IMCC Journal of Science

Abstract: 

The integration of Large Language Models (LLMs) in STEM education introduces a paradox: while AI accelerates research, it risks eroding epistemic vigilance due to “hallucinations.” This descriptive-correlational study examines the relationship between Generative AI usage frequency and source verification habits among 142 Grade 12 STEM students in Davao City, Philippines. Utilizing the validated AI-Verification Index (S-CVI=0.94), results indicate a high dependency on AI (M=4.12) contrasted with low verification behaviors (M=1.89). A significant negative correlation (r = -0.68, p < .001) confirms that increased reliance on AI tools is associated with a decline in fact-checking protocols, such as validating Digital Object Identifiers (DOIs). These findings suggest a prevalence of “automation bias,” where algorithmic fluency masks factual inaccuracy. The study concludes that STEM curricula must pivot from AI prohibition to “AI auditing,” recommending mandatory verification logs to restore academic integrity in the age of algorithmic learning.

Keywords: automation bias; epistemic vigilance; generative AI; STEM education