#interpretable ai

Researchers Challenge Accuracy-Based AI Evaluation With Symbolic-Mechanistic Framework

New position paper argues standard accuracy metrics fail to detect memorization, data leakage, and brittle shortcuts in machine learning models.