About this report — and this site

Artificial intelligence, advanced robotics, and a rapidly expanding biomedical frontier are already reshaping how physicians are trained and how they practice. If present trajectories hold, the quarter century before us will not be a period of gradual evolution. It will be a mosaic, with unfamiliar elements creeping quietly into view until, at some inflection point, the whole picture changes.

This report therefore asks an ambitious question. As human + humanoid medicine takes shape, what becomes of the human physician, and how should the profession, and the institutions that educate physicians, prepare?

The answer offered here identifies and contextualizes promising directions for thought and action. It is not a definitive reference. It is a structured synthesis of insights from a transdisciplinary group of thought leaders, drawn from within biomedicine and from far afield of it. The work is organized across three dimensions: the emerging reality of biomedicine and care; the redefinition of the physician; and the transition that carries us from the one to the other.

The findings are a snapshot in time of informed judgment, offered to galvanize conversation and to invite rapid iteration.

This report operates deliberately at a thirty-thousand-foot level. It assumes developed-world access to care, adopts a global rather than a country-centric lens, and looks toward 2050 and not beyond, limiting itself to a single generational leap.

Financial and regulatory constraints, while real, are intentionally bracketed here so as not to narrow the visioning prematurely. Health equity and the opportunity to serve low- and middle-income countries are essential, and they are flagged throughout for dedicated follow-on work.

The report is written for physicians, and for the healthcare leaders and organizations responsible for healthcare delivery and medical education. It is a starting point for discussion among these communities, not prescriptive guidance.

The Future Physician Initiative comprises workgroup participants and advisors from leading academic and research organizations worldwide, among them: Biovista, Inc.; Champalimaud Foundation; German Aerospace Center / Institute of Robotics and Mechatronics; Hebrew University of Jerusalem / Faculty of Medicine; Helmholtz Association of German Research Centres; Kyoto University / Center for Medical Education and Internationalization, and Institute for Advanced Study; The Lown Institute; Massachusetts General Hospital / Harvard Medical School; New York Academy of Medicine; Osaka Metropolitan University / Graduate School of Medicine; Rockefeller University; Santa Fe Institute; Sheba Medical Center / MSR Israel Center for Medical Simulation; Stanford University / School of Medicine; Tel Aviv University / Edmond Safra Bioinformatics Center; The Systems Biology Institute / RIKEN Center for Integrative Medical Sciences; Thomas Jefferson University / Sidney Kimmel Medical College and Computational Medicine Center; University of Maryland / School of Medicine; University of New Mexico / School of Medicine; and Yale University / School of Medicine, and School of Nursing.

Mark L. Tykocinski, MD, established the Future Physician Initiative in early 2025 and serves as its lead organizer. Priscilla Markwood is the administrative lead.

A working group was assembled under the Future Physician Initiative banner, drawn from the institutions named above. Early conversations among those who joined at the outset were exploratory and agenda-setting: they established the questions worth asking and shaped much of what followed. Three subgroups each held a series of online meetings to brainstorm the future of biomedicine and healthcare, and the projected impact of biomedical innovation and disruptive technologies on the physician of the mid-twenty-first century. Those discussions, distilled by Angela Song-Ruiter and supplemented with additional written input from the working group, informed the drafting of this report. Mark Tykocinski, as Lead of the working group and principal author, drafted and shaped the document, drawing upon the rich discussions and shared ideas developed collectively by the working group over the course of its meetings. After further iterative conceptual and textual refinement by the working group offline, the report was finalized at a workshop hosted at the Santa Fe Institute in early November 2026.

The report has been supported by a generous gift from Dr. Yoshihara Asano of Tokyo, Japan.

Cover image: Tobi Kahn, from the Dancers series.

One word recurs across every chapter, and it emerges as a cross-cutting meta-theme: convergence.

Trends, disciplines, and capabilities that developed apart are being knitted into one reconfigured field. Given convergence, the object a physician must hold is no longer divisible along the lines the profession inherited. Many structures built on those lines, including the specialty, the department, the journal, the funding stream, and the certifying board, now tie to a field that has moved. The chapters touch on what converges in each case, and points to what the profession would have to do about it.

In Chapter I, elements of three earlier epochs of physicianship converge in setting the stage for a fourth, the collective of experts, and the physician's many past and projected personas converge into one fluid, plural identity.

In Chapter II, knowledge domains converge to generate new biologies, and in turn, new biologies themselves converge with one another to spawn derivative fields.

In Chapter III, advanced technologies and emerging social determinants converge in pioneering new modes of medicine, and in turn, these converge into an interlaced constellation of healthcare trends, with the learning health system as the mechanism by which that constellation keeps rewriting itself.

In Chapter IV, streams of machine intelligence converge into an ecosystem of interoperating systems that underpin a new era of data-aware medicine.

In Chapter V, data, mechanism, and the individual patient converge on N-of-1 care.

In Chapter VI, human and machine intelligence converge without merging, with the human advantage carving out its niche where there are irreducible gaps.

In Chapter VII, discovery and practice converge in the physician creative, and the instrument and the question converge in technoscience, so discovery stops being something done upstream by other people. The technoscientific inclination suffuses all of physicianship.

In Chapter VIII, convergence assume center stage, as consilience is the very definition of what convergence is all about. Specialties converge and reassemble. Healthcare professionals converge and distribute intelligence. The physician embraces knowledge domains far afield from biomedicine.

In Chapter IX, a spectrum of human capabilities and qualities converge in reifying the transcendent physician.

In Chapter X, individual and population health converge, the local converges with the global, and the global physician assimilates all the other identities.

In Chapter XI, three competency cores converge into foundation elements of a new medical curriculum, the phases of training converge into a lifelong continuum.

In Chapter XII, the case is made for the institutions that control the profession to converge on shared agendas—in discovery and in formation alike.

Restating the consolidated vision set forth earlier in the report: The physician of the mid-twenty-first century is a member of a collective of experts, holding four fluid identities overlaid on specialty, working at the human plane of an intelligence ecosystem, filling the irreducible gaps that no machine closes, and formed continuously across a career rather than minted at graduation.

Beneath that vision sit eight meta-themes, each a tension to be managed rather than a problem to be solved: human meaning-making against machine computation; distributed intelligence against individual authority; systems-level stewardship against episodic intervention; fluid adaptive identities against fixed professional roles; transdisciplinary convergence against siloed expertise; human compassion and ethics becoming more important rather than less as machines grow capable; continuous learning systems against static expertise; and the orchestrator-interpreter-integrator against the sole expert.

Stating the eight as tensions is correct, and it leaves open which way to lean in practice. One principle supplies the answer. In every pair, the second term is what the system produces on its own. Machine computation, individual authority, episodic intervention, fixed roles, siloed expertise, efficiency at the expense of compassion, static expertise, and the sole expert are all defaults, arrived at by drift rather than by decision. The first term in each pair survives only where someone designs, funds, and protects it. What is distinctively human in medicine does not persist by default, and everything this report recommends is an attempt to make it persist by design.

Automation does not remove pieces of the human contribution. It concentrates that contribution by removing everything that can be specified, so a smaller share of the work carries a larger share of its consequence. The first term of each pair is the concentrate rather than the residue. This is why the report's optimism and its urgency are one claim. The physician's value rises as machine capability rises, and it rises only for a profession that has prepared for what it is being left with.

Convergence is the spine of the argument. Four further meta-themes run the length of the report, and we take each in turn.

Transdisciplinarity

The problems that matter no longer sit inside any one discipline, so they cannot be solved from inside one. Convergence science, as the National Research Council defined it, calls for shared facilities, joint appointments, and funding that does not force a project to declare a disciplinary home. That is institutional redesign, and goodwill will not substitute for it. The Harvard-MIT program in health sciences and technology draws engineers and computer scientists into medicine by deepening their own discipline. Stanford's Biodesign puts physicians, engineers, designers, and business students on a single team, and measures the result in companies founded and patients reached. Every one of the eight population health frontiers of Chapter X needs at least three disciplines simply to state the problem correctly.

The demand is not for polymaths. The sociology of expertise distinguishes contributory expertise, the tacit skill needed to do a specialty's work, from interactional expertise, full fluency in a domain's language without the ability to practice it. Interactional expertise costs a fraction as much to acquire, and most of what this report asks of the physician is interactional expertise held across many domains at once. The operative skill is recognition. Recognition means knowing to which domain a problem belongs, which is frequently not the one it presents as, and then acquiring what is needed. It is a smaller skill than fluency and a rarer one, because it depends on having seen enough problems misfiled to know how misfiling happens.

The claim has a critical institutional edge. Transdisciplinarity fails in practice for a reason seldom articulated. Investigators choose collaborators by confirmation bias, partnering with people whose assumptions they already share, so the team is convergent on paper and homogeneous in practice. It fails again at the review panel, since convergent work needs reviewers competent to judge work belonging to neither field, which is a staffing problem before it is a policy one. It fails a third time in the curriculum, where interprofessional education stops at the health professions and thereby reifies half the argument. What is called for is trans-professional formation that places physicians in real working interdependence with engineers, designers, lawyers, economists, and humanists. The return is counted in decades rather than in course evaluations, because a cohort formed across professions in its twenties is still working across them in its fifties.

Complexity

We treat complexity science as an instrument of knowing before an instrument of predicting. It also names what the physician becomes—a decipherer, navigator, and engineer of complex adaptive systems, from fundamental biological mechanism through pathogenesis, clinical diagnostics and therapeutics, the modes of medicine, healthcare system architecture, and the interfaces of care with patients, communities, and global entities. The three verbs form a sequence. Deciphering is reading a system well enough to say what kind of system it is and which regime it occupies. Navigating is acting inside it while the reading stays permanently incomplete, so timing, reversibility, and the length of the feedback loop matter more than the precision of any estimate. Engineering is shaping it. Physicians already do all three, unnamed and untaught, and naming them is what makes them teachable.

Complexity science tells a physician, in advance, when linear, single-cause, average-based reasoning is about to fail. The strongest instances invert a clinical intuition. Read disease as a perturbation of a network and multimorbidity becomes a single object with a shape. Goldberger's fractal physiology makes variability itself a vital sign and turns the orderly trace into a finding, which also means that nearly every monitor now in use cannot see it. Gatenby's adaptive therapy steers a tumor instead of killing it maximally, so dose stops being a number and becomes a control policy, and antimicrobial resistance is the same logic at the scale of a population. Beneath these sits our most consequential epistemic claim, the distinction between what a system can be predicted to do and what our models achieve. More data closes the second gap and leaves the first exactly where it was. That is the exact sense in which complexity sets a permanent ceiling on prediction, and the reason judgment has a permanent home.

The claim becomes clinically usable only when it is disaggregated. Unpredictability arises for four distinct reasons: inherent dynamical instability, prohibitive measurement cost, computational intractability even given unlimited data, and a state space that is itself unknown because the system adapts, evolves, or innovates. Each implies a different response, from shorter feedback loops to triage of measurement to honest abstention to simply asking the patient. The fourth class is the deepest and defines the physician's territory, since a tumor evolves, a pathogen emerges, and a person's circumstances, values, and goals change, so no model can assign probability to an outcome its state space does not contain. Telling these apart is the single most useful competence complexity science contributes to practice. The two errors are symmetrical and both common. Mistaking irreducible uncertainty for remediable ignorance produces over-testing and false precision, and mistaking remediable ignorance for irreducible uncertainty produces premature fatalism. Neither is currently taught.

Two corollaries follow. Krakauer's emergent engineering supplies the design posture, shaping selective context and distributions of outcome rather than specifying end states. It asks medicine to stop treating complete cure, eliminated risk, perfect stability, and total security as descriptions of what treatment does. Medicine moves a patient into a space of desirable though never optimal performance, and it then keeps them there while the patient and the world go on changing. Lloyd supplies the discipline, since the model that minimizes total information is the maximum-likelihood model, the most concise one that makes the best possible predictions given the data. That makes parsimony a criterion of correctness and deciding what not to measure a clinical skill. Lloyd's point about natural scales adds the constructive half, because finer and more frequent sampling can reach a scale at which a system that was uncontrollable becomes controllable. That is the strongest principled argument available for continuous monitoring and for the biomedical digital twin, and equally their discipline, since data gathered at the wrong scale buys nothing at all.

Creation

We advocate for discovery to be an expectation of the profession rather than a territory reserved for a physician-scientist elite. Creation is a property of systems rather than of exceptional individuals, and the six reforms of the research ecosystem in Chapter VII aim to make physician creativity structurally possible. Curiosity turns out to be an environmental variable managed across a career rather than a trait to select for at admission, and the industrialization that erodes caring erodes curiosity too. As machines take over the answering of questions, the asking of them becomes the scarce and valuable act. Machine systems are incurious by construction, because nothing can surprise a system that holds no expectations of its own. Every question a machine pursues was posed by a person. Deciding what is worth asking, and what is worth wanting, is the deepest of the irreducible gaps.

Two developments make the claim more than aspirational. The barrier to entry has moved. What AI absorbs is the part of discovery that used to require an institution, the apparatus, the cohort, the core facility, and the years, while what it cannot supply is the question. The constraint shifts from access to instruments toward access to data and the capacity to frame a question worth putting to it, and a practicing physician holds both in unusual measure. Cheap answers therefore raise the value of good questions rather than lowering it, and they raise it most for the person standing where the problems arise. Creation also has a second and consistently undervalued half. The physician as designer works where the binding constraint is arrangement rather than knowledge, and design breakthroughs in medicine have historically come from noticing an arrangement everyone had stopped seeing, demanding no sophisticated technology at all.

Two failure modes attend the claim that every physician should be a creative. Tokenism is the required scholarly project nobody supervises, assessed on completion rather than on quality, which teaches trainees that research is a hurdle to be cleared. The unfunded mandate is research output expected of clinicians whose schedules contain no protected time in which to produce it, which is a tax rather than a training model. The remedy for both is the same, and exhortation is no part of it: protected time, real mentorship, methodological instruction that precedes the requirement to produce, and honest breadth in what counts, so that the physician whose contribution is a redesigned pathway or a piece of qualitative work is recognized as having done the thing. Beyond the individual, the barrier is less capability and almost always standing. Research culture decides whether part-time, late-entering, clinically grounded contributions count as normal science or as a hobby indulged, and that decision determines whether any of this reaches the person the argument is about.

Virtue

The underlying question is what the physician is for, and our answer is that value migrates toward whatever cannot be delegated. Some of it cannot be delegated at all: answerability, first-person standing, and the capacity to bear the cost of being wrong. The sharpest formulation we can offer is that every résumé virtue of medicine is in principle machine-attainable, and not one eulogy virtue is. What remains is medical virtue, with the array of human qualities it signifies.

Naming it virtue rather than listing qualities is deliberate. A list invites the reading that these are traits some clinicians happen to possess, pleasant when present and forgivable when absent. Virtue, derived from the nature and ends of medicine rather than imported from outside it, makes the opposite claim. A physician who lacks fidelity to trust, compassion, prudence, justice, courage, temperance and self-effacement, or integrity is less competent, not merely less pleasant, because they cannot reliably achieve what medicine is for. That is why this report treats character as the profession's differentiated asset, and why virtue formation becomes the core business of medical education rather than an adornment of it. It is also why empathy is the wrong ground on which to stake the human claim. Empathy became the organizing term for the human dimension largely because it was the part that could be measured, and it is a perceptual capacity rather than a moral and relational condition, and of all these qualities the one machines simulate best.

Virtues are formed by habituation, and habituation is a property of environments. A system that prices attention will not produce attentive physicians however well it teaches attentiveness. That places a condition on everything Part III proposes, since formation depends on clinical environments, and a dean should know which battles outside the school decide whether the work inside it survives contact with the wards. The same mechanism explains why curiosity and caring rise and fall together. Both require unbilled time spent in the presence of something unresolved, neither produces anything countable in the interval, and both are restored by time the schedule protects rather than time that happens to be left over.

If convergence is the diagnosis, fragmentation is the obstacle, and it must be addressed before any of the preceding recommendations can take hold.

The current health ecosystem is fragmented at every level at once. Knowledge domains are siloed, so a breakthrough in one field reaches an adjacent one years late or not at all. Technologies are disparate and non-interoperable, so a well-integrated whole loses to a collection of brilliant but isolated tools. Medical specialties and subspecialties have hardened into fiefdoms that no longer track the underlying biology. Institutions compete where they should pool. Data is scattered across systems that cannot exchange it without loss. And the bodies that govern the profession each hold a partial veto and no comprehensive mandate.

The remedy is demonstrated rather than hypothetical. Standing consortia in Japan, the United States, and Europe, with and without industry at the table, show that cooperation without merger works. They share a four-part specification worth applying to every proposal in this report before it launches: a federation rather than a merger, an explicit neutral convener no participant controls, a prior definition of what is shared and what is not, and a commitment to publish. What that test mostly reveals is which proposals have not yet decided who convenes them.

Five priorities follow for those who lead health systems, academic medical centers, societies, and the institutions that train physicians.

• Design for the human factor, deliberately. If compassion, meaning, and ethical judgment are the physician's durable contribution, they cannot be left to chance. Build authentic caring into the architecture of care, through protected time, relational continuity, and roles defined around connection. Virtue is formed by habituation and habituation is a property of environments, so the learning environment is a curricular instrument whether or not anyone designs it as one.
• Re-empower the physician over technology. Equip physicians to evaluate, validate, and govern the AI and robotic systems they work alongside, including the authority and the literacy to overrule them. Govern the substrate as well as the output, since data is layered and a system validated only where it is observable can be sound at the surface and unsound three layers down. Set the consequence threshold above which a system whose reasoning cannot be inspected may inform a decision but may not carry it.
• Make scientific reasoning a universal competency. Close the widening gap between accelerating discovery and everyday practice by teaching research, critical thinking, and complexity science to every physician rather than to a physician-scientist elite. Teach complexity as a working competence rather than as an outlook, so a clinician can name which kind of unpredictability is in front of them. Teach the insight case, the reframing of a badly represented problem, as deliberately as medicine has always taught the search case.
• Build the team, flatten the hierarchy. Plan explicitly for horizontal, networked care in which expertise distributes across an expanding set of professionals and the patient is an active partner, and invest in the integrator and orchestrator roles that hold a distributed system together. Plan also for the professions that do not yet formally exist. AI mediation, digital health stewardship, longitudinal navigation, and virtual care coordination are already being done badly by people whose job it is not, and the physician's relation to those professions will be settled early and then be difficult to revisit.
• Treat education reform, and the rate of absorption, as the decisive lever. Every other priority routes through how physicians are formed and how fast systems adopt what they are given, so manage adoption as deliberately as invention. Protect the enduring human core first when resources contract, since it is the only layer of the curriculum with no constituency of its own. Resource longitudinal mentoring and simulation as infrastructure rather than as goodwill, since they are the two mechanisms by which everything else in this report is transmitted.

Each priority should carry an indicator, on the discipline Chapter VII demanded of the research enterprise, so progress can be judged rather than asserted. For the biomedical frontier, the interval between a clinician noticing something and running a first experiment. For technology governance, the proportion of clinical AI procurement decisions in which a practicing physician held a vote rather than an opinion. For discovery as a universal competence, the share of physicians who contribute to one published or implemented advance within five years of training. For the team, the proportion of care episodes with a named integrator answerable across the settings involved. For education, the interval between a capability entering practice and entering the required curriculum. We would add a sixth, which bears on the human factor most directly and is the most conspicuously unmeasured of all: the proportion of a physician's working week that is protected time in the presence of an unresolved question or an unresolved person. None of these is measured today, all of them could be, and a reader should be able to return in five years and determine whether anything happened.

Above all, physicians themselves must be proactive. The profession can allow the physician role to be decomposed by technology and workflow redesign, or it can reconstitute the physician around the capacities future medicine most needs. Nothing recommended here requires permission from a body that does not yet agree. The mechanism that carried comparable change in engineering, aviation, and architecture is available now: a small coalition of credible institutions demonstrates an alternative, measures it publicly, and lets the accrediting and certifying bodies follow the evidence.

Choose the latter, and the human factor will not simply persist. It will become more visible, more intentional, and more consequential than ever.

This report began with a fifty-year-old drumbeat predicting the physician's obsolescence, and it ends with the contrary conviction that the physician's future is bright because it is so different. Success will be measured by whether the relevant communities convene quickly enough that the directions offered here are superseded by sharper ones, developed through exactly the transdisciplinary collaboration this report has argued for throughout.

This report is a visioning synthesis, organized in three movements: the emerging reality of biomedicine and care; the redefinition of the physician around four identities; and the transition in education and institutions that carries us from one to the other.

Its inputs were the deliberations of a transdisciplinary working group drawn from within and far outside biomedicine, structured around the evolution of the physician; the trajectory of AI, advanced robotics, statistics and mathematics, and complexity science; and the frontier of biomedical discovery and the human meaning of care. These were distilled into cross-cutting themes, recurring structural shifts, hardwiring concepts for education, and the meta-themes that provide the analytical backbone above.

Guiding assumptions. The analysis observes several deliberate constraints. It keeps a thirty-thousand-foot view instead of drilling into any single domain. It imagines what is achievable by 2050, roughly one generation, without venturing further. It takes a global, apolitical perspective and begins from developed-world access to care, while recognizing that health equity and low- and middle-income countries deserve dedicated treatment. And it holds financial and regulatory constraints to one side, so that what medicine could become is explored on its own terms.

On the nature of this report. It is best understood not as a fixed plan but as a snapshot of informed judgment at a moment of rapid change, offered to galvanize work. The roles, scenarios, and moonshot ideas here are illustrative, meant to widen strategic conversation. Its recommendations are deliberately directional: embrace disruptors rather than resist them; hardwire transdisciplinary collaboration rather than leave it to chance; make the scientific physician a universal expectation; protect the uniquely human as the core of medicine; and reconfigure education to cultivate fluid, role-based identities across a lifetime.

On the evidence. The visioning is informed at selected points by current high-impact reviews and primary sources: on levels of autonomy in clinically deployed surgical robotics; on healthcare and biology as complex adaptive systems; on the structure and tractability of the translational gap; on digital twins and the routes to N-of-1 care; on the scope trajectories of advanced practice professions; on the psychosocial science of the patient-physician relationship, including the reliability gap that opens when opaque AI enters the consultation as a third participant; and on the multinational literature of patient attitudes, patient-reported outcomes, and decision-making preferences, which supplies the account of the patient in Chapter V and marks the limits of what that literature can currently tell us about 2050. Numbered notes follow.

The goal is not to predict a single future, but to build the imagination, the institutions, and the community that can shape whichever future arrives, so that, whatever medicine becomes, it remains unmistakably human.