AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking

Synthesis: Gerlich (2025) investigates whether frequent use of AI tools erodes critical thinking, using cognitive offloading as the proposed mediating mechanism. Using a mixed-method design with 666 valid UK participants (convenience + purposive sampling across age groups, educational backgrounds, and occupations) plus 50 semi-structured interviews, the study combined a 23-item questionnaire (drawing on the Halpern Critical Thinking Assessment and Terenzini's self-reported measures) with thematic analysis. Both quantitative and qualitative analyses converge on a central finding: heavier AI tool usage is associated with lower critical thinking, and cognitive offloading — delegating memory, decision-making, and information retrieval to external systems — significantly mediates this negative relationship. The results were robust across ANOVA, correlation, multiple regression, and random forest regression, and were corroborated by interview themes of AI Dependence, Cognitive Engagement, and Ethical Concerns.

Key Findings

• Frequent AI tool use correlates negatively with critical thinking (r = −0.68); ANOVA showed significant differences in critical thinking across AI usage levels (p < 0.001). Hypothesis 1 confirmed.
• Cognitive offloading partially mediates the relationship: total effect b = −0.42 (p < 0.001), indirect effect through offloading b = −0.25 (p < 0.001), direct effect b = −0.17 (p < 0.01). Hypothesis 2 confirmed.
• Strong inter-variable correlations: AI use ↔ offloading (r = +0.72); AI use ↔ critical thinking (r = −0.68); offloading ↔ critical thinking (r = −0.75).
• Younger participants (17–25) show higher AI dependence and cognitive offloading but lower critical thinking than older participants (46+), who used AI less and scored higher.
• Higher educational attainment is associated with better critical thinking regardless of AI use (education ↔ critical thinking r = +0.34); a significant interaction (β = 0.02, p = 0.046) indicates education moderates and mitigates the negative AI effect.
• Multiple regression: AI use negatively predicted critical thinking (β = −1.76, p < 0.001, R² = 0.244); random forest (R² = 0.370) flagged AI use and offloading as dominant predictors (permutation p = 0.0099).
• Qualitative themes — AI Dependence, Cognitive Engagement, Ethical Concerns — corroborated the quantitative results; lower-attainment users were more likely to trust AI outputs without scrutiny.

What this means for practice

• Instructors. Pair every AI task with work the tool cannot do for the learner — explicit Critical Thinking exercises and Active Learning — because heavier AI use correlated with lower critical thinking (r = −0.68) and the association ran substantially through cognitive offloading (indirect effect b = −0.25, p < 0.001).
• Instructors. Teach students to notice what they delegate: offloading mediated the AI–critical-thinking link, so short metacognitive prompts asking what was outsourced and what was worked out independently target the mechanism rather than the tool.
• Faculty developers. Build AI competency training that keeps cognitive engagement in view alongside tool fluency; higher educational attainment moderated the negative AI effect (interaction β = 0.02, p = 0.046), so instruction is a lever on the outcome.
• Administrators. Direct structured critical-thinking instruction at the youngest cohorts: 17–25-year-olds reported the highest AI dependence and offloading and scored lowest on critical thinking, while 46+ participants used AI less and scored higher.
• Researchers. Treat Trust in AI as a measurable risk factor, since the study extends the "Google effect" and cognitive offloading literature by linking trust in AI to heavier offloading and weaker scrutiny of outputs.

Limitations

• Both key variables are self-reported — AI tool usage and critical thinking were captured through a 23-item questionnaire — so the associations may partly reflect shared method variance rather than independent measurement.
• The 666 UK participants were recruited by convenience and purposive sampling; the author names potential sample bias as a limitation and cautions against broad generalization.
• The design is cross-sectional and correlational: mediation is modeled statistically, not manipulated, so the author calls for longitudinal and experimental work before causal claims about AI eroding critical thinking.
• Even the best-fitting random forest model accounted for 37% of the variance in critical thinking (R² = 0.370), leaving most of the outcome unexplained by AI use and offloading.

Citation

Gerlich, M. (2025). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(6), 6.