Alternative models
Pooling risk might prevent the clinician becoming a liability sink, but the prevalent model may have other drawbacks for the clinician, the patient, and the system as a whole. Figure 1 shows that the entire input of the patient and clinician into the decision is restricted to either accepting the AI’s recommendation, or - for this case - rejecting it (effectively switching it off and returning to standard practice, albeit likely influenced by the AI’s recommendation). This is at odds with the goal of patient-centred decision-making,23 as the AI cannot easily incorporate patient context and ideas, concerns, and expectations itself - this context is only added by the clinician choosing to accept or replace the AI’s output. It may also be frustrating for the clinician by eroding their ability to do what they do best: integrating clinical science and patient context in a dialogue to come to a shared decision.