By Paul W. Bennett, April 2, 2026
The proliferation of generative AI and digital tools is rapidly reshaping the terms of engagement in Canada’s K–12, college, and university classrooms. What began as a promising means of boosting productivity and personalizing learning is now under closer scrutiny and attracting intense debate in education circles. Generative AI opens up new possibilities for improving productivity, expanding individualized learning tailored to students and tapping into the enormous potential for 24/7 student support services. It has also accelerated the trend toward cognitive offloading – outsourcing thinking and writing tasks to machines. That’s what is driving rising concerns in the education sector about the impact AI is having on the thinking and writing capabilities of the current “screen-based generation.”
Coming to terms with the increasingly AI-driven world of education is emerging as the critical policy issue of our time. Finding a middle ground is a formidable challenge, especially with the emergence of the Alpha School model and the arrival of “Plato” (the AI humanoid), both of which utilize AI to replace human, in-person teachers. The best way forward might be found in building upon what is known as “cognitive complementarity” – or striking the right balance between human knowledge-building and tapping into expanded generative AI capacities. While taking advantage of AI tools, it is absolutely essential to ensure that students retain the capacity to use their heads instead of succumbing to the dictates of algorithms.
At the centre of this emerging debate is what American cognitive scientist Barbara Oakley has termed the “memory paradox.” The more we rely on external tools – whether calculators, search engines, or AI chatbots – the less we exercise and develop our own cognitive capacities. In the short-term, such tools appear to make learning easier and more efficient. Over time, however, they risk eroding the very mental habits and knowledge structures upon which higher-order thinking depends.
Canadian K–12 schools, already heavily invested in digital technologies over the past fifteen years, now find themselves at a crossroads. A recent bestseller, The Digital Delusion, by American cognitive neuroscientist Jared Cooney Horvath, drew a connection between falling scores on US national assessments since 2012 and the shift to public schools providing a device to every student. It’s helped to fuel a wave of American states to push for legislation to restrict screen use in schools. Getting ahead of the wave and establishing the needed guardrails has become a strategic imperative.
The invisible problem: When thinking becomes optional
Modern technological marvels have sped-up functions and have brought us unparalleled efficiency which can, at times, mask the mostly hidden risks, cognitive, mental and physical. Overreliance on search engines, navigation tools, and LLM aggregators is, for example, reducing our ability to retain information and perform mental tasks. That’s most visible when digital cash registers break down, Uber drivers get lost, or calculators go missing on a job site.
In classrooms, the same pattern is emerging. Students accustomed to instant answers and AI-generated text are less inclined to grapple with problems, sustain attention, or commit information to memory. What happens when those supports are removed? Increasingly, the answer is: far too many students get completely lost.
Cognitive offloading, once seen as a benign convenience, is exacting a higher price. It alters how students engage with knowledge, often substituting superficial familiarity for deep understanding. In doing so, it undermines the slow, effortful processes – practice, retrieval, and reflection – that build durable learning.
The erosion of writing skills
The arrival of ChatGPT in November 2022 marked a decisive turning point. Within months, it became evident that traditional writing assignments – long a cornerstone of classroom assessment – were newly vulnerable to automation. Essays could be generated in seconds, often with a fluency exceeding that of many students.
Early reactions ranged from alarm to resignation. In The Atlantic, teacher Daniel Herman declared the “end of high school English” and of assigning the conventional high school essay. Education analyst Natalie Wexler, author of The Knowledge Gap (2019), offered a more grounded perspective. Writing, she argued, is not simply a finished product but an iterative cognitive process. Through writing, students organize ideas, test understanding, and consolidate knowledge. AI may replicate the output, but it cannot reproduce the intellectual struggle that gives writing its educational value.
That distinction is critical. Many students already struggle to write effectively. Allowing AI to assume the task risks concealing weaknesses rather than addressing them. More troubling still, it may displace one of the most powerful learning mechanisms available – what cognitive scientists term “retrieval practice.”
Student assessment and its legitimacy
If generative AI destabilizes writing instruction, it poses an even greater threat to assessment. Unsupervised, take-home assignments – once central to progressive education – are now fundamentally compromised. When AI can generate essays, solve problems, and imitate personal voice, verifying authenticity becomes exceedingly difficult.
United Kingdom assessment expert Daisy Christodoulou has warned that such developments jeopardize the integrity of entire assessment systems. In Canada, the trend is already well advanced. A 2025 KPMG survey found that 73 per cent of Canadian students use generative AI for schoolwork, up from 52 per cent in 2023. Four out of five admit submitting AI-generated material as their own, while nearly half (48 per cent) report a decline in their critical thinking since adopting these tools.
Grades may be rising, but learning is not. AI detection tools remain unreliable, leaving teachers engaged in a performative, empty ritual – assessing work that may not be authentically human. The result is a quiet but consequential erosion of trust at the core of schooling.
A practical response may lie in revisiting older forms of assessment: supervised, in-person exams, whether handwritten or conducted under controlled digital conditions. Far from regressive, such measures may be necessary to ensure that grades reflect actual learning.
The memory paradox and its policy implications
Beyond issues of cheating and assessment lies a deeper concern: the nature of learning itself. Research synthesized by Barbara Oakley and trumpeted by her Canadian disciples points to a troubling dynamic – as AI capability rises, human cognitive engagement looks to be declining.
Memory, in this framework, is not passive storage but an active process involving encoding, retrieval, and integration. Repeated retrieval builds mental frameworks – “schemas” – that underpin reasoning, problem-solving, and creativity. Cognitive offloading disrupts this process. When students default to “just Google it” or “just ChatGPT it,” they bypass the very mechanisms that build understanding. Over time, this weakens knowledge structures, reduces mental flexibility, and diminishes independent thought.
Oakley’s research situates this trend within a broader pattern. The long rise in IQ scores from the 1930s to the 1980s – the Flynn effect – has plateaued and, in some countries, reversed. Declines observed in the United States, Britain, France, and Norway beg for some plausible explanation. While multiple factors shape cognitive performance, the recent decline, according to Oakley, is most likely the consequence of two converging forces: the educational shift away from direct instruction and memorization, and a growing reliance on cognitive offloading through digital tools. If generative AI use expands without any boundaries, the intellectual habits that sustain deep learning may be further eroded.
The pedagogical dilemma: Combatting cognitive atrophy
These developments also cast a critical light on prevailing pedagogical approaches. For decades, many Canadian classrooms have embraced constructivist or discovery-based models, emphasizing student autonomy and minimizing direct instruction.
In an AI-rich environment, such approaches may inadvertently exacerbate the problem. When students are given open-ended tasks with minimal guidance, they are more likely to turn to AI as a shortcut. In doing so, they may skip essential stages of learning: grappling with concepts, making errors, and refining understanding.
The result is a kind of cognitive atrophy. Students appear competent – producing polished work and achieving respectable grades – but lack the underlying knowledge and skills. It is, in effect, a form of “machine-assisted illusion.”
The case for cognitive complementarity
Generative AI is here to stay and presents us with not only a “wicked problem” but a fundamental policy dilemma. Sage observers of AI innovation like Samuel Hammond see three paths forward and possible responses:
- Openness to Technology Innovation and Cultural Evolution – the laissez faire approach promoted by the edtech industry and its secular evangelists.
- Mitigation and Adaptation – establish ethical standards and guardrails to educate AI users and limit the risks.
- Regulation and Enforcement – establish a regulatory system to constrain potentially bad actors and ensure ethical, socially-responsible use of AI tools.
Today’s educators are caught in the middle and being left on their own to come to terms with the fast-moving, relentless pace of AI innovation and incursion in K–12 schools, colleges, and universities. Generative AI has already demonstrated its value in education. The challenge is to achieve cognitive complementarity. AI should extend human capability, not replace it with “lazy” algorithmic processes.
Proponents of cognitive complementarity are attempting to develop that middle way, combining mitigation with adaptation. This requires a deliberate shift in how AI is integrated into classrooms. Students must first develop foundational knowledge and skills before being encouraged to use AI tools. They need to understand not only how to generate responses, but how to evaluate them – distinguishing fact from error, insight from fabrication.
Policy imperatives: Establish the boundaries and tap into the potential
Generative AI has proliferated and seeped into many facets of K–12 education. Our current laissez-faire approach to generative AI is fraught with potential problems. Without clear guidelines and guardrails, the risks to the quality of in-person teaching and credibility of student assessment will continue to multiply in our schools. School-based effective-use policies inevitably fall short without a broader, more robust set of policy guidelines. Given the magnitude and pace of change, it will require a coherent, coordinated, and effective response, ideally guided by the Council of Ministers of Education Canada and provincial education authorities. An effective response would focus on several key priorities:
- Guidelines for effective and ethical use: Teach students to rethink their current practices, to limit AI use and to delay cognitive offloading. Moral suasion is critical in turning the tide.
- Revamp student assessment practices: Place a renewed emphasis on supervised, in-person evaluations, as well as the development of tasks that require demonstration of knowledge in ways that AI cannot replicate.
- Integrate AI into teacher education and professional development: Address the realities of AI. Educators need a clear understanding of both the capabilities and limitations of these tools, as well as practical strategies for integrating them responsibly.
- Enhance curriculum design: Incorporate evidence-based learning strategies – retrieval practice, spaced repetition, and explicit instruction – that strengthen memory and understanding. These approaches, once unfashionable, are increasingly supported by cognitive science.
- Set guardrails for students: Clear guidelines should be established for student use of AI. This includes setting limits on when and how tools can be used, as well as teaching students to delay cognitive offloading until after they have engaged with material independently.
Teachers and students can be likened to experts and novices and, in that respect, are fundamentally different in the way they acquire, organize and utilize knowledge. With a good base of knowledge and pedagogical expertise, teachers are much better equipped to immediately tap into the real potential of Gen AI in their classrooms. Four examples of these real-life benefits for educators are:
- Improves productivity – Using AI to plan units of study and classroom lessons – reduces routine workload, freeing teachers to focus on actually delivering lessons and facilitating learning.
- Enables individualized learning – Creating individual tutorial programs and tailoring “mini-lessons” for students, much like the beta version piloted by Salman Khan at Khan Academy.
- Improves speed and quality of feedback – returning marking faster with more detailed responses to student work.
- Assists in revitalizing teaching practice – revamping proven lesson plans to keep them current and tap into the latest in edtech resources.
Each of these examples shows how teachers can, with boundaries in place, use AI-enabled tools to enhance student learning in the classroom and, at the same time, use AI tools to complement their own teaching practice.
The big lesson
The rise of generative AI has the potential to revolutionize education, for better or worse. Newly-created AI tools can greatly enhance productivity, but they also challenge the very basis of teaching and learning. If the purpose of education is merely to respond to queries and generate boilerplate answers, then machines may already be outperforming us. But if it is about developing the capacity to think, reason, and read nuances, then humans are indispensable. British education researcher Carl Hendrick put it best. Even the most advanced AI can simulate intelligence, but it cannot think for us. “That task remains,” in his words, “stubbornly and magnificently, human.”
The danger is not so much that AI humanoids like Plato, referred to as Figure 03, will replace teachers or even replicate students. It is that, without careful stewardship, AI will make thinking optional. In such a world, the erosion of memory, writing, and critical thought would not be immediately visible. Grades might rise. Work might appear more polished. But beneath the surface, the foundations of learning would be slowly and invisibly collapsing.
The emergence of AI has completely disrupted what remains of liberal education in our high schools, colleges, and universities. Early attempts to propose a “postplagiarism framework” and brush aside academic conventions as “Western industrial colonialist legacies” are indicative of the deepening crisis afflicting the academy. Reading between the lines, jettisoning academic integrity standards is tantamount to looking the other way.
Generative AI is here to stay but it’s up to us to ensure it serves our purposes. The policy choices made now – about AI’s role in teaching, assessment, and administration – will shape not only how students learn, but how they think and whether they can write on their own. It turns out, using your head still matters – and finding a way to co-exist with AI is now a strategic imperative.
Paul W. Bennett, EdD, is the director of Schoolhouse Institute, a senior fellow at the Macdonald-Laurier Institute and the founder of researchED Canada (2017–2026). Revised and updated from a Conference Presentation, researchED Atlantic Canada, Saint Mary’s University, October 25, 2025.
