Explainability
What is Explainability?
The degree to which the internal mechanics of an AI or machine learning system can be explained in human terms. Explainability is a key requirement of the EU AI Act for high-risk AI systems.
Terms that appear alongside explainability
Each of these is named in at least one of the same controls as explainability. The number is how many controls name both.
- transparency 26 shared controls
- governance 12 shared controls
- eu ai act 8 shared controls
- nist 7 shared controls
- accountability 7 shared controls
- ai governance 7 shared controls
- compliance 6 shared controls
- ai transparency 6 shared controls
Frameworks that govern explainability
What the standards actually require on explainability
Requirements naming explainability across 6 standards, quoted from the control text.
AI-related risk source: Where the system's workings or the reasons for its outputs cannot be understood, errors go undetected, accountability weakens and stakeholders lose trust.
iso-iec-23894-2023::B.3 · Lack of transparency and explainability →Approaches include using inherently interpretable models and applying post-hoc explanation techniques to opaque models.
iso-iec-tr-24028-2020::9.3.4 · Approaches to explainability →To gain relevant stakeholders' understanding and reassurance and to present evidence of how the AI operated, the party receiving explanations shares what explanation it needs so that the explaining party can analyse and grasp what is required, and takes the ne...
japan-ai-guidelines::C6.4 · 6)(4) Improving explainability and interpretability for relevant stakeholders →Member States should require AI actors to disclose and combat stereotyping in AI outcomes and data, whether by design or negligence, and ensure training data do not foster inequalities, prejudice, disinformation or restrictions on expression, with attention to...
unesco-recommendation-on-the-ethics-of-ai::PA2d · paras 66, 67, 70 Combat stereotyping and bias in outputs and training data, diversify teams and data, and set clear transparency and explainability requirements →Adopt explainability practices so the functioning of the model and how it arrives at decisions can be communicated to relevant stakeholders.
AIGE-OM-5 · Model explainability →AI systems should be transparent (relevant people are told how, when and why an AI system is used) and explainable (relevant parties can access and understand how it reaches decisions) to a degree proportionate to risk and achievable with the state of the art,...
uk-ai-regulation-framework::P2 · Principle 2: Appropriate transparency and explainability (non-statutory, applied by regulators) →Questions people ask about explainability
What is Explainability?
Why is Explainability important for compliance?
Which compliance frameworks address Explainability?
Where can I learn more about Explainability?
See how Explainability applies across compliance frameworks
Our platform maps 1,016 frameworks with 316K cross-framework control mappings. Explore how this concept is addressed across standards.