June 2025 In today’s world, artificial intelligence has become a part of many aspects of our daily lives. Unsurprisingly, AI is also making its way into the field of intellectual property law, as large language models are being used in proceedings before the EPO.
A recent example of this trend can be seen in the recent decision T 1193/23, where ChatGPT’s output was cited during the oral proceedings before the Board of Appeal. In this case, ChatGPT was used to interpret a term in the opposed patent claim to show how ‘a skilled person’ would interpret the meaning of a particular word.
Background
The patent in question concerned rotor spinning machines, specifically, the invention aimed to improve safety of the starting and stopping processes of high-speed rotors.
The patent was challenged by an opponent who claimed that the invention lacked novelty in light of a single prior art document. One of the key arguments against novelty was based on the interpretation of the term “check” as found in the opposed patent, in contrast to a to a term “monitor” which was found in the prior art document.
The opponent argued for a broad interpretation, suggesting that “check” encompassed the act of “monitoring”, making the claimed invention overlap with the prior art. Meanwhile, the patentee argued that “check” referred to a one-time verification, while “monitor” implied a continuous process. To support this interpretation, the patentee presented an output from ChatGPT which aligned with their interpretation, showing that AI understood these two terms to be essentially different.
EPO decision
After hearing this and other arguments, the Board of Appeal decided that:
- ChatGPT’s output is irrelevant because the interpretation of the claim depends on the skilled person’s understanding, and ChatGPT does not necessarily correctly reflect the understanding of the skilled person in the respective technical field at the relevant time.
- ChatGPT is trained on data that is unknown to the user and it is sensitive to the context and formulation of the question, which means that the output cannot be trusted.
- ChatGPT could not establish an interpretation of the claim alone and that it should be supported by expert evidence. The Board of Appeal specified that that additional proof is needed, for example, technical literature.
- The main request lacked novelty over the prior art.
Summary
Artificial intelligence is taking over our daily lives, and intellectual property field is not an exception. We can see how the outputs of large language models such as ChatGPT are now being tested for use in legal research, submission drafting and even as evidence in court proceedings.
This EPO decision shows that AI has some value as it does not completely rule out the use of AI for claim interpretation, but it does indicate that any evidence based on the output of an AI has to be supported by further expert evidence. Ultimately, this decision establishes that AI cannot take the role of a skilled person.
