Explainable AI Fails to Safeguard Doctors from Biased AI's Diagnostic Errors
- Study tested whether explainable AI improves doctor's diagnostic accuracy vs biased AI
- Accurate AI modestly improved diagnosis; biased AI substantially worsened it
- Researchers hoped explainable AI would help doctors realize model was biased
- But explanations did not act as a "safety valve" against biased AI
- Concerning that explainability does not safeguard against errors from biased AI