Designing for AI literacy: psychological mechanisms and self-determination.

INTRODUCTION

This study examined how instructional approach (skills-based instruction [SBI] versus thinking-based instruction [TBI]) and task type (structured tasks [STT] versus open-ended tasks [OET]) influence university students' AI literacy performance through autonomy, competence, and relatedness.

METHODS

A randomized 2 × 2 between-subjects experiment was conducted with 320 university students assigned equally to four conditions. Participants received video-based instruction and subsequently completed an AIGC image-generation task. AI literacy performance was assessed using an objective task rubric, while the three basic psychological needs were measured using validated self-report scales.

RESULTS

Two-way ANOVA showed significant main effects of instructional approach and task type and a significant interaction effect. TBI produced higher performance than SBI under both task conditions, with a larger advantage under OET. In the structural model, TBI positively predicted autonomy, competence, relatedness, and AI literacy performance. OET positively predicted autonomy, relatedness, and performance but did not significantly predict competence. Competence was the strongest psychological predictor of performance, followed by autonomy, whereas relatedness had no independent effect. Bootstrap analysis showed that instructional approach operated indirectly through autonomy and competence, while task type operated primarily through autonomy. The single factorial interaction term did not retain a significant direct effect after the psychological needs were included. fsQCA further identified multiple sufficient configurations leading to high AI literacy performance.

DISCUSSION: These findings demonstrate that instructional design affects AI literacy through differentiated motivational pathways rather than through a uniform activation of all three psychological needs. Within this single-institution, short-duration experiment, the findings should be interpreted cautiously and require longer, multi-site replication.