Explainability
Definition
Explainability describes how comprehensibly the reasons or the process of an AI result can be presented.
In AI, explainability refers to the ability to provide an explanation for a result that is understandable and accurate for the target audience. The NIST report on explainable AI distinguishes explanations, comprehensibility, explanation fidelity, and system limitations.
Simply explained
A calculation does not become comprehensible just because it receives an arbitrary comment. As with a comprehensible invoice, an explanation must match the actual procedure. The comparison does not replace a technical review.
Why it matters in practice
Explainability can help teams critically assess results. Depending on the target audience, different levels of detail are required, and an explanation alone does not prove model reliability.
Example from business
Possible example: A company documents which inputs influence the output of a predictive model and checks whether this representation accurately describes the model process.
Limits and pitfalls
- Plausible-sounding explanations can be incorrect. Understandability, technical fidelity, data quality, and legal requirements must be evaluated separately.
Origin & history
Phillips and co-authors presented the report "Four Principles of Explainable Artificial Intelligence" for NIST in 2021 and described principles for the quality of AI explanations.
Use in mid-sized companies
- Possible fields of application include documentation of AI predictions, design of explanatory user interfaces, and evaluation of transparency requirements.
Frequently asked questions
Must an explanation be only easily understandable?
No. It should also accurately represent the actual process.
Does explainability automatically make a system fair?
No, fairness must be assessed separately.
Sources
- Four Principles of Explainable Artificial Intelligence · Phillips et al. (2021): Four Principles of Explainable Artificial Intelligence
Sources verified on 20.10.2026. Editorial application examples are hypothetical; no evidence of real customer results.
Editorially reviewed on: