← The frontier
Technology & AIJul 10, 2026

ConceptSMILE: Auditing the Trustworthiness of Concept-Based Explainable AI

Concept-based explainable artificial intelligence (AI) can make model reasoning more human-understandable, but concept-level outputs are not automatically trustworthy.

Concept-based explainable artificial intelligence (AI) can make model reasoning more human-understandable, but concept-level outputs are not automatically trustworthy. We introduce ConceptSMILE, a model-agnostic perturbation-based auditing framework for evaluating the…

The frontier is open to all. Sign in to learn this from first principles and save it to your knowledge base.