Explainability of AI decisions is the ability to clearly state the basis on which a system arrived at a given output. It gains practical importance everywhere a decision affects a person – in assessing applications, selecting candidates, evaluating creditworthiness, or allocating a service. For high-risk systems, the data subject has the right to an explanation of the system's role in the decision-making and of its main elements. Complex models, meanwhile, function as a black box, so explainability is secured through a combination of measures: documentation of input factors, operational records, supplementary explanation tools, and, above all, the fact that the final decision is made by a human who can justify it. Companies should therefore not deploy unexplainable models where people's rights are at stake.
See also: Human oversight of AI, High-risk AI system, Records of AI system operation.
