A study of 140 radiologists indicates that AI assistance's impact on diagnostic accuracy is inconsistent, benefiting some while impairing others.

Radiologists' years of experience and previous AI tool usage did not predict how AI would affect their diagnostic performance.

The inherent accuracy of AI tools is critical, as more precise AI tools enhance performance, whereas less accurate tools may hinder it.

Further research shows that experience and baseline performance are unreliable in forecasting how radiologists will be affected by AI; top performers tend to remain so with or without AI.

The studies highlight that while AI has the potential to improve medical imaging outcomes, its effectiveness can vary greatly, necessitating careful planning for its integration into healthcare.

The research received support from the Alfred P. Sloan Foundation, J-PAL Health Care Delivery Initiative, and MIT's School of Humanities, Arts, and Social Sciences and was published in Nature Medicine.