A Cornell University fellow develops strategies to extract more than correlations from algorithms’ predictions.
The method has two main features: it evaluates how AI models reason through problems instead of just checking whether their ...