Skip to content

Read-only version: For security reasons, the Research Information System (FIS) has been temporarily replaced by this read-only version. All public content and full texts remain available; login, editing, submission and statistics are not available. Questions: fis@uni-bamberg.de

Home › Publications › How explanations from XAI-based decision support affect human task performance : a meta-analysis

Publication

How explanations from XAI-based decision support affect human task performance : a meta-analysis

Haag, Felix (2026): How explanations from XAI-based decision support affect human task performance : a meta-analysis, in: Journal of decision systems, Abingdon: Taylor & Francis, Jg. 35, Nr. 1, 2616693, S. 1–28, doi: 10.1080/12460125.2026.2616693.

Details

Faculty/Chair
Faculty/Professorship
Author
Title of the Journal
Journal of decision systems
ISSN
2116-7052
1246-0125
Publisher Information
Taylor & Francis : Abingdon
Year of publication
2026
Volume
35
Issue
1, 2616693
Language
English
DOI
Abstract

Artificial intelligence (AI) increasingly supports human decision-making across domains. Yet, many AI-based decision support systems (DSS) rely on machine learning models that are ‘black boxes’ to humans. This opacity has driven the development of explainable AI (XAI) methods that explain model outputs in human-understandable terms. Empirical findings, however, remain inconsistent regarding whether and how such explanations affect users’ decision performance: some studies report improvements, while others find negligible or even negative effects. To reconcile these inconsistencies, we conduct a meta-analysis with 4589 participants comparing XAI-supported decisions to unaided decisions and 7706 participants comparing XAI-supported to AI-only supported decisions. Our analyses indicate that, on average, XAI-based decision support is associated with higher task performance compared to no support, whereas the additional gains of explanations over AI-only support are small. We also find that studies’ risk of bias levels are associated with the magnitude of reported effects, suggesting larger reported performance gains in studies with higher risk of bias. Interestingly, explanation type alone does not show a significant moderating effect on task performance across studies. Overall, these findings provide a better understanding of how XAI explanations influence human decision-making and inform the design of XAI-based DSS.

Keywords
Explainable artificial intelligence (XAI)
interpretable machine learning
decision performance
explanation types
risk of bias
Type
Article
URI
Activation date
2026-06-18
Project(s)

Permalink: https://fis.uni-bamberg.de/handle/uniba/115619 · Data as JSON · Cite: BibTeX · RIS

Data as of: 2026-11-16

Restricted read-only access: login, editing and submission are temporarily disabled.