Explainability (explainable AI)
The ability to make an AI model's decisions understandable, a matter of trust, regulatory compliance and liability.
Explainability is the ability to make an artificial intelligence model's decisions or predictions understandable to a human being. Its opposite is the black box: the most complex models often buy their performance with such opacity that even their own designers struggle to say why a given decision was made. The most powerful are frequently the least interpretable, and that tension between performance and transparency is one of the dilemmas of applied AI. The subject has moved out of the technical corner and become a regulatory and legal one. Data protection law regulates fully automated decisions and recognizes a right to information about the underlying logic, and the new frameworks applicable to AI impose transparency requirements on high-risk systems. Dedicated techniques try to shed light, after the fact, on how models work. For insurance and liability, explainability weighs heavily: fault is hard to attribute, a decision hard to contest and a harm hard to compensate when no one knows what happened inside the model. Opacity dilutes responsibility and complicates proof.
Hypothetical case: a bank turns down a loan on the recommendation of a model whose decision it cannot explain. The customer challenges it, the regulator asks questions, and the bank struggles to justify a refusal that carries heavy consequences.
explainable AI, XAI, interpretability, explainability, AI explainability
Follow the research
One letter a week. First, the week's underwriting note, when one is published: a dated, argued decision, signed by Paul Bernardski. Then what the observatory published on cyber risk, political risk and insurance. No account, no tracking.
No spam, one-click unsubscribe, no data ever sold.