In our study with 220 clinicians, machine learning recommendations did not improve the quality of clinicians' treatment selections even though the (simulated) machine learning system was nearly twice as accurate as clinicians working on their own. Read more.