EP 3 118 356 B1 relates to methods or securely starting and/or stopping a rotor of a rotor spinning machine and rotor spinning machine.
Brief outline of the case
The OD maintained the patent according to AR1. The opponent appealed.
The board decided that claim 1 as maintained lacked N over a document not mentioned in the ESR.
The same applied to AR1-3. AR 4 and 5 lacked IS.
The patent was thus revoked.
The case is interesting in that the proprietor referred to ChatGPT to defineIinterpret some claimed features.
The proprietor’s point of view
At the OP before the board, the proprietor referred to various terms used in claim 1, in particular “position control” and “control” as opposed to “monitor”, with answers received from the chatbot ChatGPT to corresponding queries.
The proprietor merely presented orally the answers received from ChatGPT.
The board’s decision
The board observed that the proprietor did not submit, in writing for the file, the extensive oral answers, some of which were only presented briefly. Therefore, the exact content of the answers received from ChatGPT could not be taken into account for the decision.
The board noted in this context that ChatGPT’s response is in itself irrelevant, since the interpretation of the features in the claim and the claim itself, has to be based on the understanding of the skilled person. In this context, the board referred to T 206/22, Reasons 1.
The generally increasing prevalence and use of chatbots based on “large language models” and/or “artificial intelligence” alone does not justify the assumption that an answer received necessarily correctly reflects the understanding of the skilled person in the respective technical field at the relevant time.
The answers given in reply to queries to a chatbot are based on training data unknown to the user and may also depend sensitively on the context and the exact formulation of the question.
Evidence of how certain terms in the claim of a patent, or patent application, are interpreted by the skilled person can, for example, be provided by appropriate technical literature.
Comments
It was to be expected that, sooner or later, parties would resort to answers given by chatbots to interpret claimed features.
The board rightly observed the such replies depend clearly from the training data of the chatbot and of the question adressed to it. It is well known that training data of a chatbot can introduce a bias in the later replies and that a answer can vary with the question put to the chatbot.
Even if, in the present case, the answers from the chatbot would have been filed in written form, they could not have been used in the interpretation, as their publication date would either be unknow, or if known, they could be based on data input in the chatbot after the effective date of the claim.
As observed in T 206/22, the information used by the chatbot to determine its interpretation is based, at least in principle, on all the information available to it, including documents published well after the effective date of a claim. It is therefore not an interpretation that necessarily corresponds to what a skilled person would have understood at the priority date.
It is even possible to imagine that, further than the mere interpretation of claimed features, chatbots could be used to create prior art. For the above reasons, such created prior art would certainly not represent prior art which could be used against the validity of a patent.
Such uses of chatbots should be nipped in the bud.
When quoting internet citations as prior art, it has to be shown that what comes from the internet has been published before the effective date of the claim. This does not appear not possible for data outputted from a chatbot. Wayback machines for chatbots are still to be invented.
At least as far as interpretation of claimed features is concerned, a chatbot will not replace the skilled person. Technical literature will be need, be it in paper or in digital form, but always with precise information about the publication date.
Creating prior art with a chatbot is certainly possible. Here again the its publication date, the training data and the question to the chatbot leading to said prior art are relevant aspects which will have to be taken into account. The enablement of AI generated prior art is also a question which will require careful attention.
One question which also arises is whether inventions generated “as such” or with the help of AI are patentable. I refer here for example to a conference held in Alicante in 2019, commented on IPKat.
The situation is still moving and we all know what happened to the two DABUS applications.
We certainly live in interesting times.
Comments
1 reply on “T 1193/23 – A chatbot does not replace the skilled person”
It was a desperate argument from a desperate representative. It is somewhat disappointing that the argument the representative tried to make probably was right. LLMs, after all, can only output what is likely in view of their training data, so there probably were citations that could have supported the representative’s position if only they could have found those sources. Rather than ChatGPT, the representative should have at least used a LLM with RAG that could provide sources.