The association between AI usage and critical thinking in AI contexts: examining the mediating roles of automation bias and cognitive offloading
JOURNAL OF COMPUTING IN HIGHER EDUCATION, cilt.38, sa.4, ss.1-12, 2026 (SSCI, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 38 Sayı: 4
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s12528-026-09507-2
- Dergi Adı: JOURNAL OF COMPUTING IN HIGHER EDUCATION
- Derginin Tarandığı İndeksler: Social Science Premium Collection (ProQuest), Education Collection (ProQuest), Education Source Ultimate (EBSCO), Scopus, Technology Collection (ProQuest), Aerospace Database, Agricultural & Environmental Science Database, Social Sciences Citation Index (SSCI), IBZ Online, Education Abstracts, ERIC (Education Resources Information Center), INSPEC, EBSCO Education Source, Psycinfo
- Sayfa Sayıları: ss.1-12
- Bursa Uludağ Üniversitesi Adresli: Evet
Özet
With the proliferation of generative artificial intelligence (AI) tools, a growing body of research reports negative associations between the use of these technologies and students' critical thinking. However, the psychological pathways that may account for this association remain insufficiently understood. This study examines the relationship between university students' generative AI use and their critical thinking in AI-usage contexts, and tests whether the perception of automation bias and cognitive offloading statistically mediate this relationship. A cross-sectional survey was administered to 526 undergraduate students recruited from three universities in western Türkiye. Generative AI usage intensity and critical thinking in AI usage were measured with previously published instruments; the automation bias perception and cognitive offloading measures were developed for this study across two independent pilot samples, and their psychometric properties are reported in detail. Structural equation modeling indicated that generative AI usage intensity was negatively associated with critical thinking in AI contexts, with a total effect (β = -.40, p < .001) and a substantially smaller direct effect (β = -.12, p = .014). Indirect paths accounted for a considerable portion of the total association through automation bias perception (β = -.15, p < .01) and cognitive offloading (β = -.13, p < .01). The model accounted for 50% of the variance in critical thinking in AI usage. Because the design is cross-sectional, these results describe patterns of association rather than causal or temporal sequences; alternative model specifications, including a reverse ordering, fitted the data comparably well. Instructional strategies targeting trust calibration and deliberate cognitive engagement are discussed as hypotheses for future intervention research.