Obvious to Whom? Rethinking the Person Skilled in the Art in the Age of AI
The test for inventive step appears straightforward until a more fundamental question is asked: obvious to whom? For decades, patent law has answered that question through the legal fiction of the Person Skilled in the Art (PSITA). Yet, as scientific research increasingly relies on artificial intelligence, the assumptions underlying that benchmark have become less certain. This article examines whether one of patent law's oldest concepts is equipped to meet one of its newest challenges. It argues that the overlooked question is not simply whether the PSITA should be presumed to use AI-assisted research tools, but when such tools become part of ordinary scientific practice, and it proposes a principled framework for answering that question.
Introduction
Every patent system begins with a deceptively simple question: is this invention obvious? The question appears straightforward until one asks the next, far more difficult, question. Obvious to whom? The answer is certainly not the inventor, since an inventor may be exceptionally gifted, unusually fortunate, or simply persistent, none of which has much to do with the invention. Measuring inventiveness against the person who created it would make patentability depend on personal ability rather than the quality of what was made.
The Benchmark for Inventiveness
Patent law avoids that problem by relying on one of its most enduring legal fictions, the Person Skilled in the Art (PSITA). Instead of asking whether this inventor would have found the invention obvious, the law imagines an ordinary worker in the relevant technical field and asks whether that person would have reached the same result. What matters is not the inventor's brilliance, but whether the invention rises above ordinary technical skill.
Section 2(1)(j) of the Indian Patents Act, 1970 defines an "invention" as a new product or process involving an inventive step and capable of industrial application.1 Section 2(1)(ja) then defines an inventive step as a feature of an invention that involves technical advance as compared to the existing knowledge, or has economic significance, or both, and that makes the invention "not obvious to a person skilled in the art".2 The Supreme Court, likewise, in Bishwanath Prasad Radhey Shyam v. Hindustan Metal Industries,3 applied the standard of the person skilled in the art rather than that of the inventor. The PSITA is therefore the benchmark against which every claim of inventiveness succeeds or fails.
The same benchmark recurs well beyond India, though its label changes. In Graham v. John Deere Co.,4 the level of ordinary skill in the art became a central component of the obviousness inquiry, while KSR International Co. v. Teleflex Inc.5 emphasised that the skilled person is one of ordinary creativity, not an automaton. In England, the Court of Appeal in Pozzoli SpA v. BDMO SA,6 restating the structured inquiry first formulated in Windsurfing International Inc. v. Tabur Marine (Great Britain) Ltd.,7 required courts to identify the notional person skilled in the art together with the common general knowledge attributed to that person, viewing obviousness through the eyes of the normally skilled but unimaginative addressee.
What unites these formulations is a shared assumption that inventiveness can be measured against a stable, objective standard of ordinary skill. For most of the last century, that standard barely needed defending. Whether a chemist searched journals by hand or through an electronic database, the underlying method of research stayed within recognisably human limits, since the tool made the search faster without deciding what to look for or drawing any conclusions. It is precisely that relationship between the researcher and the tool that artificial intelligence (AI) now changes, and the existing literature on this shift has focused overwhelmingly on a different question, to which we turn next.
A Different Question from the One Most Often Asked
Most legal scholarship on AI and patents has concentrated on inventorship: whether an AI system can itself be named an inventor. That debate has largely been settled in the negative across major jurisdictions, with courts and patent offices consistently holding that conception requires a human mind, and that an AI system, however capable, cannot itself be an inventor.8 The 2025 revision of the USPTO's inventorship guidance for AI-assisted inventions confirmed this by discarding earlier joint-inventorship-style tests in favour of the ordinary human conception standard.9
That settlement, however, resolves the wrong question for most cases. The far more common scenario is not an AI-generated invention with no human inventor, but a human invention made with AI assistance, and here the unresolved question is not who invented it, but how obvious it was. A pharmaceutical scientist may run an AI-assisted literature review before opening a journal article; a structural engineer may use generative-design software to explore load-bearing configurations before a single physical prototype is built; a semiconductor firm may use AI-assisted layout tools to explore chip architectures that a human team would take months to search manually. In each case, the human researcher remains the inventor, and the invention remains entirely human throughout. Yet no settled rule tells a patent examiner, once they sit down to test the inventive step, how to answer one basic question.
Should the PSITA also be assumed to use those same tools?
If the answer is no, the law keeps testing inventions against a researcher who no longer resembles anyone actually working in the field, and the benchmark quietly drifts away from the practice it was meant to represent. If the answer is yes, the threshold for patentability rises, since work that once demanded real technical effort may now look obvious, not because the invention changed, but because the yardstick used to measure it grew more capable.
Neither answer is comfortable, and both rest on an assumption patent law has never openly examined: that the PSITA can evolve at all. The assumption itself is not unreasonable, since a legal fiction representing an ordinary researcher cannot stay frozen while research changes around it. Accepting that the benchmark can evolve, however, turns out to be the easy part. The harder part is answering a second question.
When should that presumption actually begin?
In 2024, the USPTO itself opened a public consultation asking precisely this question, namely how the proliferation of AI should affect the knowledge attributed to the person of ordinary skill in the art, and closed it without issuing binding guidance.10 The doctrinal vacuum this article addresses is therefore one that the world's largest patent office has identified but not yet filled.
When Does AI Become Ordinary?
Inventive step, under the Indian Patents Act and its counterparts elsewhere, is assessed as of the priority or filing date of the application, not as of whatever later date a dispute happens to reach a court or examiner.11 This forecloses an obvious worry, that a patent granted today could be struck down years later simply because AI has since become more capable.
Fixing the date, however, only tells us when to look. It does not tell us what to attribute to the PSITA once we are looking there. On the priority date of a given application, had a particular AI tool already become part of ordinary practice in that field, such that the PSITA should be presumed to have used it, or was the tool still novel enough that no ordinary researcher could realistically have relied on it? A tool may have been dependable in one field on that date and purely experimental in another, and it may have been technically available to everyone yet realistically affordable to only a few, so the same calendar date can produce different answers depending entirely on the tool and the field in question.
Towards a Principled Framework
The way out is to stop asking whether AI in general has become ordinary, and ask instead whether a specific tool has become ordinary within a specific field. In General Tire & Rubber Co. v. Firestone Tyre & Rubber Co.,12 the English Court of Appeal distinguished information that had genuinely entered the common stock of knowledge held by skilled practitioners from information that merely happened to exist somewhere within the prior art, holding that information does not become part of the skilled person's knowledge merely because it exists somewhere; rather, the relevant inquiry is whether it forms part of what a competent skilled addressee ought to know.
Patent law has therefore developed an approach for determining when knowledge becomes part of the ordinary equipment of the skilled person. It has, however, not developed an equivalent framework for determining when research tools become ordinary. The statutory language itself does not need to change to accommodate this gap, since the Indian Patents Act and its counterparts elsewhere already ask whether an invention is obvious to a person skilled in the art, and that language is broad enough to absorb an evolving account of what such a person's ordinary practice actually consists of. What follows is therefore simply an interpretive framework for applying that existing test.
The reasoning underlying General Tire13 can also inform the treatment of research tools. In deciding whether a tool has become ordinary in a field, a court could ask four questions: whether the tool is routinely used by practitioners generally, rather than by a handful of specialists; whether it is realistically accessible and affordable, rather than confined to a well-resourced few; whether it is professionally regarded as reliable, rather than experimental or unproven; and whether it is taught to newer entrants in the field as a matter of course, rather than treated as a niche skill.
Here, no single factor should decide the question alone. Taken together, however, these considerations let a court or examiner determine whether a research tool has become part of ordinary practice within the relevant field, evidenced in the same way as common general knowledge, through expert testimony, industry surveys, training curricula, and standard textbooks, rather than by its mere commercial availability.
Seen this way, the proposed framework is capable of operating across very different technical fields, and two examples make the point concrete. An AI literature tool used across the industry to organise publications and prioritise experiments would likely satisfy this test before a proprietary molecular prediction platform trained on one company's confidential data and licensed only within that company. The same distinction holds outside biology: a generative-design tool that has become a standard feature of mechanical and structural engineering curricula and practice would satisfy the test in that field long before a bespoke AI system built in-house by a single manufacturer solely for its own product line, even though both fall under the broad label of AI. In each case, both technologies qualify as AI, yet only one reflects how research is ordinarily conducted, because the PSITA has never been presumed to possess a competitor's exclusive technological advantages, and there is little reason why the emergence of AI should alter that principle.
The same reasoning explains why the benchmark cannot evolve at a single, uniform pace, since a tool indispensable in structural biology may be irrelevant in mechanical engineering, and the technologies attributed to the skilled person must evolve field by field, exactly as scientific practice itself does.
One qualification remains, and it is important enough to state expressly. Determining that an AI tool has become part of the ordinary equipment of the Person Skilled in the Art does not mean that every output generated by that tool is automatically obvious. Suppose an AI design system, now routinely used by structural engineers, generates two hundred possible bridge designs. Under the proposed framework, the skilled person is presumed to have access to that same set of possibilities because the tool itself has become ordinary. The inventive contribution, however, may still lie in recognising which of those designs was worth pursuing and demonstrating, through ordinary engineering analysis and testing, that it solved the technical problem where no ordinarily skilled person would have reasonably expected it to succeed.14 The framework proposed here therefore answers only the threshold question of what research tools are attributed to the skilled person. Whether the claimed invention would nevertheless have been obvious remains governed by the ordinary principles of inventive step, ensuring that attributing an AI tool to the PSITA does not, without more, predetermine the outcome of the obviousness inquiry.
Conclusion
Seen this way, AI is not forcing patent law to redefine inventiveness. Rather, it is forcing patent law to articulate when an evolving research technology becomes part of the benchmark against which inventiveness has always been measured, applying to research tools the same discipline it has long applied to common general knowledge. Until that happens field by field, the answer to the deceptively simple question with which every patent case begins, is this invention obvious, will keep depending on a harder one. Obvious to whom?
*****
Footnotes
1. The Patents Act, 1970, No. 39 of 1970, § 2(1)(j) (India).
2. The Patents Act, 1970, No. 39 of 1970, § 2(1)(ja) (India).
3. Bishwanath Prasad Radhey Shyam v. Hindustan Metal Indus., (1979) 2 SCC 511, 521 (India) (decided under the Indian Patents and Designs Act, 1911, but consistently applied under the 1970 Act).
4. Graham v. John Deere Co., 383 U.S. 1, 17 (1966).
5. KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398, 421 (2007).
6. Pozzoli SpA v. BDMO SA, [2007] EWCA Civ 588, [2007] FSR 37 (Eng.).
7. Windsurfing Int'l Inc. v. Tabur Marine (Gt. Brit.) Ltd., [1985] RPC 59, 73 (Eng.).
8. Thaler v. Vidal, 43 F.4th 1207, 1213 (Fed. Cir. 2022); Thaler v. Comptroller-Gen. of Patents, Designs & Trade Marks, [2023] UKSC 49, [2024] 1 WLR 1 (UK).
9. Revised Inventorship Guidance for AI-Assisted Inventions, 90 Fed. Reg. 54,636 (Nov. 28, 2025), https://www.federalregister.gov/documents/2025/11/28/2025-21457/revised-inventorship-guidance-for-ai-assisted-inventions.
10. Request for Comments Regarding the Impact of the Proliferation of Artificial Intelligence on Prior Art, the Knowledge of a Person Having Ordinary Skill in the Art, and Determinations of Patentability Made in View of the Foregoing, 89 Fed. Reg. 34,217 (Apr. 30, 2024), https://www.federalregister.gov/documents/2024/04/30/2024-08969/request-for-comments-regarding-the-impact-of-the-proliferation-of-artificial-intelligence-on-prior.
11. The Patents Act, 1970, No. 39 of 1970, §§ 2(1)(l), 13(1), 64(1)(f) (India); 35 U.S.C. § 103 (2018); Patents Act 1977, c. 37, §§ 2(2), 3 (UK).
12. General Tire & Rubber Co. v. Firestone Tyre & Rubber Co., [1972] RPC 457, 482 (Eng.).
13. Id.
14. See KSR, 550 U.S. at 421 (treating a solution as obvious where the known options are finite and predictable and the anticipated success follows, which implies the converse where neither condition holds).