Explainability is the property of an AI system that allows its behaviour and outputs to be understood by humans, including the factors that drove a particular decision and the limits of the model's competence.
In practiceThe NIST AI Risk Management Framework treats explainability as one of the trustworthiness characteristics, alongside reliability, safety, security, accountability, privacy and fairness. Practitioners distinguish between intrinsic interpretability, where the model itself is transparent, and post-hoc explanation, where techniques such as SHAP or counterfactuals approximate why a black-box model produced a given output. A common confusion is to treat any feature importance plot as a sufficient explanation, when regulators and affected individuals often require reasons that are meaningful in context.
A bank deploying a machine learning model for small business lending pairs each adverse decision with a counterfactual explanation showing which inputs, if changed, would have led to approval, supporting both customer communications and adverse action notices.
This definition is maintained by Moweb partners and used in live client engagements. For how Explainability applies to your estate, or to challenge a working definition, speak to a partner.