By Christina McKendrick
Introduction
ChatGPT was released to the public in 2022, thrusting generative AI into the mainstream overnight and leaving educators on the frontlines of a rapidly evolving digital landscape. While a growing body of guidance continues to emerge, navigating this space still feels like the Wild West. As students increasingly adopt AI to complete everyday tasks, teachers are left weighing its productivity benefits against growing concerns over academic integrity. But beneath the surface of cheating and shortcuts lies a deeper, more insidious question: what is the outsourcing of thinking doing to actual human cognition?
Synthesizing findings from empirical and secondary research, this paper explores the cognitive risks of generative AI on student development. It seeks to bridge these findings with the Canadian School Libraries’ Digital Media and Literacy Toolkit (DMT), providing teacher-librarians with a practical framework for implementing teaching strategies that prioritize human reasoning within the Library Learning Commons (LLC).
What’s the Problem with AI?
Generative AI is already being used widely by students. According to data compiled by the Center for Universal Education at Brookings Institution, as of 2025, 91% of teens in the United States, 94% of teens in Australia and 67% of teens in the United Kingdom have reported using generative AI. In Taiwan, 58% of junior and senior high school students use AI at least once a week (Burns et al., 2026). Similarly, it was reported that 64% of young people (ages 16–24) across the EU reported using generative AI (Chounta, 2025) and according to a survey conducted by the accounting firm KPMG, 73% of Canadian post-secondary students reported using generative AI for school work (KPMG, 2025). The Brookings Institution report also noted that due to a lack of policy, both students and teachers often hide their use of AI for fear of being perceived as less competent (Burns et al., 2026). This would mean that in any of these hidden instances, students do not have any teacher guidance. While none of these numbers specifically reflect Canadian secondary students, the fact of the matter is that students across the globe are using AI in staggering numbers- with or without any guidance. Left unchecked, this could pose a serious problem.
Some may argue that AI is an incredible tool and students should harness its capabilities to enhance their work. While generative AI does come with some great benefits to productivity, it also comes with some pretty serious risks. Of the post-secondary students who participated in the survey conducted by KPMG, 48% said they believe their critical thinking skills have deteriorated since they started using AI (KPMG, 2025), and they have good reason to be concerned.
A randomized controlled trial involving nearly 1,000 high school students in Türkiye compared math results of students using a standard GPT-4 interface against a version designed for tutoring that provided hints instead of direct answers. They compared both groups back to a control group, who did all of their studying without any access to AI. The group of students using a standard version were more productive in completing practice problems, but ultimately scored 17% lower than their peers on the final closed-book evaluation. Researchers identified a “Crutch Effect,” where students used standard generative AI to bypass the productive struggle required for learning. This led to metacognitive laziness, where critical phases of learning—like diagnosing misunderstandings or evaluating reasoning—were skipped. They also noted that students using standard AI often did not perceive any negative effect on their learning, suggesting they are unaware of how these tools can impede their development (Bastani et al, 2025). This controlled trial demonstrates that standard AI tools can inadvertently undermine student learning by fostering a false sense of competence. Students trade away deep cognitive engagement for short-term convenience, remaining largely unaware of how it negatively impacts them.
In a briefing prepared for the EU’s Committee on Culture and Education (CULT), Dr. Irene-Angelica Chounta warns that “A student can be autonomous as a technology user while being deeply heteronomous as a cognitive agent — dependent on AI not just for tools but for the very judgements that constitute independent thinking.” (Chounta, 2025, p.5). She outlines some key dangers of the removal of effortful thinking by students, highlighting the negative impact on cognitive development and fundamental skills (including attention, memory, and problem-solving), and a loss of metacognition and autonomy (Chounta, 2025). This warning exposes a dangerous digital illusion: students can effortlessly operate an AI interface while completely surrendering their internal reasoning. Technical fluency can mask a total loss of independent thought and fundamental cognitive development.
In his 2025 article, Cognitive Atrophy Paradox of AI–Human Interaction: From Cognitive Growth and Atrophy to Balance, Igor Kabashkin explores the phenomenon of cognitive atrophy (sometimes also referred to as cognitive decline or debt). He explains that increased offloading of creative and analytical tasks to AI eventually results in structural cognitive weakening analogous to muscular atrophy, where unused neural circuits and the capacity for original thought deteriorate through disuse. This dependency manifests through automation bias, where machine outputs are automatically trusted, comprehension erosion and decline in intellectual curiosity (Kabashkin, 2025). While adults risk losing pre-existing competencies through excessive delegation, developing students face an even graver consequence: the complete prevention of cognitive development before critical thinking skills can even be established.
Studies have also shown that young people are particularly susceptible to these risks. In Michael Gerlich’s 2025 study, AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking, they found that younger participants (ages 17–25) were significantly more vulnerable to the negative cognitive effects of AI tools than older groups, displaying more dependency and overall lower cognitive scores (Gerlich, 2025). This makes sense given that the human brain is still developing well into the 20s. If a young individual has been using generative AI unchecked for a while now, they’ve likely fallen victim to some of the dangers previously outlined. This does reinforce a serious need for educators to guide students on how to use AI in a way that does not bypass mental effort.
While generative AI is being rapidly and widely adopted by students around the globe, its unguided use carries severe developmental risks. From inducing metacognitive laziness, cognitive heteronomy, and automation bias to driving cognitive atrophy and the erosion of creativity in young, developing minds. Unchecked AI reliance bypasses the essential productive struggle required for learning. Because young people are especially vulnerable to these effects, intentional educator guidance is desperately needed to protect student thinking, preserve foundational cognitive skills, and protect student intellectual autonomy.
Promising Practices: Good Cognitive Hygiene Rituals
Generative AI exists. There’s no turning back now. We need to find a way to protect the cognitive development of students in a world where generative AI is readily available. So what can be done?
The changes required to completely protect student development is beyond the scope of what can be achieved at the individual school level alone. The AI that is most available to students are the general versions which allow for the bypassing of mental effort, and learners are heavily incentivized to continue using AI as a cognitive surrogate due to the perceived transactional nature of education – completing assignments for grades (Burns et al., 2026). Increased trust and heightened dependence are also mutually reinforcing mechanisms. As usage deepens, it can become seemingly logical for humans to do nothing, creating an automatic reflex of cognitive offloading (Burns et al., 2026). Schools are fighting against a technology that is structurally designed to replace the very thinking skills schools are meant to build. Long-term sustainable change requires reform from policymakers and developers to ensure that the AI tools available to students are pedagogically sound and can’t negatively impact cognitive development the same way these general models do.
That being said, the risks are far too high to simply wait around for policies and tools to change. Action must be taken immediately. Kabashkin identifies five key risk factors that contribute to cognitive atrophy: excessive delegation of tasks (which manifests in automatic use of AI for problem solving), surface-level understanding (which manifests as acceptance of outputs without further analysis), cognitive laziness (which manifests in the avoidance of mental effort), automation bias (which manifests as an overtrust in AI), and creativity erosion (which manifests in the repetition of the same AI-generated patterns). To counteract these risks, he proposes what he calls Applied Cognitive Management (ACM)—strategies for practicing what we can think of as “good cognitive hygiene” rituals. These strategies help counteract the risk factors that could lead us toward cognitive atrophy (or in the case of students, preventing them from developing critical thinking in the first place). The risk factors, their manifestations, and the proposed strategies identified by Kabashkin are outlined in the figure below:
Note. From “Cognitive Atrophy Paradox of AI–Human Interaction: From Cognitive Growth and Atrophy to Balance,” by I. Kabashkin, 2025, Information, 16(11), 1009, p. 30 (https://doi.org/10.3390/info16111009) CC BY 4.0.
While Kabashkin was not specifically discussing the school context, his countermeasures can be applied when working with students who are still developing their skills. As systemic changes and AI tools continue to develop, modelling and instilling these habits is a promising practice that could help protect student thinking in the interim.
LLC Leadership and Leveraging the Digital Media Toolkit
In Canada, a lack of policy and professional development leaves educators unequipped to instruct students on responsible AI use. In a study conducted by the Fraser Institute, an astonishing 63.6% of Canadian teachers reported no training at all on how to instruct students to use AI reasonably. (Fraser Institute, 2026). This is an alarming statistic given that unguided AI use hinders critical thinking. This lack of guidance leaves educators ill-prepared to support students with AI. How can we possibly expect teachers to protect student learning, when they have not received adequate guidance on how to do it?
This is where the teacher-librarian can help. While we wait for changes further up the chain, schools need knowledgeable leaders to facilitate AI literacy instruction and drive forward teaching practices that prioritize human thinking. Teacher-librarians can be that leader. Part of the teacher-librarian’s role is to instill critical thinking and foster literacies to empower life-long learners, (Canadian School Libraries, 2016-2023) which now includes AI literacy as part of digital and media literacy. Taking the lead when it comes to AI literacy is inherently part of the modern teacher-librarian’s role. Additionally, teacher-librarians are in a perfect position to support the development of better AI practices and AI literacy instruction through collaborative projects with classroom teachers. This can help support staff development. Furthermore, the leadership from the LLC would help ensure professional AI mentorship is available to all students regardless of home resources or their classroom teacher’s current level of comfort. Therefore, teacher-librarians are uniquely positioned to bring equity, collaboration, and expertise together. By taking the lead on AI literacy, they not only ensure every student gets the support they need to think critically and practice good cognitive hygiene, but they also empower classroom teachers with the confidence and guidance required to navigate these new digital tools.
The Digital Media and Literacy Toolkit and Good Cognitive Hygiene
In late 2025, Canadian School Libraries (CSL) published the Digital Media and Literacy Toolkit as a guide for school library professionals to support their school communities with an expanded notion of literacy, including AI literacy. It uses a framework of FEEL, THINK, and DO action strands to help learners ethically engage with, evaluate, and create content in an increasingly digital world. The DMT provides key questions to consider, suggested activities, and some possible resources to support their implementation. To combat cognitive atrophy, we can use the Digital Media and Literacy Toolkit as the framework for AI literacy, while also embedding and modeling good cognitive hygiene practices throughout. For the LLC purposes, I have adapted Igor Kabashkin’s ACM countermeasures to the following five “good cognitive hygiene rituals” :
1. Planned AI-Off Intervals: days/periods dedicated to manual tasks.
2. Reflection Checkpoints: engaging in metacognitive reflection.
3. Manual-First Rule: complete tasks manually, then compare AI generated results. 4. Dual Validation: students lateral read to validate AI outputs.
5. Human-Focused Creative Tasks: tasks designed to be more challenging for AI to complete effectively and that encourage human creativity.
Before going any further, it is worth noting that primarily the secondary context was being considered in the researching and writing of this paper. The cognitive and developmental differences between a 7 year old and a 16 year old are inherently different, and may require different pedagogical approaches, which have not been considered within the scope of this paper.
FEEL: Digital Agency is Digital Wellness. As part of the FEEL strand, one of the key questions we are asked to consider is How might wellness include our digital selves? (Canadian School Libraries, 2025, p.8-9). This is the perfect opportunity to incorporate explicit discussions on the risks of AI use on cognition and development, and introduce some good cognitive hygiene practices as important wellness strategies. This could be an excellent time to begin modelling, AI-off Intervals, reflection checkpoints and/or the manual-first rule as an important part of digital wellness. For example, teacher-librarians can “Help students create personalized plans (using AI or a digital tool of their choosing) that fit their lifestyle and needs around digital wellness” (Canadian School Libraries, 2025, p.9). Before starting the plan, students can be instructed to maintain a week-long AI Diary (a form of reflection checkpoint) to track when they use chatbots to replace effort versus when they use them to actually enhance learning. In a later lesson, the teacher-librarian can facilitate a discussion on cognitive atrophy—explaining that the brain needs the productive struggle of thinking to remain healthy. To model this, the teacher-librarian mandates AI-off intervals during the initial reflection and planning of their digital wellness plan, requiring students to rely on their own thoughts to create it. The plans they created can later be compared to an AI created plan, or AI could be left out of the task entirely. By incorporating reflection checkpoints, students can begin to assess their own AI habits, and determine if they are healthy or unhealthy. Rituals like ai-off intervals or the manual-first rule then protect the mental effort necessary for intellectual development.
THINK: Critical Analysis, Manual Reasoning, and Dual Validation. The THINK strand is specifically focused on developing student thinking skills, so this is perhaps the best area to model and implement good cognitive hygiene rituals when using generative AI. The activities and concepts presented in the THINK section of the framework provide an excellent opportunity to embed the manual-first rule, reflection checkpoints, and dual-validation. For example, if the teacher-librarian plans to “do guided exercises with texts in different mediums to focus on skill building in areas like identifying the main idea, biases, perspectives and how different formats can change how we interpret information” (Canadian School Libraries, 2025, p.18), they can apply the manual-first rule by requiring students to complete a manual analysis of a text to identify main ideas and biases before using AI to generate a report. Students can then compare the AI outputs to their own ideas, critique the quality of the outputs, and look for any bias or inaccuracies in the AI output itself. This ensures initial cognitive activation and prevents students from accepting machine results without evaluation.
Another example from the THINK strand could be to “develop units that focus on specific aspects of information literacy, such as evaluating sources through lateral reading, triangulation, understanding bias, and synthesizing information” (Canadian School Libraries, 2025, p. 23). Students can practice dual validation by using trusted LLC databases to verify claims generated by an AI tool. Students then complete a reflection checkpoint to explicitly compare their personal understanding with the AI output, noting where AI may have hallucinated. This is probably a familiar practice already, which you may know as lateral reading. The importance of verifying information and finding multiple sources is not a new concept, however the increasing overtrust in AI is a new and growing problem. We must stress the importance that the human needs to verify information elsewhere. The idea of reflection checkpoints will also sound familiar. Encouraging metacognition in students has long been a part of teaching practice. In a way, we are just implementing these familiar practices in a new context.
DO: Human-Led Creation. Lastly, we can leverage the DO strand to combat the erosion of creativity. When we explore the question: How might we support students in creating original content or remixing existing content? (Canadian School Libraries, 2025, p.29-30), we can design tasks that encourage or require real human creativity, and that are challenging for AI to complete. For example, if we “facilitate design thinking challenges centered around the UN SDG goals, where students work collaboratively to come up with creative solutions to real-world issues” (Canadian School Libraries, 2025, p.30). The teacher-librarian can facilitate a localized problem-solving challenge where students identify and solve an issue in their immediate school or neighborhood, such as improving the school cafeteria’s food waste system. This task is challenging for AI to complete effectively because it centers on lived experiences and local stakeholder needs that AI cannot easily observe or feel. Students begin with a manual-first phase—where they rely on peer interviews, walkthroughs, and personal reasoning to formulate their plan. AI would only be permitted in the final stages for presentation or data organization. Students would use a justification log (a form of reflection checkpoint) to answer reflection questions like Why was AI needed here? and How did I change the AI output to ensure it still reflects my original human ideas? By requiring students to focus on a task that is not as straightforward for AI to complete, to defend their original creative intent when employing AI, and requiring a manual-first approach, the teacher-libarian can help keep the students in charge and prevent them from relinquishing control of the creative thinking process to AI.
Conclusion
While protecting student thinking and development will require efforts from many different levels, the LLC plays an indispensable role in this task at the school level. By grounding AI literacy in the framework of the Digital Media and Literacy Toolkit and embedding intentional cognitive hygiene practices, teacher-librarians can keep the library a hub for human-driven inquiry. Navigating the future of education requires us to ensure technology serves as a tool to support, rather than undermine, the intellectual growth of our students.
Some questions for further consideration:
● How else might we design tasks that prioritize human thinking over machine output, and that make students want to engage with the productive struggle?
● How might we support classroom teachers (and all other staff members) in developing their competencies? What kind of PD could be led at the school level to support them?
● How can the LLC bridge the communication gap between home and school regarding healthy AI use and digital wellness?
References
Bastani, H., Bastani, O., Sungu, A., & Kabakcı, Ö. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences (PNAS), 122(26), Article e2422633122. PubMed Central. https://doi.org/10.1073/pnas.2422633122
Canadian School Libraries (CSL). (2020). . Leading learning: Standards of practice for school library learning commons in Canada. http://llsop.canadianschoollibraries.ca
Canadian School Libraries (CSL), Casa-Todd, J., Anderson, N., Fowler, C., Jeffrey, J., Krys, T., Loffredi, L., & Maliszewski, D. (2025). Digital and Media Literacy Toolkit. Canadian School Libraries. https://www.canadianschoollibraries.ca/digital-and-media-literacy-toolkit/
Burns, M., Winthrop, R., Luther, N., Venetis, E., & Karim, R. (2026). A new direction for students in an AI world: Prosper, prepare, protect. Brookings Institution. https://www.brookings.edu/articles/a-new-direction-for-students-in-an-ai-world-prosper-prepare-protect/
Fraser Institute & MacPherson, P. (2026). Survey of Canadian Grades 6–12 Teachers on AI Training and School Policies for Its Use. Fraser Institute. http://www.fraserinstitute.org
Gerlich, M. (2025). AI Tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6. https://doi.org/10.3390/soc15010006
Irene, C. (2026). Artificial intelligence in classrooms: Cognitive dimensions. European Parliamentary Research Service. https://www.europarl.europa.eu/thinktank/en/document/IUST_BRI(2026)784575
Kabashkin, I. (2025). Cognitive Atrophy Paradox of AI–Human Interaction: From Cognitive Growth and Atrophy to Balance. Information, 16(11), 1009. https://doi.org/10.3390/info16111009
KPMG. (2025). Generative AI boom among Canadian students raise dilemmas. https://kpmg.com/ca/en/media/2025/10/generative-ai-boom-among-canadian-students-raises-dil emmas.html
OECD. (2026). OECD Digital education outlook 2026: Exploring effective uses of generative AI in education. OECD Publishing, Paris. https://doi.org/10.1787/062a7394-en.
Topali, P., Ortega-Arranz, A., & Molenaar, I. (2026). Transitioning from general-purpose to educational-oriented generative AI: Maintaining teacher autonomy. In OECD, OECD digital education outlook 2026: Exploring effective uses of generative AI in education (pp. 147–166). OECD Publishing. https://doi.org/10.1787/062a7394-en
Christina McKendrick is an educator stepping into a hybrid English as a Second Language (ESL) and teacher-librarian role at Greenwood Secondary School in the Toronto District School Board. She has previously served in both ESL and library positions within her school’s all-newcomer, ESL-focused community. A graduate of Queen’s University who began teaching in 2017 and is currently a Teacher Librarian Specialist candidate, she is passionate about diversifying library resources to reflect many different identities and thoughtfully integrating technology to enhance student learning. She strives to cultivate an inclusive school library learning commons rooted in student voice, equity, and support for the wider school community.