The trust–appropriate reliance relationship is non-linear, with the decline attenuating at higher trust and no reversal within the observed data¶

Claim · All claims
Evidence · 1 study (2 entries) · associational r1–r2 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? A quadratic trust term improved model fit (R2 = .185 vs. .173) and was significant (b = 0.56, p = .011). → Griffin Pitts 2026 q3 i? The fitted curve did not reverse direction, providing limited evidence of a mid-range optimal trust pattern. → Griffin Pitts 2026 (2)

Evidence¶

Griffin Pitts 2026¶

Griffin Pitts, Neha Rani, and Weedguet Mildort. (2026). Trust and Reliance on AI in Education: AI Literacy and Need for Cognition as Moderators. Preprint, accepted to AIED 2026. https://arxiv.org/abs/2604.01114 doi:10.48550/arxiv.2604.01114

q3 · i? · associational · r2

Model-fit comparison in the Results section of the 432-participant regression study; the quadratic model's ΔR2 = .012 and both terms were significant. No standardized effect size is printed, so impact is null.

"For appropriate reliance, the quadratic model explained more variance than the linear model (R2 =.185vs.R 2 =.173;∆R 2 =.012). Both the linear trust term (b=−3.18,SE= 0.43,t(429) =−7.45,p < .001) and the squared term (b= 0.56,SE= 0.22,t(429) = 2.55,p=.011) were statistically significant."

Griffin Pitts 2026 (2)¶

Griffin Pitts, Neha Rani, and Weedguet Mildort. (2026). Trust and Reliance on AI in Education: AI Literacy and Need for Cognition as Moderators. Preprint, accepted to AIED 2026. https://arxiv.org/abs/2604.01114

q3 · i? · associational · r1

The Results section reports the shape of the quadratic fit shown in Figure 3b: the decline attenuated at higher trust and "the fitted curve did not reverse direction within the observed data". No effect size is printed.

"the estimated decline in appropriate reliance was steeper from lower to moderate trust values and then attenuated at higher trust values; however, the fitted curve did not reverse direction within the observed data, providing limited evidence of a mid-range “optimal trust” pattern in this task context."

Discussion¶

Related Claims¶

• Higher post-task trust in an AI assistant predicts lower appropriate reliance during AI-assisted programming problem-solving — a broader claim this one bears on
• AI literacy negatively moderates the trust–appropriate reliance relationship, with the trust slope becoming more negative as AI literacy increases — related
• Need for cognition negatively moderates the trust–appropriate reliance relationship, with the trust slope becoming more negative as need for cognition increases — related