A multi-institutional study using artificial intelligence to provide reliable and fair feedback to surgeons

Surgeons who receive reliable feedback on their performance quickly master the skills necessary for surgery. A recently developed AI system assesses surgeon skills from surgical videos and highlights pertinent aspects. This study quantifies the reliability of AI-based explanations by comparing them to human expert explanations. To improve reliability, the authors propose 'Training with Explanations' (TWIX), using human explanations as supervision. Results show that while AI explanations often align with human explanations, they are not equally reliable for all surgeons (explanation bias). TWIX enhances reliability, mitigates bias, and improves AI system performance across hospitals. These findings extend to a training environment for medical students.

Dani Kiyasseh, Jasper Laca, Taseen F. Haque, Brian J. Miles, Christian Wagner, Daniel A. Donoho, Animashree Anandkumar, Andrew J. Hung