Human-AI teaming in healthcare: 1 + 1 > 2?
Peng Fei Liu, Jiaxin Zhang, Shuaiqi Chen, Shanguang Chen, Shanguang Chen, Shanguang Chen
Abstract
While humans and AI-powered machines are expected to complement each other—for example, leveraging human creativity alongside AI’s computational power to achieve synergy (“1 + 1 > 2”)—the extent to which human–machine teaming (HMT) realizes this potential remains uncertain. We investigated this issue through reliability analysis of data from 52 empirical studies in clinical settings. Results show that medical AI can augment clinician performance, yet HMT rarely achieves full complementarity. Two factors matter: (1) teaming mode, with the simultaneous mode (clinicians review diagnostic cases and AI outputs concurrently) yielding greater benefits than the sequential mode (clinicians make initial judgments before reviewing AI outputs); and (2) clinician expertise, with juniors benefiting more than seniors. We also addressed two practical questions for medical AI deployment: how to predict or explain HMT reliability, and how to achieve clinically significant improvements. These findings advance understanding of human–AI collaboration in safety-critical domains.