What If We Told the AI the Rest of Her Story?

Response to Michael Karch’s The Paradox of More

Michael Karch’s “The Paradox of More”[1] asks what physicians contribute when artificial intelligence can supply increasingly capable answers. His hypothetical patient makes the question concrete. A woman with advanced knee arthritis appears likely to benefit from replacement surgery. Her physician then learns that she is recently bereaved, lives alone, and fears losing her independence during recovery. A favorable prediction does not settle what she should do.

I agree. But I would give the example one more turn: What would the AI recommend if we told it the rest of her story?

As presented, the physician has information that the hypothetical system has not been given. The different recommendation may reflect that difference in information. The example establishes why the patient’s circumstances matter. It leaves open whether an AI system could recognize their importance and revise its advice.

We should not supply an imaginary answer and declare the matter settled. The system might recommend postponing surgery, arranging help at home, or exploring what the woman hopes to regain and what she fears losing. It might acknowledge every concern in sympathetic language while continuing to steer toward surgery. It might also become too cautious. Bereavement and fear do not automatically make an operation the wrong choice. She may still decide that relief of pain and improved mobility justify its burdens.

That uncertainty brings us directly to the human in the loop. What, exactly, is the human being asked to do?

If the clinician’s task is to approve a recommendation already framed as the sensible course, the patient’s concerns may become problems to manage on the way to surgery. Arranging postoperative help could be appropriate. But it is different from asking whether surgery remains the choice she wants. A recommendation can accommodate new facts without allowing them to change the decision.

The same is true of an AI response that sounds attentive to the patient. Mentioning her bereavement is not enough. The question is whether the information changes the options considered, the questions asked, or the recommendation itself. We should apply that standard to the physician as well.

Recent work on AI and clinical attention raises a related concern.[2] Once a plausible recommendation organizes the case, the clinician may begin checking whether its conclusion is defensible rather than asking what could make it inappropriate. That is a possibility to investigate, not something Karch’s hypothetical demonstrates. It is also why human presence alone cannot establish meaningful oversight.

A useful next step would be to compare what clinicians and AI systems ask before recommending an intervention, then examine how each responds when the patient’s circumstances change. Can either identify information that should reopen the decision? Can the patient correct the account on which the recommendation rests? Does the clinician have the time, evidence, and authority to pursue a different course?

Karch is right to insist that medicine involves judgment and responsibility. His example becomes stronger if we leave open what AI can contribute to that judgment. The physician’s responsibility remains even when the machine responds thoughtfully. Meaningful human oversight requires the opportunity to reconsider the decision with the patient—and the willingness to let the rest of her story change where it leads.

References

1. Karch MM. The Paradox of More. GrandRoundsMD. October 18, 2026. The Paradox of More
2. Newhall SC. Artificial intelligence-induced attentional displacement in clinical diagnosis. Version 1.0. Zenodo; September 12, 2026. Preprint; not peer reviewed. https://doi.org/10.5281/zenodo.22727468