AI readiness tracks clinical tasks, not specialties: 330-task scoring finds task category explains ten times more variance
Synopsis
Whether a medical specialty is ready for AI depends more on which clinical tasks it contains than on its name. A framework scored 330 clinical tasks, 15 for each of 22 specialties drawn from the American Board of Medical Specialties list, on six dimensions: codifiability, data-structure dependence, transferability of decision authority, physical dependence, clinical governability and specialty centrality, each on a three-point scale, then aggregated them into a Medical Specialty AI Readiness (MSAIR) score. Task category explained 71.6% of adjusted R² versus 6.4% for specialty; documentation scored highest on average (2.874) and emergency decision making lowest (1.639), while specialty means spanned only 2.010 to 2.543.
Interpretation
Moving the unit of analysis from the specialty to the clinical task shows that variation within a specialty's tasks matters more than variation between specialties. Earlier debate about which specialties AI will automate treated specialties as whole units, extrapolating strong performance on single abilities such as image interpretation to an entire field. MSAIR means across the 22 specialties ranged from 2.010 to 2.543, a spread of 0.533, while task-category means ran from 1.639 for emergency decision making to 2.874 for documentation, a spread of 1.235, more than twice as wide.
Task category carries far more explanatory power for AI readiness than specialty membership, and adding specialty to the model buys little. In task-level regressions, specialty alone as a categorical variable gave an adjusted R² of 0.064, while task category alone gave 0.716. Combining both raised adjusted R² to 0.734, an additional 1.7 percentage points; after excluding specialty centrality, task category gave 0.778, specialty 0.098 and the combined model 0.799, so the result did not depend on including or double-weighting centrality.
Each task is scored on six dimensions and combined into a weighted score, and a specialty's value is the arithmetic mean of its 15 task scores. Five of the six dimensions describe whether a task can be handed to AI, while specialty centrality separately describes how important that task is to the specialty's identity. The primary score is (C+D+T+P+G+2S)/7, with specialty centrality double-weighted; tasks came from the American Board of Medical Specialties certificate list, with task lists based mainly on Accreditation Council for Graduate Medical Education competency frameworks, totaling 330 tasks.
Perspective
The framework reports aggregated readiness of the selected task portfolios for relative comparison, not absolute probabilities of AI adoption, automation or workforce displacement. Its intended readers are healthcare leaders, policymakers, medical educators and professional societies judging which parts of clinical work may change first. The source proposes future evaluation through broader international expert consensus, larger specialty-specific task inventories and periodic updates reflecting advances in AI capabilities, and notes the same task-based approach could be adapted to other healthcare professions or health-system functions.
Task selection and scoring involved expert judgement, the database was not built through a formal Delphi process, and alternative investigators might select, define or score some tasks differently. The six dimensions and their weighting are methodological choices, and other frameworks could yield different estimates. AI capabilities and medical practice are evolving rapidly, so readiness should be treated as dynamic rather than fixed and will require reassessment as technologies, tasks and specialty practice change.