Why LLM Health Copilots Cannot Replace Clinical Common Sense

In his 1950 paper 'Computing Machinery and Intelligence', Alan Turing proposed that textual indistinguishability equaled thought. That philosophical bet quietly underpins every consumer health copilot on your phone. A critique published by computer scientist Peter J. Denning and philosopher Ted G. Lewis in their 2025 analysis through Cambridge University Press demonstrates why that premise fails inside the clinic. Denning argues that human intelligence relies on practical know-how, social context, and intuitive observation that cannot be encoded as computational algorithms, regardless of context-window size.

For clinical work, this distinction matters immediately. A language model predicts the most probable next word based on billions of ingested sentences. A physician evaluates the shallow breathing of a patient who insists she feels fine, the faint yellow tint in someone's sclera, or the inconsistent phrasing of a parent describing a child's fall. That diagnostic instinct is embodied know-how, not text matching.

This boundary protects patient agency. You can use large language models productively to clarify medical jargon on lab reports, draft questions for your next appointment, or compare nutritional labels. Treat the interface as an unusually articulate library index, not an artificial clinician. When your body signals distress through pain, fatigue, or autonomic shifts, consult a person whose judgment is grounded in physical biology, not probability tables.