During a recent seminar at the Brooklyn Tech Expo, Professor David Lefer of NYU Tandon School of Engineering assessed the fundamentals of AI innovation from a direction most patent practitioners have not considered. While acknowledging the strengths of what AI can do, Professor Lefer identified what AI clearly lacks: curiosity.
Lefer drew an analogy between modern AI systems and Koko the Gorilla. Koko demonstrated remarkable communication and mimicry capabilities, but did not exhibit the kind of self-driven inquiry that characterizes human cognition. AI systems similarly excel at generating responses from patterns and data. They do not possess epistemic curiosity—the drive to ask questions because understanding itself has value.
That distinction maps surprisingly well onto a core requirement of patent law.
Conception Requires a Mental Act
Under U.S. patent law, conception has long been the cornerstone of inventorship. The Federal Circuit explained in Burroughs Wellcome Co. v. Barr Labs., Inc., 40 F.3d 1223 (Fed. Cir. 1994) that conception is "the formation in the mind of the inventor, of a definite and permanent idea of the complete and operative invention." The emphasis on the mental act of conception is deliberate. Conception requires more than generating useful outputs—it requires recognizing a problem, understanding the significance of a potential solution, and arriving at a sufficiently definite inventive concept.
The USPTO's Revised Inventorship Guidance for AI-Assisted Inventions, published in the Federal Register on November 28, 2025 (90 FR 54636), reaffirmed this principle with clarity. Issued under Executive Order 14179 of January 23, 2025 ("Removing Barriers to American Leadership in Artificial Intelligence"), the new guidance rescinds the February 2024 guidance in full and replaces it with a streamlined framework that applies the traditional conception standard uniformly, with no separate or modified rules for AI-assisted inventions. The USPTO states that "AI systems, including generative AI and other computational models, are instruments used by human inventors," analogous to "laboratory equipment, computer software, research databases, or any other tool that assists in the inventive process." The framing is doctrinal rather than rhetorical: AI is not a category of contributor whose role must be separately measured, but an instrument whose use does not alter the inventorship inquiry.
As practitioners have assessed, the 2025 guidance is unambiguous: invention remains a human venture, and inventorship is limited to natural persons regardless of how substantially AI contributed to the development process.
As other practitioners have observed, the guidance also withdraws Pannu-factor analysis in single-human-plus-AI scenarios. The USPTO's reasoning is direct: Pannu addresses joint inventorship among multiple natural persons and is inapplicable when only one natural person is involved in developing an invention with AI assistance, because AI systems are not persons and therefore cannot be "joint inventors." Pannu remains operative for joint inventorship among multiple humans, regardless of whether AI tools were used in the development process.
The guidance is comprehensive in scope, applying not only to utility patents but also to design and plant patents and applications. The guidance also imposes specific requirements on priority claims: a U.S. patent application cannot claim priority to a foreign patent application naming only an AI as inventor. For foreign applications that list both a natural person and a non-natural person as joint inventors, the U.S. application data sheet must list only the natural person(s), including at least one in common with the foreign application. Companies with global portfolios and their practitioners must ensure inventorship and applicant designations align with AI-assisted innovation practices across jurisdictions.
Why Curiosity Illuminates the Legal Standard
Professor Lefer identified several forms of curiosity relevant to innovation, including empathic curiosity, diversity curiosity, and "need for cognition"—the desire to think deeply and challenge assumptions. The latter holds particular significance for patent law.
Inventive activity frequently begins with a form of intellectual dissatisfaction: noticing an inefficiency, questioning a conventional assumption, or asking why a persistent problem has not been solved. The subsequent process of testing, refinement, and iteration reflects exactly the kind of deep, reflective inquiry Lefer described.
Patent conception similarly requires more than arriving at an answer. It requires recognizing that a problem exists, formulating it in a way that reveals its inventive potential, and exercising judgment about why a particular solution matters. Current AI systems can propose technically plausible candidates at remarkable speed. They do not independently identify which hidden problems are worth investigating or become intellectually unsettled by unanswered questions. That evaluative, question-driven process remains a distinctly human contribution.
The analogy is illustrative rather than doctrinal. The U.S. standard for conception does not reference curiosity as a legal element, and court decisions will not turn on whether an inventor was sufficiently curious. The narrower point is that the cognitive work the law protects under the conception standard—recognizing a problem, formulating it productively, and exercising judgment about a solution—maps closely onto the kind of self-directed inquiry that current AI architectures do not perform.
The Practical Documentation Imperative
As generative AI tools integrate into R&D workflows, inventorship disputes increasingly turn on the quality of documentation showing the human role. Following the November 2025 guidance, practitioners have highlighted the value of contemporaneous records that demonstrate how the human inventor directed and shaped the process. Useful documentation may include:
- The original technical question or problem statement that initiated the inventive process
- Human-generated hypotheses, assumptions, or constraints formulated before AI tools were engaged
- Iterative refinements following AI outputs, including which outputs were selected or discarded and the rationale
- Decision-making records explaining why one technical path was chosen over alternatives
- Evidence showing why particular results were viewed as meaningful and worth pursuing further
These records help establish that the human inventor formed the "definite and permanent idea" required under Burroughs Wellcome. The operative question is rarely whether AI participated, but rather how the human inventor identified the problem worth solving, prompted the AI tool to develop solutions, and exercised judgment over the resulting outputs.
Implications for Obviousness
The 2025 guidance provides welcome clarity on inventorship while leaving other patent law questions open. The USPTO's 2024 Request for Comments on the Impact of AI on Prior Art and PHOSITA sought input on whether widespread AI tool access should raise the baseline knowledge attributed to the hypothetical person having ordinary skill in the art (PHOSITA) for obviousness analysis. That question remains unresolved and carries sharper practical stakes than the inventorship debate. If AI capabilities are imputed to the PHOSITA, the universe of combinations a skilled artisan would find "obvious to try" expands substantially, and inventions that survive obviousness analysis today may not survive one tomorrow.
Curiosity does not resolve the obviousness question doctrinally, but suggests where the line may eventually be drawn. On one side, combinations and outputs an AI can readily surface from prior art would become presumptively obvious if AI capabilities are imputed to the PHOSITA. On the other, the human judgment about which AI-generated possibilities are worth pursuing and reducing to practice—which is harder to mechanize—may continue to anchor non-obviousness.
A Global Consensus on Human Inventorship
The requirement that inventors be natural persons is now settled across major jurisdictions. The UK Supreme Court confirmed this in Thaler v Comptroller-General of Patents, Designs and Trade Marks [2023] UKSC 49 (Dec. 20, 2023). The Federal Circuit reached the same conclusion in Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. Aug. 5, 2022). The EPO's Legal Board of Appeal did likewise in J 8/20 (Dec. 21, 2021), with subsequent consistent rejections. In each forum, the same plaintiff, the same AI system, and the same result: only natural persons may be named as inventors.
As the Congressional Research Service's analysis has noted, the USPTO's position is that ordinary conception standards apply uniformly, regardless of AI involvement in the inventive process. While AI systems may someday simulate inventive curiosity, the law has defined inventorship around a human mental act that current AI architectures do not perform.
The most valuable contribution to an AI-assisted invention thus remains the one machines still cannot replicate: the ability to ask the question that nobody else thought to ask—and to recognize why the answer matters.