The crucial responsibilities doctors must never offshore to AI

Two doctors from prestigious medical schools are cautioning that human doctors must retain crucial responsibilities even as artificial intelligence advances into healthcare. Andrew S. Parsons, M.D., M.P.H., from the University of Virginia and Adam Rodman, M.D., M.P.H., from Harvard Medical School note that AI has become adept at certain tasks, like diagnosing from written cases and electronic health records.

Yet, they maintain that only humans can determine the appropriate course of treatment, taking into account a patient's goals, comorbidities, social circumstances, and the associated uncertainties.

The authors argue that while AI can identify conditions, it cannot decide which risks a particular patient should accept. Real patients often fall outside the norms established by the populations studied in guidelines and trials, making it challenging to apply a "correct" answer universally. Guidelines and trials may lean towards average cases, but real-life patients frequently have complex conditions and unique social circumstances, or different goals than those in a trial's endpoint.

Parsons and Rodman emphasize that deviating from the technically correct answer can often be beneficial, and someone must bear responsibility for those deviations. They argue that even as AI takes on more diagnostic roles, clinicians must still consider each individual's circumstances, work with patients to make hard choices, and ultimately shoulder the responsibility for the outcomes.

The authors claim that accountability in medicine will always lie with the clinician, who must reason under uncertainty and answer for the decisions made.

This ownership is crucial only when the clinician knows they will answer for the result, as it influences their decision-making differently than a system with nothing to lose. Furthermore, a clinician who expects to continue seeing a patient may consider how choices will affect their relationship if things go wrong. The authors conclude that while AI will continue to improve the information fed into these decisions, it cannot assume ownership of what happens next.

In their conclusion, Parsons and Rodman urge clinicians to welcome AI's growing accuracy while retaining the responsibility of acting for each patient. They stress that the patient, not the computer, will have to live with the consequences, and it is the clinician's ethical duty to ensure the best possible care for each individual.