Safety · Established · Intermediate

Explainable AI

Also known as: XAI

Methods that make a model's individual decisions understandable to the people affected by them.

What Explainable AI is

Where interpretability studies internals, explainable AI focuses on decision-level accounts: which factors drove this loan rejection, which region of the scan triggered this flag.

How it works

Techniques include SHAP and LIME attributions, counterfactual explanations, attention and saliency visualisation, and using inherently interpretable models where the stakes justify the accuracy trade-off.

Why it matters

Regulation in credit, employment and healthcare increasingly requires meaningful explanation, and users reasonably refuse to act on unexplained outputs.

Common uses

→Credit decision reasons
→Clinical decision support
→Model debugging
→Regulatory documentation

Strengths

✓Builds trust and enables recourse
✓Reveals spurious features

Watch for

✓Post-hoc explanations can be unfaithful
✓Can create false confidence

Continue exploring

Interpretability Bias in AI AI Governance Decision Tree