CoChat

Cognitive transparency and trust calibration in explainable AI systems: a psychological model of human-AI decision reliance

Open access
Rehman, Usman · 2026 · Frontiers in Psychology

The DOI matches its Crossref record; 42 of the 42 references it cites were found in a registry or an index; 4 of the 80 sources its claims cite were checked for support.

Abstract

Introduction: Explainable artificial intelligence (XAI) is widely expected to improve transparency and support appropriate human reliance on AI-assisted decision-making. However, the psychological processes through which explainability relates to reliance remain insufficiently understood, and explainability, transparency, and trust have largely been examined separately. Objectives: This study develops and tests a psychological model of cognitive transparency in which AI explainability, algorithmic transparency, and perceived AI accuracy are associated with human reliance on AI through cognitive understanding, cognitive trust, and trust calibration, with AI literacy as a boundary condition. Methods: A cross-sectional online survey was completed by 342 AI users in South Korea in May and June 2026. Each participant evaluated one self-selected AI system that they had personally used; all constructs were measured by self-report. The data were analyzed using partial least squares structural equation modeling (PLS-SEM) in SmartPLS 4. Results: AI explainability (β = 0.349) and algorithmic transparency (β = 0.297) were positively associated with cognitive understanding. Cognitive understanding (β = 0.388) and, more strongly, perceived AI accuracy (β = 0.492) were associated with cognitive trust. Cognitive trust was associated with trust calibration (β = 0.631), which was associated with human reliance on AI (β = 0.600). All hypothesized indirect effects were significant, and AI literacy strengthened the association between AI explainability and cognitive understanding (β = 0.192). The model explained 36%–54% of the variance in the endogenous constructs and showed medium predictive power. Conclusion: The findings are consistent with a sequence in which explainable AI relates to reliance through users’ understanding and trust judgments. Because all constructs were self-reported and measured at a single point in time, the results describe associations among users’ perceptions rather than causal effects, independently verified calibration, or objectively appropriate reliance. The findings suggest that AI developers in domains such as healthcare, education, and finance, where calibrated reliance is critical, should pair user-facing explanations with information about system accuracy and with AI literacy support.

Overview

CoChat Summary

TL;DR

This cross-sectional survey of 342 AI users in South Korea tests a PLS-SEM model in which perceived AI explainability and algorithmic transparency relate to human reliance on AI through cognitive understanding, cognitive trust, and (self-assessed) trust calibration, with AI literacy as a moderator. All ten hypotheses were supported, but the authors stress that the design captures associations among self-reported perceptions, not causal effects or objectively appropriate reliance.

Key findings

  • AI explainability (β = 0.349, p < 0.001) and algorithmic transparency (β = 0.297, p < 0.001) were positively associated with cognitive understanding; cognitive understanding (β = 0.388) and, more strongly, perceived AI accuracy (β = 0.492) were associated with cognitive trust.
  • Cognitive trust was associated with trust calibration (β = 0.631) and trust calibration with human reliance on AI (β = 0.600); all hypothesized indirect effects were significant, including the full serial path from explainability to reliance (β = 0.051, 95% CI [0.032, 0.075]).
  • AI literacy strengthened the explainability → cognitive understanding association (interaction β = 0.192, p < 0.001; simple slopes 0.156 at −1 SD vs. 0.541 at +1 SD), though its own main effect on cognitive understanding was not significant (β = 0.078, p = 0.153).
  • The model explained 38.1% of variance in cognitive understanding, 54.3% in cognitive trust, 39.8% in trust calibration, and 36.0% in human reliance; SRMR = 0.048, with medium predictive power overall but weak for reliance (Q²predict = 0.102).

Methods

  • Quantitative cross-sectional online survey conducted in South Korea in May–June 2026; 463 responses received, 121 removed (93 incomplete, 28 duplicates), leaving 342 valid responses (73.9% usable rate). Sample was 54.1% male, 68.1% aged 20–30, 75.1% frequent/very frequent AI users; recruited by non-probability convenience sampling among university students and early-career professionals.
  • Each participant evaluated one self-selected AI system they had personally used (system type not recorded); all eight constructs were measured with three self-developed reflective items on a five-point Likert scale, with AI literacy items referring to AI in general.
  • Data analyzed with PLS-SEM in SmartPLS 4 (path weighting, 3,000 iterations), with measurement model assessment (loadings, α, ρA, ρC, AVE, Fornell-Larcker, HTMT, cross-loadings), full collinearity assessment for common method bias, bootstrapping (5,000 subsamples), PLSpredict and CVPAT for out-of-sample prediction, and a two-stage approach for moderation with simple-slope probing.

Limitations

  • Stated by the authors: the cross-sectional design precludes causal or temporal ordering, and alternative orderings (e.g., trust shaping perceived understanding) are plausible; all constructs were self-reported, so understanding, calibration, and reliance were not objectively verified; one trust calibration item ("I rely on the system appropriately") refers to reliance and may have inflated the calibration–reliance association; participants evaluated self-selected, unrecorded systems; items were newly developed with no pilot test or expert panel review, and the AI literacy measure overlaps with self-efficacy; the English-language questionnaire and a convenience sample of South Korean students and early-career professionals limit generalizability; common method variance cannot be ruled out entirely.
  • Evident from the text: the study did not measure system performance, mental models, or affective trust, so the calibration construct is explicitly a metacognitive self-judgment that "may diverge from actual calibration"; the authors also note reliance was only weakly predicted (Q²predict = 0.102), and the two non-hypothesized indirect effects (full serial path; algorithmic transparency → cognitive trust) are exploratory and require confirmation.

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

Integrity signals

Registry

Crossref · 10.3389/fpsyg.2026.1924913

Cites

42 of 42 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Image 1
CC BY · From the PDF
Image 2
CC BY · From the PDF

Verification

Claims the paper makes

1 Valid · 4 Not valid · 46 Not checked · 111 Own results and interpretations

65 more claims

Create a free account to see all 162 claims and to verify the 68 references still open.

See all claims — free account
  • It may state two separate claims as one. 64
  • Probably not a substantial claim. 48

Create a free account to see these claims.

Citations

Cites

Research works printed but not linked: 3

  1. Hair, 2022, A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM)
  2. Rehman, AI-based performance feedback systems and employee motivation: Evidence from Korean workplaces., J. Indus. Distrib. Bus., 17, 1, 10.13106/jidb.2026.vol17.no2.1 · doi:10.13106/jidb.2026.vol17.no2.1
  3. Ringle, 2024, SmartPLS 4.

Metadata

Authors
Rehman, U.
Affiliations
Hoseo University (KR)
Corresponding author
Rehman, Usman
Venue
Frontiers in Psychology
Published
2026-10-14
Volume
17
Pages
1924913-1924913
ISSN
1664-1078 (ISSN-L)
OpenAlex
W7221796136
Cited by
1
References
39
Citations by year
2026: 1
FWCI
1.08
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Psychology (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
Hoseo University
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Social Sciences › Human Factors and Ergonomics › Human-Automation Interaction and Safety
  • Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
Keywords
explainable AI, algorithmic transparency, trust calibration, cognitive trust, AI literacy, cognitive understanding, human-AI collaboration

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 2026-11-18, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-18 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash).

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

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.