- AI KATANA Classroom
- Posts
- AI and the Future of Critical Thinking
AI and the Future of Critical Thinking
Navigating opportunities and risks for today’s students
AI KATANA
April 18, 2025
Why the Conversation Matters
Generative AI systems such as ChatGPT, Claude and Gemini entered higher‑education corridors at break‑neck speed. In February 2025 the Higher Education Policy Institute (HEPI) and Kortext reported that 92 % of UK undergraduates now use generative AI tools, up from 66 % in 2024. Only a third have received any formal training on how to use them responsibly.
That rapid adoption is forcing educators to ask a hard question: What happens to the development of students’ critical‑thinking skills when algorithms can draft, summarise and even “reason” on command?
What Do We Mean by “Critical Thinking”?
Most educational psychologists define critical thinking as the ability to analyse, evaluate and synthesise information in order to form reasoned judgements. Facione & Facione’s six‑dimension model (inquisitiveness, open‑mindedness, systematicity, analysis, truth‑seeking and confidence in reasoning) is still widely cited today.
These dispositions underpin the higher‑order processes in Bloom’s revised taxonomy—analysis, evaluation and creation—and are precisely the skills employers say are hardest to automate.
The Double‑Edged Sword of AI
Potential Gains | Potential Risks |
|---|---|
Personalised prompts can scaffold inquiry, nudging students to ask “why” rather than memorise “what”. | Cognitive off‑loading—outsourcing steps of reasoning to the machine—may weaken the very neural pathways behind critical thinking. |
24/7 formative feedback from intelligent tutors encourages iterative revision and metacognition. | Over‑reliance can mask gaps in understanding and increase the temptation to accept surface‑plausible answers. |
Socratic chatbots can model argumentative dialogue at scale. | Bias & hallucinations require skills in source‑checking that many students have not yet mastered. |
What the Evidence Says
Positive or Mixed Effects
Study / Data | Key Finding |
|---|---|
“ChatGPT in the Classroom” (International Journal of AI in Education, 2025) | Master’s students who critiqued AI‑generated essays saw a significant increase in critical‑evaluation marks, whereas students who merely submitted AI drafts did not improve overall grades. |
Student interaction with ChatGPT promotes complex argumentation (Computers & Education, 2024) | Structured prompting led to richer justification chains and counter‑arguments in undergraduate philosophy seminars. |
Meta‑analysis of AI and higher‑order thinking (Interactive Learning Environments, 2024) | Across 30 studies the mean effect size for higher‑order thinking (Hedges g ≈ 0.32) was small‑to‑moderate, with stronger gains in inquiry‑based designs. |
OECD “Putting AI to the Test” Spotlight (2023) | GPT‑4 now outperforms average 15‑year‑olds on PISA reading and science tasks, raising the bar for human comparative advantage in reasoning. |
Interpretation: When AI is used as object—something to evaluate, debug or rebut—it can act as a productive foil, stimulating metacognitive reflection and deeper analysis.
Negative or Null Effects
Study / Data | Key Finding |
|---|---|
Systematic review of dialogue systems (Smart Learning Environments, 2024) | 75 % of students in 327‑participant sample believed heavy reliance on AI reduced their own critical‑thinking capacity; similar proportions cited risks to creativity and ethical writing. |
MDPI study on cognitive off‑loading (2025) | In a survey of 666 learners, AI‑tool use predicted lower critical‑thinking scores (β = –0.42), fully mediated by habitual off‑loading of mental steps. |
Microsoft Research workplace survey (2025) | Knowledge workers reported a non‑linear relationship: moderate AI use enhanced strategic thinking, but heavy use correlated with poorer performance on open‑ended reasoning tasks. |
Guardian higher‑education commentary (2025) | Academics warn that “Google‑level convenience culture” risks a gradual atrophy of human critique, especially in essay‑based disciplines. |
Interpretation: Passive consumption of AI outputs, especially without fact‑checking, appears to erode the struggle that cements learning. The dose and design of AI use are critical moderators.
Why Results Diverge: Five Moderators
Pedagogical Framing
Generative‑writing as draft (students improve it) boosts thinking; generative‑writing as final answer suppresses it.Task Complexity
Gains are larger in ill‑structured problems (e.g., case analysis) than in algorithmic tasks where AI simply supplies the procedure.Student AI Literacy
Higher AI‑literacy scores predict better critical‑thinking outcomes, suggesting training mitigates automation complacency.Assessment Design
Where marks reward process transparency (reasoning steps, source evaluation), students engage more critically with AI output.Frequency & Purpose of Use
The Microsoft and MDPI datasets both show a U‑curve: occasional AI augmentation helps; high‑frequency “mindless prompting” harms.
Recommendations for Educators and Institutions
Action | Rationale |
|---|---|
Embed AI‑literacy modules covering prompt engineering, bias detection and source triangulation. | Students who know how models work are less likely to treat outputs as gospel truth. |
Design “AI‑required” assignments where students must critique, improve or adversarially test an AI answer. | Shifts AI from answer‑generator to thinking catalyst (e.g., compare claim vs. evidence). |
Weight assessment toward reflective commentary (process logs, metacognitive journals). | Forces students to articulate the reasoning that AI cannot supply, bolstering self‑explanation effects. |
Adopt retrieval‑practice & open‑book exams focusing on application and judgement, not recap. | Mitigates the copy‑paste temptation and rewards genuine analysis. |
Implement “critical‑thinking diagnostics” at course entry and exit. | Provides evidence of whether AI‑rich curricula are building or eroding skills over time—and lets departments adjust early. |
Implications Beyond the Classroom
Employability: As GPT‑class models excel at routine reasoning, human comparative advantage will increasingly hinge on argument quality, ethical judgement and interdisciplinary synthesis.
Equity: The HEPI data show that students from less‑privileged backgrounds express more anxiety and receive less guidance on AI use, risking a new digital divide.
Policy: Accreditation bodies may need explicit “AI‑competency” standards that include critical‑thinking outcomes, mirroring emerging OECD frameworks.
Conclusion
AI will not kill critical thinking—but it will punish complacency.
When AI is treated as a co‑participant—something to question, refine and occasionally out‑argue—it can turbo‑charge intellectual growth. When treated as an oracle, it quietly atrophies the muscles of inquiry. The research to date paints a nuanced picture: design‑dependent effects with a small‑to‑moderate average impact, positive or negative depending on use case.
For educators the message is clear: pedagogy, not the algorithm, determines the outcome. With thoughtful scaffolding, generative AI can become the most patient Socratic tutor ever invented; without it, students risk outsourcing the very skills that make them uniquely human. The stakes could not be higher—for tomorrow’s graduates, and for the societies that will rely on their judgement in an AI‑saturated world.
References
Facione, P. A. (1990). Critical Thinking: A Statement of Expert Consensus for Purposes of Educational Assessment and Instruction (The Delphi Report). American Philosophical Association.
Oates, A., & Johnson, D. (2025). ChatGPT in the Classroom: Evaluating Its Role in Fostering Critical Evaluation Skills. International Journal of Artificial Intelligence in Education.
Suriano, R., Plebe, A., Acciai, A., & Fabio, R. A. (2024). Student interaction with ChatGPT can promote complex critical‑thinking skills. Learning and Instruction, 95, 102011.
Dibek, M. İ., Şahin Kurşad, M., & Erdoğan, T. (2024). Influence of artificial‑intelligence tools on higher‑order‑thinking skills: A meta‑analysis. Interactive Learning Environments, 1–23.
Zhai, C., Wibowo, S., & Li, L. D. (2024). The effects of over‑reliance on AI dialogue systems on students’ cognitive abilities: A systematic review. Smart Learning Environments, 11(28).
Gerlich, M. (2025). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1).
Higher Education Policy Institute & Kortext. (2025, February 26). Student Generative AI Survey 2025. HEPI.
Organisation for Economic Co‑operation and Development. (2023). Putting AI to the Test: How Does the Performance of GPT and 15‑Year‑Old Students in PISA Compare? (OECD Education Spotlights No. 6). OECD Publishing.
Microsoft & LinkedIn. (2024). 2024 Work Trend Index Annual Report: AI at Work Is Here. Now Comes the Hard Part. Microsoft WorkLab.
Weale, S. (2025, February 26). UK universities warned to ‘stress‑test’ assessments as 92 % of students use AI. The Guardian.
