From Art to Algorithm: The Epistemic Displacement of Clinical Judgment in Contemporary American Medicine
The Moment Medicine Decided to Distrust Itself
In 1992, a working group at McMaster University published a manifesto of sorts in the Journal of the American Medical Association, formally introducing the term 'evidence-based medicine' to a wide clinical audience. The document was careful to position the new approach not as a repudiation of clinical experience but as a complement to it—a way of subjecting traditional practice to rigorous empirical scrutiny. Yet the philosophical implications of the framework, as it was subsequently institutionalized across American medical education and healthcare delivery, ran considerably deeper than the original authors may have intended.
What evidence-based medicine gradually installed, at least in its dominant institutional forms, was a hierarchy of knowledge in which randomized controlled trial data occupied the apex and individual clinical judgment occupied the base—or, in some formulations, did not appear on the hierarchy at all. This arrangement reflected genuine epistemological commitments: the recognition that individual practitioners are subject to cognitive biases, that anecdotal experience is an unreliable guide to treatment efficacy, and that the history of medicine is littered with interventions that seemed to work until they were rigorously tested. These are not trivial concerns.
Yet the philosophical costs of this hierarchy have received comparatively little attention, particularly as evidence-based medicine has evolved into something its founders may not have fully anticipated: a computational infrastructure in which algorithmic prediction systems increasingly mediate the relationship between clinical data and clinical decision.
Tacit Knowledge and the Epistemology of the Bedside
The philosopher Michael Polanyi introduced the concept of tacit knowledge to describe forms of expertise that cannot be fully articulated in explicit propositional terms—the knowledge that inheres in skilled performance, in pattern recognition, in the accumulated experience of doing something well over an extended period of time. 'We know more than we can tell,' Polanyi wrote, and this observation applies with particular force to clinical medicine.
The experienced diagnostician does not simply apply a decision tree to a presenting set of symptoms. She attends to the texture of a patient's narrative, the quality of her breathing, the subtle incongruities between what is said and what is observed. She draws on a vast repertoire of prior encounters, most of which are not stored as explicit propositions but as something closer to pattern recognition—a felt sense of when a presentation 'fits' a familiar category and when it does not. This form of knowledge is not infallible; it is subject to precisely the biases that evidence-based medicine was designed to counteract. But it is also not reducible to the variables that any predictive model, however sophisticated, is capable of encoding.
The philosophical question is not whether algorithmic prediction is useful—it demonstrably is, across a wide range of clinical applications—but whether the institutional privileging of algorithmic outputs over clinical judgment involves a systematic exclusion of epistemically legitimate forms of knowledge.
The Predictive Turn and Its Institutional Consolidation
The transition from evidence-based medicine as methodology to predictive medicine as infrastructure has been enabled by a convergence of technological and economic forces specific to the American healthcare context. The digitization of medical records, mandated by the Health Information Technology for Economic and Clinical Health Act of 2009, created the data substrate on which predictive algorithms could be trained. The commercial imperatives of the insurance and hospital administration sectors created powerful incentives to standardize clinical decision-making in ways that are auditable, defensible, and cost-predictable.
The result is an environment in which clinical judgment is increasingly legible only when it can be documented in terms that algorithmic systems recognize. A physician who departs from a protocol-recommended course of treatment must justify that departure in writing, often to administrators who evaluate the justification against criteria derived from population-level statistical models. The epistemic authority of the individual clinician is not formally abolished in this arrangement—it is simply required to translate itself into a language it was not designed to speak.
Resistance from the Clinic: More Than Conservatism
Physicians who express discomfort with the predictive turn are frequently characterized, in both popular and academic discourse, as resistant to change—defenders of professional prerogative masquerading as epistemological critics. This characterization deserves scrutiny. Some of the most philosophically articulate resistance to algorithmic displacement of clinical judgment comes from clinicians with substantial quantitative training who are, accordingly, well-positioned to understand both the power and the limitations of predictive modeling.
Cardiologists, for instance, have raised concerns about sepsis prediction algorithms deployed in American hospital systems, noting that models trained on population data from specific institutional contexts may perform poorly when applied to individual patients whose presentations diverge from the training distribution. Oncologists have questioned whether progression-free survival, a metric readily amenable to algorithmic optimization, is an adequate proxy for the quality-of-life considerations that experienced clinicians navigate through sustained patient relationships. These are not objections to quantitative rigor; they are objections to the conflation of what is measurable with what is medically relevant.
The Philosophical Stakes
At the deepest level, the debate over algorithmic prediction in medicine is a debate about the nature of medical knowledge itself. If medicine is fundamentally a natural science whose object is the biological mechanisms of disease, then the displacement of individual clinical judgment by population-derived predictive models represents a straightforward epistemological advance—the replacement of unreliable individual inference with more reliable statistical inference. If medicine is, as many of its practitioners and historians have argued, also an interpretive practice whose object is the suffering individual in her particular circumstances, then the picture is considerably more complicated.
The philosopher of medicine Miriam Solomon has argued that medical knowledge is irreducibly social and contextual in ways that purely statistical frameworks struggle to capture. Her work, along with that of scholars in the tradition of narrative medicine developed at Columbia University, suggests that what is at stake in the algorithmic turn is not merely a technical question about prediction accuracy but a philosophical question about what medicine is for.
Answering that question requires more than better algorithms. It requires a sustained engagement with the epistemological foundations of clinical practice—an engagement that, thus far, the American medical establishment has been only intermittently willing to undertake.
