Trust and Reliance on AI in Education: AI Literacy and Need for Cognition as Moderators

Synthesis: Pitts, Rani & Mildort (2026, AIED) show with 432 undergraduates that higher Trust in an AI assistant is associated with lower appropriate reliance: students who trusted the assistant more were worse at discriminating correct from misleading AI suggestions during Python problem-solving. The relationship is non-linear and moderated by AI literacy and need for cognition — trust is not a safe proxy for appropriate use.

The experiment

• 432 undergraduates solved Python output-prediction problems with recommendations + explanations from an AI chatbot that included accurate and intentionally misleading suggestions
• Appropriate reliance measured behaviorally: accepting correct suggestions, rejecting incorrect ones — a form of calibration
• Surveys captured trust, AI literacy, need for cognition, programming Self-Efficacy, programming literacy

Findings

• Non-linear trust→reliance relationship: higher trust → lower appropriate reliance (weaker discrimination between correct/incorrect recommendations)
• Moderators: AI literacy and need for cognition significantly shaped how trust translated into reliance behavior
• Implication: interventions should target calibration — instructional and system supports that encourage reflective evaluation of AI assistance during Problem Solving

What this means for practice

• Instructors. Do not read student trust as a proxy for good judgment: in this study higher post-task trust went with lower appropriate reliance (r = -.42, p < .001) and lower task accuracy (r = -.38), so the most confident AI users were the least selective.
• Instructors. Make verification part of the task rather than advice at the side. Since students accepted 86.03% of misleading recommendations while rejecting only 2.37% of correct ones, build in cognitive forcing functions — committing to an answer before seeing the recommendation, or justifying agreement or disagreement afterward.
• Instructors. Aim calibration support at students who already approach AI cautiously: AI Literacy and need for cognition predicted better appropriate reliance mainly at low trust, and that advantage narrowed as trust rose.
• Learners. Distinguish being able to program from being able to judge an AI suggestion: task accuracy (61.21%) and appropriate reliance (61.77%) landed at almost the same level, so following the assistant well is not the same as solving the problem well.
• Researchers. Extend reliance measurement beyond a single score. The accept/reject measure collapses careful checking, prior knowledge, and superficial cue use into one number, so pair it with justifications, confidence ratings, or response times.

Limitations

• A single-session laboratory experiment: 432 undergraduates from one institution (University of Florida) completed 14 output-prediction problems with a fixed mix of 6 misleading and 8 accurate recommendations, and the initial recommendations were pre-programmed in a Wizard-of-Oz design (follow-up turns used gpt-3.5-turbo-0125). Reliance across longer, open-ended assignments — planning, debugging, revision — was not observed.
• Every learner characteristic and the trust measure came from self-report 7-point Likert scales aggregated per participant, which capture broad perceptions and tendencies rather than the skill of verifying a recommendation under time pressure.
• The appropriate-reliance score records only observable accept-versus-reject decisions relative to recommendation correctness, so two students could reach the same score by very different strategies; no process data (think-aloud protocols or interaction logs) was collected.
• Two of the four proposed moderators were not distinguishable from zero (programming self-efficacy p = .121; programming literacy p = .437), and the Johnson-Neyman boundaries fell far into the lower tail of the moderator distributions (z = -2.37 for AI literacy, -2.10 for need for cognition), so the moderation story rests on a narrow region of the sample.

Citation

Pitts, G., Rani, N., & Mildort, W. (2026). Trust and Reliance on AI in Education: AI Literacy and Need for Cognition as Moderators. AIED 2026.