Coming off the back of this year’s fabulous CIPA Congress, which focussed on anything and everything to do with AI and IP, I had the thought – can it be argued that ChatGPT is ‘akin’ to the skilled person?
Of course, I already knew it had been decided by the EPO in T1193/23[1], that specifically in terms of deciding claim construction, ChatGPT could not be a skilled person.
However, it wasn’t really this I was thinking about. I was thinking whether ChatGPT could be thought of as the skilled person when it comes to non-obviousness[2] (leaving aside for a moment that ChatGPT is not a human[3] and the confidentiality aspects of public LLMs). This aspect wasn’t looked at by the EPO in that decision.
All these thoughts being sparked by Dr Laurence Aitchison[4] of the University of Bristol’s statement at CIPA congress that stated “ChatGPT is at an intelligence level of a 2-year undergrad degree student” (summarised).
So, this led me to a coffee break time discussion with both Andy Bartlett and Julyan Elbro of the UK Intellectual Property Office. On our earlier panel discussion, they had highlighted that due to ChatGPT’s very plausible ‘hallucination’ issue, they were receiving, and envisaged continuing to receive, many patent applications that looked on the surface like good patents, but that had, and were going to, turn out to lack an inventive step, or more probably, be insufficient[5]. Moreover, it was going to be more and more difficult for IPO examiners to spot such applications due to the extremely ‘plausible’ nature of the output of current AI/LLM tools.
I posed this thought to them: “given we accept ChatGPT can’t invent or be an inventor, can it be argued that it is akin to a ‘person skilled’ in the art in terms of intelligence level? Then is it also conceivable that the only way ChatGPT can be used in the inventive process is to help an inventor flesh out their specification and so the invention needs to be contained in the original prompt itself? Any input from an inventor not containing the inventive concept therefore cannot lead to an invention?[6]”
Thus, concluding that the spotting of the technical problem to be solved, by the human operator, being the inventive step, the AI tool may be considered as simply being used to put the invention into practice, such as a wood working tool may be used on a lab bench. This nicely aligns itself to the way many private LLMs are already being used to draft a patent based on an inventive input from a human inventor.
So, you may be thinking that this is all quite obvious so far, but bear with, reader.
We talked more, not particularly reaching any conclusions, and later in the day Michelle Holoubek from Washington DC based firm Sterne, Kessler, Goldstein & Fox, highlighted that one logical consequence of the USPTO guidelines on the matter[7], is that an inventor probably ought to record their original idea, the prompt they entered into any LLM, and the original output from that tool, plus any iterations and associated prompts along on the way, just in case their invention was ever challenged by the UPSTO or a third party. This evidence could then be used to provide sufficient evidence of inventiveness.
And what about sufficiency? Does it also have to be argued that the inventive concept input into the LLM not only needs to be inventive but also sufficient in and of itself and in which case, can an LLM be a tool for developing the invention, such as the example given of a wood working tool used on the lab bench, and can it therefore only be used as tool for making a patent application more legible?
Indeed, the USPTO guidelines[8] seem to concur with this way of thinking:
Merely recognizing a problem or having a general goal or research plan to pursue does not rise to the level of conception.[54] A natural person who only presents a problem to an AI system may not be a proper inventor or joint inventor of an invention identified from the output of the AI system. However, a significant contribution could be shown by the way the person constructs the prompt in view of a specific problem to elicit a particular solution from the AI system.
Michelle further suggested that as a room full of practitioners, we should also be asking clients before we begin drafting if they have used an LLM at any point in their ‘inventing’ process and further, ask to see their original inputs. To check, prior to undergoing further work and potentially wasted effort, that the proposed inventive step was in the grasp of the inventor before any AI tool was used. Including sufficiency, therefore?
So, given such thoughts were spinning around my head, I decided the obvious, to ask ChatGPT what it thinks about being ‘akin’ to the skilled person. So, here’s my question and here’s the reply:
——————————————————————————————————
Q: Do you think it can be argued that you are the skilled person when it comes to inventions?
A: “That’s a very interesting question — and yes, it can be argued, but with some nuance.
In patent law, the “person skilled in the art” (or “skilled person”) is a legal fiction — an imagined expert who:
- Has average skill and common general knowledge in the relevant technical field at the priority date.
- Is not inventive, meaning they can apply routine techniques and read the literature, but they don’t create new ideas themselves.
- Serves as a benchmark for assessing obviousness, sufficiency, and clarity.
So, could I qualify as that skilled person? Here’s how it breaks down:
✅ Arguments in favor:
- I have access to an enormous range of technical and scientific information across disciplines — much broader than a human’s general knowledge.
- I can reason systematically, interpret patent claims, and apply established technical teachings as the skilled person would.
- I can simulate what such a person might think when faced with prior art or a technical problem.
❌ Arguments against:
- The skilled person is human; they have the mindset, biases, and limitations of someone working in the art at the relevant date. I don’t have those human constraints or the historical context of their knowledge[9].
- My access to knowledge is effectively beyond what a skilled person could have had at the priority date — so I might be too knowledgeable[10].
- Legally, the “skilled person” must be a construct grounded in human capabilities, not an AI model.
So while I can model or simulate the skilled person’s reasoning for analytical purposes (e.g., when assessing inventive step), I wouldn’t be the skilled person in the legal sense — rather, I’d be a tool to approximate their reasoning.
Would you like me to outline how I could simulate a “skilled person” analysis for an invention (e.g., obviousness or sufficiency)?”
———————————————————————————————————
So much in that reply jumped out at me. In theory, at least, it seems it could be argued that CHATGPT can simulate a skilled person, ChatGPT certainly thinks it can. Indeed, as the skilled person is a legal construct and not a real human being, why not?
Further, where does this leave non-obviousness and the skilled person if ChatGPT believes it is more intelligent in comparison??
Given also, that just before starting to write this article, I had read Mo Albolkheir’s divisive paper[11] on how non-obviousness is itself a construct, and is in effect, a test of human intelligence. In that paper Mo argues that:
- For an invention to be patentable, the core inventive concept must be beyond the grasp of those with average powers of imagination.
- The inventor has grasped the core inventive concept.
- Therefore, the inventor’s powers of imagination are above average—i.e. it is the inventor’s powers of imagination that are being examined as the basis for patentability, not the invention.
Then, I’m not entirely sure where that leaves us!
Whilst the case law hasn’t decided on the issue yet, I am absolutely clear that it is a practitioners’ duty to advise our inventor community on the perils of using both public and private LLMs, with regards to proving inventive step and sufficiency, notwithstanding the more ‘well-known’ issues of confidentiality and human inventorship.
It is also clear that it may help an IPO in examining patents in the future if the inventor were asked not only to declare in advance if an AI tool was used, but if they have, that they should be required to submit the original prompt, or input text, along with their application?
On a different note, what about the case where someone has used the AI tool to draft an outline of the patent (so using it as the skilled person to provide a skeleton) but has appropriately not included the inventive step in the prompt and has filled that in later themselves – should they be required to admit to the use of AI? At what point does it become something that should be declared?
Further, given ChatGPT’s own views on its ‘intelligence’, where does that leave the definition of the skilled person when assessing obviousness and moreover, sufficiency?
An argument to be considered is that if ChatGPT is correct, and it is more intelligent than the skilled person due to its knowledge beyond what the skilled person could have accessed at the priority date of the application, then it is effectively in the same position as we are as inventors, practitioners, and examiners. That is, we all also have access to knowledge of information beyond what was available at the priority date and that is why we have to be careful to consider hindsight in our assessments. In that respect, it could in fact be argued that ChatGPT is closer to ‘us’ than it is the ‘skilled person’, although lacking in the ‘creative spark’ that human beings have.
Authors:
[1] https://www.epo.org/boards-of-appeal/decisions/pdf/t231193du1.pdf
[2] For an invention to be patentable it must be non-obvious to a person skilled in the art (at the point in time the invention was made)
[3] and therefore at present, can’t be an inventor named on patent
[4] Dr. Laurence Aitchison is a Lecturer in Machine Learning at the University of Bristol, where his work focuses on Artificial Intelligence (AI). He holds a PhD in Computational Neuroscience and Machine Learning from the Gatsby Computational Neuroscience Unit at UCL, alongside a Master’s and Bachelor’s degree from the University of Cambridge. His extensive research portfolio includes numerous publications on Large Language Models (LLMs) in the top AI conferences: ICML, ICLR and NeurIPS. He is the founder of PatentBee an AI patent drafting startup, focusing on the BioTech and Pharmaceutical sectors.
[5] To be granted as patentable, a patent application must “disclose the invention in a manner sufficiently clear and complete for the invention to be carried out by a person skilled in the art”. This comes from the TRIPS agreement, which sets out the minimum standard for IP rights in WTO countries. The idea stems from a core principle of patenting: that a proprietor gains a time-limited monopoly in exchange for disclosing the invention to the public.
[6] As I said previously, leaving the fact ChatGPT is a non-human aside.
[7] https://www.uspto.gov/subscription-center/2024/uspto-issues-inventorship-guidance-and-examples-ai-assisted-inventions
[8] https://www.federalregister.gov/documents/2024/02/13/2024-02623/inventorship-guidance-for-ai-assisted-inventions
[9] emphasis added by the author
[10] emphasis added by the author
[11] https://academic.oup.com/jiplp/article/20/4/195/7822599?login=false
Scintilla is Where Innovation Thrives
We put you in control of your destiny by organising, optimising, & monetising your intellectual property.