close icon-linkedin icon-twitter icon-facebook icon-mail icon-google-plus icon-search icon-phone icon-instagram
What are you looking for?
In T 1193/23, the Boards of Appeal (the Boards) revoked a patent relating to a method for safely starting and/or stopping a spinning rotor of a rotor spinning machine (electric motor). The decision notably addresses the relevance of chatbot (specifically ChatGPT) answers in claim interpretation, concluding that such answers cannot automatically be assumed to reflect the skilled person's understanding.
Case law 20 May 2025

T 1193/23: Chatbot answers alone cannot establish the skilled person's understanding of claim terms

In T 1193/23, the Boards of Appeal (the Boards) revoked a patent relating to a method for safely starting and/or stopping a spinning rotor of a rotor spinning machine (electric motor). The decision notably addresses the relevance of chatbot (specifically ChatGPT) answers in claim interpretation, concluding that such answers cannot automatically be assumed to reflect the skilled person's understanding.

Background

The Opposition Division had maintained the patent in amended form. The case concerned a method for safely starting and/or stopping a rotor of a rotor spinning machine. Key features involved checking the position control for active magnetic bearings and/or the data connection to the motor control, and blocking the start or stopping the rotor if predetermined setpoints or states were not reached.

During appeal a dispute arose regarding the interpretation of several claim terms, particularly "position control" and "checking" versus "monitoring." The patent proprietor had cited ChatGPT answers to support their interpretation of these terms.

The Board's reasoning

The Board first addressed the use of ChatGPT answers in claim interpretation. The Board noted that "the mere increasing prevalence and use of chatbots based on language models ('large language models') and/or 'artificial intelligence' does not yet justify the assumption that a received answer - which may be based on training data unknown to the user and also depend sensitively on the context and exact wording of the question(s) - necessarily correctly represents the understanding of the skilled person in the respective technical field (at the relevant time.

The mere increasing prevalence and use of chatbots … does not yet justify the assumption that a received answer … necessarily correctly represents the understanding of the skilled person in the respective technical field.

The Board explained that evidence of how certain terms in a patent claim are interpreted by the skilled person may be demonstrated through appropriate technical literature, but no such evidence was submitted in this case.

Regarding the disputed claim terms, the Board rejected the patent proprietor's narrow interpretation. The Board found that the term "checking" in the claim does not exclude "monitoring," contrary to the patent proprietor's arguments. The Board also concluded that "position control" could refer to both a device and a process, not just the device as argued by the proprietor.

The Board ultimately revoked the patent for lack of novelty over prior art document D3, which was found to disclose all claimed features either explicitly or implicitly.

Decision of the Board

The Board found all requests unallowable and revoked the patent.

Summary written by the NLO EPO Case Law Team