Two preregistered experiments with over 1,000 participants show that patients trust a physician less when the AI suggests a diagnosis than when it merely highlights information – unless the physician forms an independent judgment first.
Practising with AI-enabled virtual patients improved perceived competence and self-efficacy and reduced insecurity among future psychotherapists – but not more than watching role-play videos. Our randomized study in BMC Medical Education.
In three preregistered experiments, showing AI advice only on demand or only after a person’s own decision reduced overreliance on wrong advice – but at the cost of underreliance on correct advice.