Does Frequent Use of Generative AI Undermine Critical Thinking?

Related concepts: Artificial Intelligence | Cognitive Offloading | Critical Thinking | Psychology

Brainstorm

->AI and the effect is has on the psyche/brain

1. On loneliness

-> Does AI play a part in the loneliness epidemic -> How do GPT transformers ease lonliness -> Is GPT an effective therapist

2. On cognitive function

-> AI and the decline of cognitive function -> AI and the decline on critical thinking -> Effects of AI on language processing in kids

3. AI and addiction

-> Is AI addictive

Non-Ai Related.

-> Impact of porn on a developing brain.

Title: Does frequent use of generative AI for study tasks undermine critical thinking through cognitive offloading?

Introduction:

Generative artificial intelligence (AI) systems, such as ChatGPT, are increasingly embedded into everyday learning environments. Their rapid uptake in higher education has prompted urgent debate about whether such tools enhance or undermine students’ cognitive capacities. Central to this concern is the process of cognitive offloading, whereby individuals externalise mental demands onto technological systems (Risko & Gilbert, 2016). Cognitive offloading is not inherently detrimental - humans have long relied on calculators, maps, or smartphones to ease cognitive load. However, when overused in study tasks, such strategies may compromise the development of critical thinking, a core goal of tertiary education. This essay examines whether frequent reliance on generative AI promotes efficiency at the cost of independent analytical reasoning.

Early studies highlight how digital tools reshape memory processes. Sparrow et al. (2011) demonstrated that when individuals expected information to be retrievable online, they were less likely to retain content knowledge and more likely to remember where to find it. This “Google effect” illustrates a shift from internalizing knowledge to outsourcing recall. Extending this, smartphone research suggests that habitual reliance on devices encourages cognitive miserliness, favoring intuitive “Type 1” processing over deliberate “Type 2” reasoning (Barr et al., 2015). Such patterns suggest that technologies designed to assist cognition may inadvertently displace deeper reflective engagement.

Generative AI introduces a novel dimension: rather than merely storing or retrieving information, these systems actively produce structured outputs, from essays to problem-solving steps. While this can scaffold learning when used judiciously, it also risks fostering automation bias, the tendency to over-trust machine recommendations even in the face of conflicting evidence (Goddard et al., 2012; Lebovitz et al., 2021). Recent findings in clinical and decision-making contexts demonstrate that users often conform their judgments to AI outputs, raising concerns about erosion of personal evaluation skills.

This essay argues that frequent use of generative AI for study tasks does undermine critical thinking via cognitive offloading. Although AI can serve as a supportive tool when integrated with metacognitive strategies, its uncritical or habitual use promotes dependency, reduces analytical engagement, and heightens overreliance on automated systems. Accordingly, the implications extend beyond individual students to broader questions about education’s role in cultivating independent thought in an AI-saturated era.

Evidence from transactive-memory research shows that easy access to information changes what is remembered. In Sparrow et al.’s (2011) experiments, participants who believed information would remain accessible later were less likely to recall the content itself and more likely to remember “where” it could be found, consistent with a shift toward externalized memory stores. In education, this tendency can translate to students recalling “which prompt produced the answer” rather than the answer’s structure or justification - an early sign that retrieval is being displaced by tool-location cues rather than internalized knowledge.

Related work on smartphones supports a broader “cognitive miserliness” account: people often prefer intuitive, low-effort strategies when a device makes answers immediately available. Barr et al. (2015) found that heavier reliance on smartphones for search was associated with lower engagement in analytic (Type 2) reasoning on classic reasoning problems, consistent with substituting device-mediated lookups for reflective problem solving. Though correlational and open to alternative explanations such as inattention, the pattern coheres with the idea that easy external access can reduce practice with analysis unless tasks explicitly demand it.

Importantly, offloading does not merely affect factual recall; it can erode deeper structural knowledge when tools replace the need to build internal models. In navigation, habitual GPS use predicts poorer hippocampal-dependent spatial memory and landmark encoding (Dahmani & Bohbot, 2020). By analogy, delegating decomposition, evaluation, and synthesis to generative AI can impair students’ internal “maps” of an argument space-how claims, warrants, and counter-arguments fit together-unless pedagogy deliberately requires students to construct and defend those maps themselves.

A second pathway is overreliance on AI outputs (automation bias). Reviews of human–AI collaboration document a robust tendency to favor automated recommendations even when they are wrong, with explanations sometimes increasing acceptance without reliably improving accuracy (Romeo & Conti, 2025). In applied studies, local saliency explanations (e.g., highlights or boxes) can speed agreement with AI and raise accuracy when advice is correct, but they can also accelerate alignment with incorrect advice-lowering performance (Prinster et al., 2024; see also Cecil et al., 2024, in personnel selection). For student use, this means polished rationales or confident tone in a model’s answer may inflate perceived credibility, nudging learners to accept conclusions rather than interrogate premises and evidence. Guardrails must therefore address not only information quality but also how presentation features shape trust and deference.

Taken together, these literatures suggest a conditional conclusion. Unstructured, default use of generative AI for core study tasks likely undermines critical thinking via cognitive offloading and automation bias. However, structured integration can invert the risk: when students must generate counter-prompts, justify acceptance or rejection of outputs, and document revisions, AI becomes a scaffold for metacognition rather than a substitute for it. Contemporary education frameworks describe practical designs- e.g., requiring students to log prompts, critique model reasoning, and submit side-by-side human/AI drafts with reflective commentary - that maintain analysis “in the loop” (Ní Uanaicháin & Aouad, 2025). The pedagogical challenge is not whether to allow AI, but to obligate students to use it in ways that exercise evidence evaluation, conceptual linkage, and self-explanation.

References

Barr, N., Pennycook, G., Stolz, J. A., & Fugelsang, J. A. (2015). The brain in your pocket: Evidence that smartphones are used to supplant thinking. Computers in Human Behavior, 48, 473–480. https://doi.org/10.1016/j.chb.2015.02.029 ScienceDirect

Cecil, J., Lermer, E., Hudecek, M. F. C., Sauer, J., & Gaube, S. (2024). Explainability does not mitigate the negative impact of incorrect AI advice in a personnel selection task. Scientific Reports, 14, 9736. https://doi.org/10.1038/s41598-024-60220-5 DOI

Dahmani, L., & Bohbot, V. D. (2020). Habitual use of GPS negatively impacts spatial memory during self-guided navigation. Scientific Reports, 10, 6310. https://doi.org/10.1038/s41598-020-62877-0 BIC MNI McGill

Ní Uanaicháin, D. M., & Aouad, L. I. (2025). Generative AI in education: Rethinking learning, assessment & student agency for the AI era. Thresholds, 48(1), 111–132. https://files.eric.ed.gov/fulltext/EJ1468033.pdf

Prinster, D., Mahmood, A., Saria, S., Jeudy, J., Lin, C. T., Yi, P. H., & Huang, C.-M. (2024). Care to explain? AI explanation types differentially impact chest radiograph diagnostic performance and physician trust in AI. Radiology, 313(2), e233261. https://doi.org/10.1148/radiol.233261 CTisus

Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002 UCL Discovery

Romeo, G., & Conti, D. (2025). Exploring automation bias in human–AI collaboration: A review and implications for explainable AI. AI & Society. Advance online publication. https://doi.org/10.1007/s00146-025-02422-7 SpringerLink

Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google effects on memory: Cognitive consequences of having information at our fingertips. Science, 333(6043), 776–778. https://doi.org/10.1126/science.1207745 Harvard Scholar

2025 PSY1102 ESSAY RUBRI

Arguments:

1. The idea of cognitive offloading and the effect it has on critical thinking has already been demonstrated by $x$ below.

• Offloading already impacts critical thinking -> You arent critically analyzing a task, rather just putting it through pathways. -> I’d say part of math is also NOT critical thinking
• https://www.sciencedirect.com/science/article/abs/pii/S0747563215001272
• Evidence for how Navigational Aids impair memory and does have an effect when offloading. At least when it comes to information retention
• The perception of AI systems & the damages it has. Cognitive phenom where people display an overreliance on automated systems favoring automated recommendations over their own judgement even when displayed otherwise.
• Discusses how Smartphones Supplant thinking and how critical thinking is made redundant -> Symbiotic relationship between humans and technology. Like the above article…

PSY1102 Essay - AI & the Psyche (Due: 18/09/2025)

Phase (rubric-aligned)	Focus / Output	Start	End
Topic refine & aim	Lock a narrow title + one-sentence aim and final position	04/09	05/09
Research – map & gather	Scope related areas; search, screen, and select relevant scholarly sources	05/09	09/09
Research – extract & evaluate	Extract key findings; appraise methods/limits; organise evidence to thesis	08/09	12/09
Body drafting (60%)	Argument & Drafting	10/09	14/09
Introduction (10%)	Significance, key terms, aim, and clear position	12/09	13/09
Conclusion (10%)	Synthesis, final position, limitations/implications	15/09	15/09
Written expression (15%)	Clarity, coherence, signposting, academic tone	12/09	17/09
Referencing (APA7, 5%)	In-text + reference list; ≥4 scholarly beyond starters; formatting checks	12/09	17/09
Finalise & submit	Proof, export, portal check, submission	18/09	18/09

Articles to read

Cognitive Offloading

• https://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(16)30098-5
• https://scholar.harvard.edu/files/dwegner/files/sparrow_et_al._2011.pdf - Knowing where information can be found, and less about knowing information
  • Discuss the idea of how Google used to offload information and LLMs now replacing google
    • Research: Attempts to analyze whether the brain will default to “Googling” or breaking down a problem and thinking about it.
    • Experiment 2: Between-Subject experiment - Only one group experiences x. One group thought their data was erased -> They had better recall.
    • Experiment 3: Tested memory of participants ->
  • Evidence:
    • It would seem from this pattern that people don’t remember “where” when they know “what” but do remember where to find the information when they don’t recall it. This is preliminary evidence that when people expect information to remain continuously available (such as we expect with Internet access), they are more likely to remember where to find it than to remember the details of the item.
    • Relying on our computers and the information stored on the Internet for memory depends on several of the same transactive memory processes that underlie social information-sharing in general. These studies suggest that people share information easily because they rapidly think of computers when they find they need knowledge (experiment 1). The social form of information storage is also reflected in the findings that people forget items they think will be available externally and remember items they think will not be available (experiments 2 and 3). T
  • [ ]
• https://www.sciencedirect.com/science/article/abs/pii/S0747563215001272
  • Discusses how Smartphones Supplant thinking and how critical thinking is made redundant -> Symbiotic relationship between humans and technology. Like the above article…
    • Research:
    • Interesting Points:
      • Critical analysis vs Knowledge -> We’d have a ton of knowledge, we just cant critical analyze them
      • Type 1 Cognitive process -> Intuitive
      • Type 2 -> Working memory required
      • Thus concluded that mental shortcuts are how human brains work (cognitive miserliness)
      • [ ]
    • Evidence
      • Extremely important discussion as a subset of students were also assessed. High levels of boredom led to smartphone usage which causes cognitive miserliness, as the brain expends some energy to doing that instead of thinking.
        • Counter point 1: Could just be due to inattention…? Why would they conclude Cognitive Miserliness instead..?
        • https://www.sciencedirect.com/science/article/pii/S0747563215001272#b0105 - Discuss this paper too.
• https://www.sciencedirect.com/science/article/pii/S0747563215001272#b0105
  • More on Smartphones and how it effects thinking
    • Points:
• https://link.springer.com/content/pdf/10.1007/s00146-025-02422-7.pdf
  • AI Overreliance & Automation bias
    • The perception of AI systems & the damages it has. Cognitive phenom where people display an overreliance on automated systems favoring automated recommendations over their own judgement even when displayed otherwise.
    • Experiment
      • Users change answers to match AI recommendations. ->Weight of Advice, listens to AI more..? https://www.nature.com/articles/s41746-021-00385-9
      • Has two groups - physicians and radiologists ->Asked to view a sample, AI then came up with a report ->experts(radiologists) were more skeptical to Ai recommendations than humans.
    • Evidence
    • Discussion
    • [ ]
• https://epub.ub.uni-muenchen.de/125262/

Concepts:

1. Cognitive Offloading -> Cell.com,

   • DE

   • D: Theory where you use a physical item/trigger to offload cognitive demand of a task

   • E: Setting an Alarm to do the dishes after 15 minutes of study/Using GPT to create a schedule.

     • A computer works by having a set amount of RAM/Memory -> Similarly, according to Cell, memory works much the same way, you could break the RAM into a multiple smaller chunks working on selected tasks but there is a clear limit. -> Possibility of increasing the level? - YES! -> Jacob Collier & Hank Green interview, watch he Collier Sings, Listens and plays the guitar then cognitive offloads most of the understanding/learning.
     • Possible to increase the RAM instead..?

   • Paper discusses:

     • Active Memory
     • Prospective Memory
     • Intention Offloading
     • ‘Minimal Memory’ theory of cognition -> Theoretical framework where

   • E:

     • https://www.tandfonline.com/doi/abs/10.1080/13875868.2015.1059432
     • Evidence for how Navigational Aids impair memory and does have an effect when offloading. At least when it comes to information retention