From ChatGPT to ChatG-“PSA”: Can ChatGPT be a Person Skilled in the Art?

The increasing use of artificial intelligence, including Large Language Models (LLMs), in the life sciences and healthcare industries (for example, in drug discovery, research and development), and the strategic use of patents to secure market advantage or exclusivity in these industries, lead to several questions about the impact AI technology may have on the way validity and infringement of life sciences and healthcare related patents may be assessed. One of these questions is whether LLMs, such as ChatGPT, could be considered to be a “person skilled in the art” (PSA), which is a key legal construct affecting construction and validity in patent matters.  

AI is already transforming the life sciences and healthcare sectors. In this article, we consider whether and how AI might further affect these industries by changing the concept of the PSA, including a recent decision by the European Patent Office and whether Australian courts and the Australian Patent Office are likely to come to the same conclusion.

Person skilled in the art 

The PSA is not a real person but an artificial construct (or a “legal fiction”) that is used as a tool of analysis in patent law. “Who” the PSA is in each case changes, depending on the field of the invention claimed in the patent at issue, so that the PSA represents a hypothetical skilled but non-inventive worker in the relevant field. In some cases, the PSA can also be a team of hypothetical skilled but non-inventive workers across a variety of disciplines that would have worked together at the priority date in the relevant field (e.g. a pharmacologist, a medicinal chemist and a formulation chemist). 

The PSA is central to consideration of several validity grounds. In particular, when assessing whether or not an invention claimed in a patent is obvious (i.e. lacks inventive step) in Australia, the question is whether the PSA or notional research group at the priority date, in all the circumstances, would have been directly led as a matter of course to try the claimed invention in the expectation that it might well produce a useful desired result. The way a PSA would understand and interpret the patent is also a key aspect of claim construction where there is a “term of art”, making it relevant to both infringement and validity in a patent proceeding.

The EPO Board of Appeal decision in Rieter

The question of whether a LLM, such as ChatGPT, could be considered to be a PSA was dealt with in the European Patent Office Board of Appeal decision in Rieter CZ s.r.o. v Saurer Spinning Solutions GmbH & Co. KG (T 1193/23), where the Board answered in the negative, finding that ChatGPT does not accurately reflect the understanding of a PSA.  

In Rieter, the patent related to a method for safely starting and stopping a rotor, and the opponent had asserted that the claims lacked novelty and inventive step. During the proceedings, the patentee sought to rely on responses from ChatGPT as to the interpretation of terms used in a claim. However, the Board found that ChatGPT’s responses were irrelevant in this context because claim construction depends on the understanding of a PSA.  

The Board held that the increased use of chatbots based on LLMs and AI does not justify the assumption that ChatGPT’s responses accurately reflect the understanding of the PSA in the technical field at the relevant time. Not only are LLMs based on training data, which is unknown to the user, but their responses depend on the context and precise formulation of the questions a user asks, making those responses unreliable and irrelevant to the Board’s consideration.   

Could ChatGPT be a PSA in Australia?

Australian courts have yet to consider the question of whether a LLM can be considered a PSA. Some guidance can be taken from the landmark decision in Commissioner of Patents v Thaler (2022) FCAFC 62, where a special 5-member Full Court of the Federal Court of Australia considered the nature of personhood in the context of an invention alleged to have been made by an AI.

In its decision in Thaler, the Full Court concluded that an inventor has always been considered to be a natural person and that having a non-human inventor in a patent application contravenes the Patents Act 1990 (Cth). The High Court refused special leave to appeal so the Full Court decision stands.  

Australian courts frequently point out that the PSA is a tool for legal analysis, not an actual person, meaning there is not necessarily the same requirement for the PSA to be a natural person as there was in Thaler when considering the statutory definition of an “inventor” for the purposes of a patent application. For example, in AstraZeneca AB v Apotex Pty Ltd [2015] HCA 30, the High Court stated that AstraZeneca had conferred upon the PSA “more human characteristics of volitional and purposive action than are necessary for its function”, and that the PSA “is a pale shadow of a real person”.

However, part of the Full Court’s reasoning in Thaler is equally applicable to the concept of a PSA when, for example, considering evidence of obviousness (i.e. lack of inventive step). The Full Court held (emphasis added):

“it is plain from [the considered case law] that the law relating to the entitlement of a person to the grant of a patent is premised upon an invention for the purposes of the Patents Act arising from the mind of a natural person or persons. Those who contribute to, or supply, the inventive concept are entitled to the grant. The grant of a patent for an invention rewards their ingenuity.”

If a patent must be granted based on human ingenuity, then the same reasoning should fairly be applied to the notional “mind” of a PSA whose hypothetical actions and expectations of success are to be considered in determining whether the alleged invention involves an inventive step.

Moreover, as a legal construct, courts and patent offices are called upon to set the boundaries of what a PSA does and does not know. It would be difficult, if not impossible, for a party to put on sufficient evidence of the data underlying a LLM, or of the limits and parameters asserted by its creator. For example, ChatGPT will not promote or endorse illegal drugs or substances, even though substances currently deemed illegal could one day form the basis of new therapeutic treatments (e.g. consider the evolving legal status of psylocibin over time). Without understanding these things, it would not realistically be possible for courts and patent offices to ascertain the extent of the LLM’s “knowledge”, nor to exclude particular aspects of that knowledge, depending on the relevant field of the invention, for example.

Given all this, Australian courts are unlikely to find that current LLMs like ChatGPT are representative of a PSA, nor that they can give evidence relevant to the steps or state of mind of a notional PSA.