---
title: Explainability
description: 'The ability to provide a human-usable account of why a system produced a particular output or action.'
canonical_url: 'https://darkfactory.dev/glossary/explainability'
markdown_url: 'https://darkfactory.dev/glossary/explainability.md'
collection: glossary
date_published: '2026-08-03T00:00:00-04:00'
date_modified: '2026-08-26T00:00:00-04:00'
---

# Explainability


## Definition

Explainability is the ability to give a person a useful account of why an AI system produced a particular output or action. "Useful" depends on the audience and the decision. A model engineer debugging a failure, an applicant challenging a denial, and an auditor testing compliance need different kinds of explanation.

Explanations may come from an inherently understandable model, analysis of internal behavior, or a post-hoc method applied after the prediction. They should be tested for faithfulness, stability, and relevance. A fluent rationale generated by the same model can be persuasive while bearing little relationship to the mechanism that produced the answer.

## Distinguish it from nearby terms

Interpretability focuses on understanding a model's representations, mechanisms, or decision process. Explainability focuses on communicating an account that serves a human purpose. Transparency discloses facts about the system and its development; disclosure alone does not explain a particular result.

## Check your understanding

A lending system lists "recent credit history" as the reason for a denial, but changing that feature does not affect the result. The statement may be readable, yet it is not a faithful explanation of this decision.

## Related terms

- [Interpretability](https://darkfactory.dev/glossary/interpretability)
- [Post-hoc explanation](https://darkfactory.dev/glossary/post-hoc-explanation)

## Evidence and further reading

- [NIST AI Resource Center Glossary](https://airc.nist.gov/glossary/)
- [Google Machine Learning Glossary](https://developers.google.com/machine-learning/glossary/)