Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 55 of the 59 references it cites were found in a registry or an index; 7 of the 11 sources its claims cite were checked for support.

AI Trust: Can Explainable AI Enhance Warranted Trust?

Open access
de Brito Duarte, Regina, Correia, Filipa, Arriaga, Patrícia, Paiva, Ana · 2023 · Human Behavior and Emerging Technologies

Abstract

Explainable artificial intelligence (XAI), known to produce explanations so that predictions from AI models can be understood, is commonly used to mitigate possible AI mistrust. The underlying premise is that the explanations of the XAI models enhance AI trust. However, such an increase may depend on many factors. This article examined how trust in an AI recommendation system is affected by the presence of explanations, the performance of the system, and the level of risk. Our experimental study, conducted with 215 participants, has shown that the presence of explanations increases AI trust, but only in certain conditions. AI trust was higher when explanations with feature importance were provided than with counterfactual explanations. Moreover, when the system performance is not guaranteed, the use of explanations seems to lead to an overreliance on the system. Lastly, system performance had a stronger impact on trust, compared to the effects of other factors (explanation and risk).

Overview

CoChat Summary

## TL;DR

This experiment tested whether explainable AI increases trust in an AI recommender system, using a mushroom-judgment game with 215 participants. Explanations helped trust in some cases—especially feature-importance explanations (LIME)—but AI performance had a stronger effect than explanations or risk.

## Key findings

  • In the main sample, 211 participants remained after attention checks; the study used a 3 × 2 × 2 design with XAI type, risk level, and model performance.
  • High AI performance increased all five measured outcomes: MDMT (5.34 vs 4.05), direct subjective trust (4.84 vs 3.46), behavioral trust (0.87 vs 0.80), AI understandability (4.53 vs 3.92), and XAI quality (4.85 vs 4.11); all were significant at p < 0.001.
  • Feature-importance explanations (LIME) outperformed counterfactual explanations (DICE) and control on self-reported trust and understanding: MDMT (4.98 vs 4.60 and 4.50), AI understandability (4.70 vs 3.99 and 4.00), and XAI quality (5.01 vs 4.40 and 4.04).
  • Behavioral trust depended on an XAI × performance interaction (F2,205 = 25.40, p < 0.001, η²p = 0.20): with high performance, LIME had the highest adherence to recommendations (0.93 ± 0.02), but with low performance LIME had the lowest adherence (0.75 ± 0.02) versus DICE (0.81 ± 0.02) and control (0.84 ± 0.02).
  • Risk mattered mainly when performance was low: in that case, trust and related ratings were lower in the high-risk condition than the low-risk condition; the paper reports delegation was more pronounced in low risk when performance was low (χ²1 = 9.52, p = 0.002).

## Methods

  • Experimental user study built around a web-based mushroom game in which participants made 12 eat/avoid decisions per round with advice from an AI recommender system.
  • Participants were recruited from Prolific; 215 were recruited, 211 remained after attention checks, and they were non-experts presumed to be lay users. The sample included 114 women, 94 men, and 3 nonbinary participants; mean age was 27 years.
  • XAI type was between-subjects (LIME feature importance, DICE counterfactuals, or no explanation), risk was between-subjects (high vs low), and AI performance was within-subjects (high vs low, counterbalanced). Analysis used mixed-design ANOVA in SPSS 26 plus a chi-squared test for delegation, with Bonferroni correction.

## Limitations

  • The paper states it only examined two XAI categories: counterfactual explanations and feature-importance explanations.
  • The authors note the study used online participants assumed to be lay users, so findings may not extend directly to expert–AI collaboration. They also say the trustworthiness meaning of the 60% “low performance” condition is debatable and may not generalize to even lower-performance systems.
  • Evident from the text: the study used a game-based mushroom task, so external validity to other real-world decision contexts is limited.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1155/2023/4637678

Cites

55 of 55 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Image 1
CC-BY · From the PDF
(b) LIME layout
(b) LIME layoutCC-BY · From the PDF
(c) DICE layout
(c) DICE layoutCC-BY · From the PDF
Figure 2 Experiment work fl ow.
Figure 2 Experiment work fl ow.CC-BY · From the PDF
Figure 3 The average values of the three trust measures in relation to the XAI condition and AI system performance are presented. In terms of subjective trust measures, it becomes evident that when the AI model ' s performance is reduced, reported trust diminishes in comparison to instances where the system performs better. When considering each type of explanation, participants who were exposed to LIME explanations consistently reported greater trust compared to the other type of explanations (DICE and no explanations), regardless of the performance of the AI system ' s performance.
Figure 3 The average values of the three trust measures in relation to the XAI condition and AI system performance are presented. In terms of subjective trust measures, it becomes evident that when the AI model ' s performance is reduced, reported trust diminishes in comparison to instances where the system performs better. When considering each type of explanation, participants who were exposed to LIME explanations consistently reported greater trust compared to the other type of explanations (DICE and no explanations), regardless of the performance of the AI system ' s performance.CC-BY · From the PDF

Verification

Claims the paper makes

5 Valid · 1 Not valid · 4 Not checked · 71 Own results and interpretations

41 more claims

Create a free account to see all 81 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 16
  • It may state two separate claims as one. 21
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 59 references

Cited by

See all 52

Metadata

Authors
de Brito Duarte, R., Correia, F., Arriaga, P., & Paiva, A.
Affiliations
  • INESC-ID, Instituto Superior Técnico, Universidade de Lisboa, Portugal
  • University of Lisbon (PT)
  • ISCTE-Instituto Universitário de Lisboa (IUL), CIS-IUL, Lisboa, Portugal
  • Instituto de Engenharia de Sistemas e Computadores Investigação e Desenvolvimento (PT)
Venue
Human Behavior and Emerging Technologies
Publisher
Wiley
Published
2023-10-31
Volume
2023
Pages
1-12
ISSN
2578-1863 (ISSN-L)
OpenAlex
W4388040162
Cited by
52
References
59
Citations by year
2026: 26 · 2025: 22 · 2024: 4
FWCI
4.89
Type
journal article
Language
en
Open access
Locations
  • Human Behavior and Emerging Technologies (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Repositório do ISCTE-IUL (submittedVersion, other-oa)
Funders
  • LARSyS (UIDP/50009/2020, LA/P/0083/2020, UIDB/50021/2020, C628696807-00454142, H2020-ICT-48-2020/952215, H2020-ICT-48-2020/952026)
  • INESC-ID (UIDP/50009/2020, LA/P/0083/2020, UIDB/50021/2020, C628696807-00454142, H2020-ICT-48-2020/952215, H2020-ICT-48-2020/952026)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
  • Social Sciences › Human Factors and Ergonomics › Human-Automation Interaction and Safety
Keywords
explainable AI, trust in AI, AI recommendation systems, counterfactual explanations, system performance, risk perception, overreliance on AI

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: AI Trust: Can Explainable AI Enhance Warranted Trust? https://app.cochat.ai/papers/10.1155/2023/4637678
The paper
de Brito Duarte, R., Correia, F., Arriaga, P., & Paiva, A. (2023). AI Trust: Can Explainable AI Enhance Warranted Trust? Human Behavior and Emerging Technologies. https://doi.org/10.1155/2023/4637678

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.