Which is more accurate: a doctor, a doctor with AI, or AI without a doctor?
It seems obvious: a doctor plus AI should beat both the doctor and the algorithm on their own. Yet in a number of studies the most accurate performer turned out to be AI on its own. Let us look at why the human plus machine pairing sometimes loses to the machine.
In several diagnostic tasks (scans, ECGs, pathology) the standalone algorithm was more accurate than the same algorithm in the hands of a doctor. After seeing the AI's suggestion, the specialist often worsened the final decision.
Automation bias: the person blindly agrees with a wrong suggestion from the machine. Algorithm aversion: the opposite, rejecting a correct suggestion out of distrust in the algorithm. Both traps chip away at the pairing's accuracy.
The benefit of AI does not appear on its own simply because it has been integrated. It takes physician training, quality audits, oversight, and interfaces that nudge toward critical checking rather than blind agreement.
This is not a single experiment but a collection of different studies with different tasks and metrics. In some cases the human plus AI pairing is still the best. There is no universal answer, only a dependence on the task and the design.
The question is not "AI versus the doctor" but how to design their work together. The stronger clinician is not the one who has AI, but the one who knows how to disagree with it.