Does relying on AI dialogue systems harm student critical thinking?

Zhai, Wibowo and Li conducted a PRISMA-guided systematic review of 14 articles drawn from ProQuest, IEEE Xplore, ScienceDirect, and Web of Science, examining how student over-reliance on AI dialogue systems affects "critical cognitive capabilities including decision-making, critical thinking, and analytical reasoning." They define overreliance as accepting AI-generated recommendations "without question, leading to errors in task performance," and their synthesis across the 14 studies concludes that "over-reliance stemming from ethical issues of AI impacts cognitive abilities, as individuals increasingly favor fast and optimal solutions over slow ones constrained by practicality."

The mechanism the review assembles runs from named ethical failures to a cognitive habit. Gao et al. (2022) found users accepting hallucinated AI outputs "without validation"; Grassini (2023) found algorithmic bias from skewed training data gets mistaken for objectivity; Dempere et al. (2023) and Dergaa et al. (2023) raised privacy and transparency failures in how AI systems handle student data. Krullaars et al. (2023) argue the downstream effect is motivational as much as cognitive: over-reliance "might diminish students' drive and commitment to learning" and curtails "students' freedom to convey their unique thoughts and viewpoints," while Dergaa et al. and Marzuki et al. (2023) tie regular dialogue-system use to "a decline in abilities of cognitive abilities" and "a diminished capacity for information retention." The through-line the review draws is that unresolved ethical issues don't just create compliance risk — they condition users toward heuristic shortcuts over effortful reasoning.

This extends Does frequent AI tool use reduce our critical thinking? by supplying a named set of triggers — hallucination, bias, privacy, transparency failures — for the offloading that survey measures only correlationally. It converges with Does ChatGPT harm informal learning compared to Google Search? and Does AI assistance weaken our brain's ability to think independently? as a third, literature-synthesis line of evidence for habitual-reliance cognitive erosion, though unlike those two it reports no new data of its own. Against Does polished AI output trick audiences into trusting it?, the review's heuristic-substitution account offers a reason users let polished-looking AI output stand unchallenged in the first place.

As a synthesis of 14 secondary studies' claims, the excerpt gives no effect sizes, sample details, or quality appraisal for those studies, so whether over-reliance causes cognitive decline rather than merely co-occurring with it can't be weighed from this review alone — the claim is only as strong as the literature it pools. The warranted implication is still useful at that strength: interventions aimed at over-reliance should target the specific ethical failures (flagging hallucination, exposing bias, clarifying data use) rather than treating "overreliance" as a single undifferentiated habit to correct.