Dictionary of Applied Machine Learning
Typeset PDF version — the authoritative form of this entry
Explainable artificial intelligence (XAI) is the subfield of artificial intelligence (AI) concerned with making the predictions of machine learning (ML) methods understandable to humans. XAI can be framed as function approximation: a learned hypothesis is explained by a simpler function that a human can comprehend, either locally around a data point (local interpretable model-agnostic explanations (LIME), counterfactuals) or via additive feature contributions (SHapley Additive exPlanations (SHAP)). Such post hoc explanations treat the learned hypothesis as fixed; alternatively, explainable empirical risk minimization (EERM) builds explainability into training. If the hypothesis space is sufficiently simple, a learned hypothesis is its own explanation (see interpretable machine learning (interpretable ML)).
Explainable artificial intelligence (XAI) is the subfield of artificial intelligence (AI) concerned with making the predictions of machine learning (ML) methods understandable to humans. Each prediction is complemented by an explanation of how it has been obtained. If the ML method uses a sufficiently simple hypothesis space, the learned hypothesis may be inherently interpretable and no separate explanation is needed (Rudin, 2019). The term XAI was popularized by a program of the US Defense Advanced Research Projects Agency (Gunning and Aha, 2019); in the context of ML, the synonymous term explainable machine learning (explainable ML) is also used. The underlying property is explainability, which the international terminology standard for AI defines for artificial intelligence systems (AI systems) in general (Standardization and Commission, 2022, Sect. 3.5.7).
XAI can be framed as function approximation. A trained
hypothesis
$\learnthypothesis: \featurespace \rightarrow \labelspace$,
e.g., delivered by an opaque deep net, is typically a
highly non-linear function on a potentially
high-dimensional feature space $\featurespace$. Explaining
$\learnthypothesis$ amounts to describing its behavior locally,
around a given data point, in a form that a human can
comprehend (see Fig. 1). One form of
explanation is a simple local approximation of
$\learnthypothesis$, such as the linear function fitted by
local interpretable model-agnostic explanations (LIME), which makes the contribution of each feature
to the prediction explicit (Ribeiro et al., 2016).
SHapley Additive exPlanations (SHAP) decomposes the prediction into additive
feature contributions based on Shapley values
(Lundberg and Lee, 2017; Molnar, 2025).
Another form of explanation is a counterfactual:
instead of approximating $\learnthypothesis$, it points out
changes of the features that change the value of
$\learnthypothesis$, i.e., the prediction. A
counterfactual identifies the smallest such change,
measured by a chosen metric such as the Euclidean norm
(Wachter et al., 2018).
Concept activation vectors (CAVs) probe internal representations of a trained
hypothesis with linear classifiers to test
sensitivity to human-defined concepts (Kim et al., 2018).
These are post hoc methods: they treat the learned
hypothesis as fixed and construct the explanation
afterwards, from a simpler hypothesis out of a
user-parseable hypothesis space
(see Fig. 2).
A goal of XAI is to quantify the influence of features on the prediction of a learned hypothesis, whether the feature is an input attribute of a data point or, in the sense of mechanistic interpretability, a concept encoded internally, e.g., by the activations of an artificial neural network (ANN).
Synonyms: explainable machine learning, explainable ML.
See also: explainability, interpretability, explanation, interpretable machine learning, explainable empirical risk minimization, local interpretable model-agnostic explanations, SHapley Additive exPlanations, counterfactual, feature, mechanistic interpretability.
@misc{dictml_xaiterm,
author = {Jung, Alexander},
title = {explainable artificial intelligence},
howpublished = {Dictionary of Applied Machine Learning (course edition)},
year = {2026},
doi = {10.5281/zenodo.21569296},
note = {ISBN 978-952-64-3013-3, CC BY 4.0, retrieved 2026-08-06},
url = {https://dictionaryofml.org/terms/xaiterm.html}
}