The extent to which humans can direct, influence, or override the decisions and behaviors of an AI system.
Describes the degree of human–AI interaction points, from adjustable parameters (e.g., risk thresholds) to kill switches and override workflows. High controllability is crucial in high-stakes domains (healthcare, finance), ensuring that operators can intervene when AI outputs conflict with policy or ethics.
In an automated loan-approval pipeline, under-writable controls allow risk officers to manually adjust application scores. If the AI recommends denial but an officer identifies extenuating circumstances, they can override the decision, triggering a mandatory review log entry for audit purposes.
