How Does AI Deception Erode Academic Trust? — The Mediating Role of Psychological Contract Breach
Description
Research Hypothesis 1: H1a. AI origin → stronger O effect; H1b. Fabrication → stronger O effect; H1c. O mediates S → R and is regulated by W. Results: All three hypotheses were confirmed. AI sources were more strongly associated with O than human sources (b=2.99); fabrication was more significant than omission; and O→R showed a significant relationship (b=-0.22). The moderated mediation index was 0.35, CI [0.04,0.72]. Figure 2 shows that the AI + fabrication group had the lowest trust score (1.65), while the human + omission group achieved the highest score (4.07). Second research hypothesis: H2a. Technical implicit regulation from S to O is significant in entity theory but not in growth theory; H2b. Indirect regulatory effects. Results: Both hypotheses were confirmed. The interaction term showed b = –0.42 (p = 0.006). Under the entity theory, fabrication versus omission was significant (b = 0.26); under the growth theory, it was not significant (b = –0.16). Index = 0.29, CI [0.09,0.50]. Figure 3 shows that under the growth theory, the two lines converge, indicating no difference between the two types of deception O. Three research hypotheses: H3a. Early warning buffer effect from S to O; H3b. Indirect regulatory effects of early warning. Results: The hypothesis was not supported, with opposite directions. The interaction term was b = 1.64 (p <0.001). No warning group showed b = 1.02, while the warning group showed b = 2.67 (see enlargement). The index was −0.88, with a confidence interval of [-1.43, −0.43]. Figure 4 shows that under the warning condition, Group O's value dropped to 3.36, whereas the fabricated group still reached as high as 6.03, indicating a pronounced amplification effect. Research Hypotheses 4: H4a. High-quality explanations mitigate the S→O effect; H4b. They reduce the direct negative effect of S→R; H4c. They modulate the indirect effect. Results: Hypothesis H4a was not supported (the O-effect increased from 2.15 to 3.35); Hypothesis H4b was supported (the direct effect decreased from-3.65 to-2.95); Hypothesis H4c was supported (Index = -0.12, CI [-0.24, -0.02]). Figure 5 shows that under high-quality interpretation, the decline in trust significantly decreases.