The New Education Challenge: Staying Critical When AI Makes Thinking Easier

Generative AI has rapidly moved from novelty to routine academic infrastructure, but the key question for universities is shifting from whether students use AI to how they judge what it produces. In "From Critical Thinking to Passive Acceptance: The Moderating Role of Decision Fatigue in Students' Use of Generative AI," published in the European Journal of Investigation in Health, Psychology and Education, Marco Zuin of Istituto Universitario Salesiano Venezia and Vanessa Donadel of LUISS University examine when students critically assess AI-generated material and when they simply accept it.

The findings challenge the assumption that teaching critical thinking alone will protect students from overreliance on generative AI. Critical thinking generally reduced passive acceptance of AI output, but its protective value weakened under conditions of decision fatigue, particularly in tasks requiring students to review, evaluate or supervise AI-generated material. The implication is significant: responsible AI use may depend not only on what students know, but also on whether they have enough cognitive capacity left to apply that knowledge.

The study comes as generative AI use becomes deeply embedded in university life. Existing research cited by the authors suggests that roughly 70% to 90% of students use tools such as ChatGPT for study-related activities, including information searches, text generation and synthesis. The scale of adoption makes the quality of human oversight increasingly important, because students are no longer simply completing academic tasks themselves; they are often evaluating, refining and approving work produced with machine assistance.

AI Is Shifting the Student's Role From Producing Work to Supervising It

The research is built around a subtle but important change in the nature of learning with AI. Generative systems can reduce the effort required to generate ideas, search for information or draft text, but they do not eliminate the need for human judgment. Instead, the cognitive burden shifts from producing the answer to evaluating whether the answer is reliable, relevant and appropriate.

The authors draw on this distinction to examine "passive acceptance", the tendency to use AI-generated output without substantial personal modification or revision. Critical thinking is treated not simply as an academic skill, but as a disposition involving openness, reflective skepticism, willingness to question evidence and the habit of checking information before accepting it. In an AI-mediated learning environment, those characteristics become the human quality-control layer separating assistance from dependence.

The study surveyed 383 Italian university students drawn from 35 institutions, with an average age of 22. Of these, 332 students, or 86.7%, reported using generative AI during the previous three months and were assessed on six activities: idea generation, text generation, text revision, quiz creation, receiving feedback and information search. Data were collected during March and April 2026.

The researchers then examined whether critical thinking and decision fatigue predicted students' tendency to accept AI output without modifying it. Decision fatigue refers to the cognitive exhaustion that can follow repeated decision-making, leaving people with fewer mental resources for evaluation, reflection and self-control. Rather than assuming fatigue directly causes passive AI use, the study tests whether fatigue changes the ability of critical thinking to function effectively.

Critical Thinking Helps, But Its Protective Power Is Not Constant

Across all six academic activities, higher critical-thinking disposition was generally associated with lower passive acceptance of generative AI output. The relationship was statistically clearest for information search, suggesting that critical thinking may matter particularly when students must assess, compare and integrate information rather than simply generate content.

The educational risk from AI is not simply that students will "use it too much." The deeper concern is that they may stop interrogating the material it produces. When learners retain ownership of verification and judgment, generative AI can function as a scaffold; when they simply accept outputs as sufficiently authoritative, it can become a substitute for the very cognitive processes education is intended to develop.

The study also reinforces an emerging view that AI changes where intellectual effort is concentrated. Instead of searching for information manually, students increasingly need to verify it; instead of constructing every answer independently, they need to integrate machine-generated content into a wider argument; and instead of merely performing a task, they may need to supervise a system performing part of it. That turns critical oversight into a central academic capability rather than an optional safeguard.

However, critical thinking was not equally protective across all tasks. The researchers found no significant predictors for passive acceptance in text generation, while the relationships differed across activities such as information search, revision and feedback. This task-specific variation suggests that universities should be cautious about treating "AI use" as a single educational behavior with a single set of consequences.

Decision Fatigue Can Weaken the Safeguard Universities Rely On

What happens when students are cognitively tired? Decision fatigue did not independently predict passive acceptance of AI-generated output across the academic activities examined. However, it altered the relationship between critical thinking and passive reliance in several tasks, meaning fatigue appears to affect whether students can actually put their critical-thinking disposition into practice.

The strongest evidence emerged for text revision and receiving feedback. In both activities, the moderating effect of decision fatigue remained significant even after the researchers applied multiple statistical corrections designed to reduce the risk of false-positive findings. Evidence for information search and quiz creation was weaker and survived only the less conservative correction procedure, while no moderation effect emerged for idea generation or text generation.

Text revision provides the sharpest illustration. At low levels of decision fatigue, greater critical thinking was associated with significantly less passive acceptance of AI output. At high levels of fatigue, that relationship reversed in the model, with greater critical thinking associated with more passive acceptance, producing what the study describes as a robust interaction.

For information search, the protective relationship between critical thinking and passive acceptance was present when fatigue was low or average but effectively flattened when fatigue was high. Students with relatively strong and weak critical-thinking dispositions then behaved more similarly in their tendency to accept AI output.

The results are task-specific, and some relationships require replication, but the broader insight is powerful: possessing critical-thinking ability or disposition does not guarantee that students will consistently deploy it, especially when the cognitive cost of checking AI output becomes high.

Universities May Need to Design for Cognitive Sustainability, Not Just AI Literacy

The findings create a practical challenge for higher education. Universities have understandably focused on AI literacy, academic integrity and critical thinking, but the study suggests that these interventions may be incomplete if they ignore workload and cognitive conditions. A student may understand hallucinations, bias and verification requirements yet still skip scrutiny when exhausted by repeated choices, deadlines and information overload.

It raises questions about assessment design. If generative AI shifts students toward constant supervision of machine-generated content, educators may need to consider the cumulative cognitive burden created by requiring learners to repeatedly verify, compare, edit and judge AI outputs. Teaching students to challenge AI is important, but institutions may also need to structure assignments so that careful evaluation remains cognitively feasible rather than becoming another layer of invisible workload.

The implications extend beyond universities. Employers increasingly expect workers to use AI systems while retaining responsibility for final decisions, and the same tension between automation and human oversight applies to professional settings. If cognitive fatigue weakens quality control, then organizations deploying AI may need to treat workload, decision density and opportunities for reflection as part of responsible AI governance rather than merely employee-wellness concerns.

For developing countries and the wider Global South, the issue has an additional dimension. Expanding access to generative AI could help reduce educational resource gaps, but unequal learning environments may determine whether students use these systems critically or depend on them passively. Institutions with overcrowded classrooms, high student workloads or limited faculty support may therefore need to pair AI access with deliberate investments in teaching design, digital literacy and human supervision.

The research itself warrants cautious interpretation. It is cross-sectional and based on convenience sampling in Italy, meaning it cannot establish causation or be assumed to represent students in other countries. Reliance on AI, critical thinking and decision fatigue were self-reported, the Italian critical-thinking measure was still undergoing validation, and the final statistical models explained only around 6% to 9% of the variation in passive acceptance.

The limitations also point toward the next research agenda. Experimental and longitudinal studies could test how fatigue changes real-time verification behavior, whether students become more vulnerable to incorrect AI output after sustained cognitive effort, and which instructional approaches preserve critical oversight. Comparative studies across education systems would also help determine whether workload, teaching cultures and patterns of AI adoption alter these relationships.