AI governance glossary
AI explainability
What is AI explainability?
AI explainability is the ability of an AI system, or the people operating it, to give reasons for an output that are accurate and understandable to the person receiving them, and to recognise when the system is operating outside the conditions it was designed for.
Why it matters
- People affected by certain AI-assisted decisions have a right to an explanation: the EU AI Act gives one for decisions based on high-risk systems that significantly affect them (Article 86), alongside GDPR rights on automated decisions.
- Credit rules already require specific reasons for adverse decisions, such as adverse action notices in US consumer credit, whatever model produced them.
- Explanations are how errors and bias are found and how human oversight works in practice.
In detail
NIST's four principles of explainable AI (NISTIR 8312) are explanation, meaningful, explanation accuracy and knowledge limits. Explainability differs from transparency (disclosing that AI is used and how it works in general) and from interpretability (how far a model's internal workings can be understood). Laws usually require the result rather than a technique: reasons a person can understand and contest.
Source: NISTIR 8312, Four Principles of Explainable Artificial Intelligence. A plain-language explanation for orientation, not the legal text; follow the source for the binding wording.
Adverse decision explanation and appeal kit →Decision explanation and appeal route control →
Laws and policies on record that use it
- AI Basic ActSouth Korea · In force
- PCPD Model Personal Data Protection Framework for AI (2024)Hong Kong SAR · Guidance
- ASEAN Guide on AI Governance and Ethics (2024)ASEAN (Association of Southeast Asian Nations) · Guidance
- UK AI regulation frameworkUnited Kingdom · Guidance
- NIST AI RMFUnited States · Voluntary standard
- UNESCO Recommendation on the Ethics of AIInternational organisations (OECD, UNESCO, UN, G7, ISO) · Adopted
- Singapore Model AI Governance FrameworkSingapore · Voluntary standard
- OECD AI Principles (Recommendation on AI)International organisations (OECD, UNESCO, UN, G7, ISO) · Adopted
Duties that mention it
- Manage data quality, model development and monitoring across the lifecycleSingapore Model AI Governance Framework · Singapore
- Operators of high-impact AI must be able to explain outputs and the main criteria behind themAI Basic Act · South Korea
- Provide appropriate transparency and explainabilityUK AI regulation framework · United Kingdom
- Carry out a data protection impact assessment for high-risk AI processingICO AI guidance · United Kingdom
- Measure and test trustworthiness characteristics (Measure)NIST AI RMF · United States
Related concepts
- Human oversightDesign and operating measures that let people effectively oversee an AI system while it is in use: understand...
- Algorithmic biasAlgorithmic bias is systematic error in an AI system's outputs that produces unjustified differences in treatm...
- Model cardA short document published with a machine-learning model that states its intended use, the data it was evaluat...
- Red teaming (AI)AI red teaming is structured adversarial testing in which testers try to make an AI system fail, by producing...
All glossary terms · Think a definition is off? Tell us; corrections are logged on the corrections page.
Frequently asked questions
What is AI explainability?
AI explainability is the ability of an AI system, or the people operating it, to give reasons for an output that are accurate and understandable to the person receiving them, and to recognise when the system is operating outside the conditions it was designed for.
Where does the term AI explainability come from?
This explanation follows NISTIR 8312, Four Principles of Explainable Artificial Intelligence. It is a plain-language paraphrase for orientation; the source has the binding wording.
Which laws and policies use the term AI explainability?
Among the records on this site: AI Basic Act (South Korea); PCPD Model Personal Data Protection Framework for AI (2024) (Hong Kong); ASEAN Guide on AI Governance and Ethics (2024) (ASEAN); UK AI regulation framework (UK) and NIST AI RMF (US).