The Curious Physician: Nurturing Wonder

The explorer and the designer both rest on something prior to method and prior to infrastructure. An explorer must first find something worth exploring, and a designer must first notice that a thing is badly made. Behind both stands a single motivating disposition, and it can be cultivated or extinguished by the way physicians are trained.

Curiosity is the active desire to know, an appetite for information, understanding, or novel experience that pulls a person toward what they do not yet grasp. The quality most tightly bound up with it, and the one discovery depends on most, is wonder: the capacity to be struck by something as puzzling or surprising in the first place. Wonder supplies the initial jolt; curiosity supplies the sustained drive to resolve it. Without wonder, nothing registers as a question. Without curiosity, the question dies unexplored. Discovery lives in the space between them.

The two great traditions approach it differently, and both are relevant to medicine. Philosophers have asked whether curiosity is a virtue or a vice, setting Augustine's curiositas as restless appetite against Aristotle's opening claim that all human beings by nature desire to know. The modern settlement treats it as an epistemic virtue disciplined by purpose, which is the distinction between being curious about a patient and being curious at their expense.

Psychologists ask instead how it works and what it produces: Berlyne's separation of perceptual from epistemic curiosity; Loewenstein's information-gap account, in which curiosity is the felt discomfort of a recognized hole in one's knowledge; and the now substantial evidence linking curiosity to memory, learning, and persistence.³⁷

Precedent in Medicine

The phrase is borrowed. Bugaj, Schwarz, and colleagues have argued for nurturing wonder as an explicit aim of medical formation, not a temperament some students happen to bring. A related literature documents the same erosion curve that empathy shows, with curiosity declining across training as the volume of required content rises and the tolerance for open questions falls. Dyche and Epstein took up the question a decade later, and a small literature has followed, including a medical curiosity scale.³⁸ This body of work remains thin relative to the claims made for it.

Curiosity in the physician must be deliberately unleashed, because healthcare as currently organized erodes it, and nothing about that erosion is intentional. It is the aggregate of arrangements each defensible alone: visit lengths that price the follow-up question; documentation that consumes the margin in which a question might be pursued; protocols that answer before wondering can begin. Curiosity and caring are eroded by the same mechanism, and the same remedies restore both.

Machine systems are incurious by construction, because nothing surprises a system that holds no expectations of its own. As machines take over the answering of questions, the asking of them becomes the scarce and valuable act.

Curiosity and caring both require unbilled time spent in the presence of something unresolved: a question that has not yet become a diagnosis, or a person whose distress has not yet become a task. Neither produces anything countable in the interval, which is why both are the first casualties when throughput is optimized, and why neither is restored by instructing clinicians to try harder.

What restores them is the same in both cases: time the schedule protects rather than time that happens to be left over. Chapter IX makes the same argument on the caring side. The profession is defending one thing rather than two.

The profession must allow that wonder is a legitimate thing for a physician to feel and to say.

Institutions around the world are already rebuilding the front end of medical training around discovery, and their experiments show how curiosity gets cultivated instead of assumed.

A systematic case comes from Barry Coller at Rockefeller University, whose argument is that translational research is a discipline with teachable content and not a temperament some physicians happen to have. He defines it as applying the scientific method to a defined health need, and names three cardinal skills: defining that need with the precision a basic scientist brings to a hypothesis; designing an assay reductionist enough to be tractable yet real enough biologically that a positive result will extrapolate to humans; and conceptualizing, from the outset, a realistic path to regulatory approval or clinical adoption.

With Robert Califf he set out a framework for judging whether a discovery can survive the passage to therapy before heavy investment is committed, and he has more recently argued for putting questions of global access to a technology transfer office at the point of invention rather than after it.³⁹

Coller built machinery to match the argument. Clinical Scholars is a three-year experiential master's in which trainees design, conduct, and report their own human-subjects protocols under senior mentorship, wrapped in biostatistics, ethics and regulatory practice, team science, and leadership.

Alongside it runs Translational Research Navigation, in which translational science experts walk investigators, and especially basic scientists and trainees, through protocol development to a high standard, with a documented educational effect: trainees need progressively less support as they go. The ambition, pursued across successive awards, was a learning clinical research enterprise that improves on its own outcome data and can be shared through the national consortium.

Harvard Medical School has approached the same question from the opposite end. Where Rockefeller teaches physicians to do research, the Artificial Intelligence in Medicine doctoral track, launched in 2024 in the Department of Biomedical Informatics under Isaac Kohane and Sebastian Schneeweiss, recruits from computer science, engineering, and mathematics and teaches those students medicine. Demand settled the question of appetite: more than four hundred applications for seven places.

Two design decisions carry the pedagogy. Every student takes required clinical rotations at affiliated hospitals alongside the coursework, so the clinical problem is encountered rather than described. And every student is paired, by the end of the second year, with two mentors, one technical and one hospital-based, which institutionalizes the translation that most such programs leave to chance. Coller's trainees learn to ask a research question; these learn what the question is for. Both are answers to the same problem, which is that research capability is transmitted by apprenticeship and not by curriculum.

Several institutions have reached the same conclusion by making research a structural requirement of the degree instead of an opportunity within it. At the University of Maryland School of Medicine, E. Albert Reece as dean required every medical student to train in biomedical research methods, reasoning that a physician who has never designed a study cannot properly read one. Israel's newest medical school, the Gutwirth school at the Weizmann Institute of Science, goes furthest: every student is enrolled in a combined MD-PhD pathway, most of the time is given to practice and research rather than lectures, and doctoral work may be taken in physics, chemistry, mathematics, or computer science as readily as in the life sciences.

Others have worked on the pipeline itself. At the Hebrew University of Jerusalem and Hadassah Medical Center, Dina Ben-Yehuda as dean opened tracks aimed at students already versed in computer science, and now leads the faculty's center for computational medicine.

At Yale, under Nancy Brown as dean, the same conviction took structural form: in 2024 the school elevated its section of biomedical informatics and data science to a full department under Lucila Ohno-Machado. The choice is consequential, since a department carries faculty lines, promotion pathways, and a seat in governance where a center carries none of them. Around it sit a certificate in medical software and medical AI and a master's degree admitting computer scientists alongside clinicians. Yale's medical students moved in the same direction unprompted, founding an AI in medicine interest group whose organizer put the case plainly: understanding artificial intelligence, even at a basic level, is becoming a necessary competency for the 21st century physician.

Japan has approached the problem through the culture of training, not its structure. At Kyoto University, work associated with Hitomi Kataoka and Nagahiro Minato has pressed on the formative dimensions of medical education, including empathy, professional identity, and the conditions under which they survive training. The university's Center for Medical Education and Internationalization has made international exposure and research engagement part of how students are formed rather than an elective bolted on. What these programs share is a conviction that curiosity is not preserved by exhortation but by putting learners early, and repeatedly, in rooms where the answer is not yet known.

Each of these designs costs something. The universal MD-PhD at Weizmann buys depth at the price of selection, since a school admitting only students willing to commit to doctoral work will not admit some of the people who become the best clinicians, and the model does not scale to a system that must produce a workforce. A technically selected intake buys fluency at the price of breadth, since a cohort recruited from computer science is a cohort not recruited from the humanities, and Chapter IX argues that the second matters as much as the first.

A universal research requirement buys exposure at the price of clinical time, and it is the design most vulnerable to the tokenism named earlier, since a requirement applied to everyone is a requirement someone must supervise. None of this argues against any of the three. It argues for choosing deliberately, and for knowing which of the goods on offer a given institution most needs.

One provocative instrument follows: tooling for personalized, curiosity-driven exploration, with curiosity vectors that follow a learner's prior interests across domains, surfacing the adjacent unfamiliar.

Creation, however, is only one of the four identities, and the next concerns the physician as bridge, not as discoverer.