more information about the potentially novel elements.
As a narrower example, we could change our focus to a bicycle having “a shifter cable formed from carbon fiber filaments”. In that case, the ChatGPT output changes to:
A bicycle comprising a frame, a fork, at least one wheel, a shifter cable formed from carbon fiber filaments, a shifter mechanism for controlling the shifting of the gears of the bicycle, wherein the shifter cable is operatively connected to the shifter mechanism and extends from the shifter mechanism to the gears of the bicycle.
This scope is an improvement over the generic bicycle claim, but it also introduces new problems, including 1) an antecedent basis problem for “the gears”; 2) potentially extraneous elements such as “a fork” and “at least one wheel;” 3) potential “means plus functions” interpretations with “a shifter mechanism for…”; and 4) in our example output, several multiple-dependent claims that would draw huge fees from the USPTO. Problems aside though, from a patent practitioner’s high-level point of view, ChatGPT could speed up the process of drafting a patent application by rapidly generating numerous candidate claim iterations.
In some cases, this might mean that, instead of spending time wordsmithing a set of claims, a patent attorney could focus more closely on an invention’s specific point of novelty and work with the inventor(s) to develop a plain English summary of the specific differences between a generic device/system/method and the innovation.
Such a focused summary could provide ChatGPT with better inputs from which to form more relevant candidate patent claims and be used as a seed to ask ChatGPT to provide an initial patent specification. For example, if we input “write a patent specification for a bicycle with a shifter cable formed from carbon fiber filaments,” ChatGPT provides a relevant title, field of the invention, background, summary, and a brief, but reasonably complete, detailed description. Such outputs could potentially be used as a base template or an outline to draft a more complete patent specification.
In the near term, patent practitioners may also consider using ChatGPT alongside other claim drafting tools, such as ClaimMaster or Juristat, among others. These tools can help patent attorneys avoid potential issues, including antecedent basis, means-plus-function, indefiniteness, Section 101, among other possibilities.
Remember, your conversations aren’t private
One important issue involving large language models is how they use the information from user prompts. For example, a patent attorney must make sure to not disclose a client’s confidential information through interactions with ChatGPT. Prompts provided to ChatGPT, and likely most other large language models, are not private at this time. Specifically, OpenAI’s FAQ provides the following:
Can you delete specific prompts?
No, we are not able to delete specific prompts from your history. Please don’t share any sensitive information in your conversations.
Communication Information: If you communicate with us, we may collect your name, contact information, and the contents of any messages you send (“Communication Information”).
As such, patent attorneys must take care not disclose confidential information to publicly-accessible large language models like ChatGPT. A court could consider the content of the messages as public disclosure of the invention because OpenAI has no obligation to secrecy. Inadvertent disclosures could result in a loss of patent rights and/or a time bar if a utility application is not filed within one year of the disclosure.
Even with these cautions and potential issues, ChatGPT and other large language models have the potential to reduce drafting work for patent attorneys by rapidly providing candidate patent claims and corresponding detailed description. This may give patent attorneys more time to focus on the creative and interactive aspects of patent law.