For AI systems supporting high-stakes medical decisions, PMI-CP/CPMAI and responsible AI guidance emphasize human-in-the-loop oversight as the primary way to manage inherent uncertainty and risk. In clinical diagnosis, symptoms are often ambiguous, overlapping across multiple conditions, and influenced by patient history and context. No matter how advanced the model, there will be edge cases, rare diseases, and conflicting signals.

Rather than attempting to eliminate uncertainty purely through more complex models, more input variables, or ever-growing rule sets, best practice is to design the AI as a decision-support tool, not an autonomous decision-maker. That means physicians retain ultimate responsibility, reviewing AI suggestions, over-riding them when clinically necessary, and using their expertise to weigh patient-specific factors the model may not capture.

Human-in-the-loop design also supports explainability and trust: clinicians can question outputs, cross-check with other evidence, and provide feedback that can be used later for model improvement. CPMAI’s lifecycle framing for regulated and safety-critical domains is clear: when outcomes materially affect health or life, the appropriate way to handle uncertainty is to keep a human in the loop for all decision-making, which aligns directly with option A.