The Cognitive Cost of Convenience: Learning & the Use of GenAI

Learning is not produced by the final answer, but by the cognitive effort required to reach it. What happens when we increasingly delegate thinking, writing and problem-solving to artificial intelligence? An opinion paper by Marine Hadengue.

Opinion paper by Marine Hadengue, Ph.D., Associate Dean at SKEMA Business School and CEO of the Higher Education for Good Foundation (Youth Talks). May 2026.

Disclaimer: I do not hold a Ph.D. in neuroscience, cognitive science, or learning processes. This paper is therefore not intended to be read as a scientific article or as an expert review of the literature. It is based on scientific sources that I have tried to select and interpret carefully, with the aim of making a complex issue more accessible. Its sole purpose is to help open a broader conversation about what may happen when we increasingly delegate thinking, writing, problem-solving, and learning processes to artificial intelligence and what this could mean for human cognition, education, and development.

1. How the brain learns: the cognitive and neural mechanisms

The core idea is that learning is not produced by the final answer, but by the cognitive effort required to reach it. The brain changes through active engagement: searching, struggling, retrieving, testing, making mistakes, correcting them, and reflecting on the process.

Learning relies on several interacting systems:

1. Encoding and memory formation: When we learn new information, the brain must first encode it. The hippocampus, a key structure in the medial temporal lobe, plays a central role in forming new declarative memories and linking new information to existing knowledge. This is why effortful memorization, navigation, explanation, and retrieval can physically shape neural circuits. The famous study on London taxi drivers showed structural differences in the hippocampi of drivers who had intensively learned London’s street network, illustrating experience-dependent neuroplasticity.
2. Practice and automatization: With repeated practice, some skills move from effortful, conscious processing toward more automatic execution. This involves cortico-striatal circuits, including the basal ganglia, which are important for procedural learning, habit formation, and automatization. In simple terms: theory gives you a mental model, but repeated practice turns it into usable competence.
3. Retrieval practice and consolidation: Learning is strengthened when the learner actively retrieves information rather than simply re-reading or receiving an answer. This is known as retrieval practice or the testing effect. Research shows that retrieval practice improves long-term retention because it forces the brain to reconstruct knowledge, strengthen neural pathways, and identify gaps.
4. Metacognition: The brain also learns by monitoring its own understanding. This is metacognition: the capacity to ask, “Do I really understand this?”, “Where am I uncertain?”, “What strategy should I use next?” Reflection, self-explanation, feedback, and error correction help consolidate learning and build transferable expertise.
5. Neuroplasticity through effort: The broader mechanism is experience-dependent neuroplasticity. The brain adapts to what it repeatedly does. If we repeatedly reason, search, write, calculate, explain, or solve problems ourselves, the relevant neural networks become stronger. If we stop using those circuits, we risk cognitive weakening through underuse. This is the “use it or lose it” logic.

A concise formulation would be:

Learning = encoding + effortful retrieval + repeated practice + feedback + metacognitive reflection.

2. The problem AI creates for these cognitive learning mechanisms

The issue is not AI itself, but cognitive outsourcing: using AI to bypass the very mental operations that create learning.

When AI gives the answer too early, it can short-circuit several essential processes:

1. It reduces desirable cognitive effort: If the learner asks AI for the answer before struggling with the problem, the brain receives the product without doing the work. But the work — searching, comparing, failing, revising — is what builds neural connections. AI may therefore produce performance without learning.
2. It weakens encoding: When information is generated externally, the learner may not encode it deeply. The answer looks clear, but the learner has not built the internal representation needed to retrieve or apply it later.
3. It disrupts retrieval practice: Instead of retrieving knowledge from memory, the learner retrieves it from the machine. This can reduce the strengthening of memory traces. The learner becomes fluent in prompting, but not necessarily in knowing.
4. It undermines procedural learning: For skills such as writing, coding, problem-solving, analysis, or argumentation, competence emerges from repeated execution. If AI performs the execution, the learner may skip the repetition required for automatization.
5. It reduces metacognitive monitoring: AI can create an illusion of understanding. Because the output is polished, coherent, and plausible, the learner may feel they understand the topic without being able to reproduce, critique, or transfer the reasoning independently.
6. It creates cognitive debt: Recent research has described this as cognitive debt: the short-term gain in efficiency may come at the cost of long-term cognitive development. A recent MIT Media Lab preprint by Kosmyna et al. (2025), Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task, explored the neural, linguistic, and behavioral effects of LLM-assisted essay writing. NB: The study should be interpreted cautiously, as it is a preprint and has not yet undergone peer review.

3. The key distinction: AI as a tutor vs AI as a substitute

The healthiest framing is not “AI or no AI.” It could be:

Bad use: AI does the thinking, writing, solving, summarizing, or structuring for the learner.
Good use: AI makes the learner think harder, explain more clearly, test themselves, receive feedback, and identify gaps.

AI is cognitively dangerous when it becomes a prosthesis for thinking. AI is pedagogically valuable when it becomes a sparring partner for learning.

For education, the practical principle should be:

Use AI after the learner has first attempted the task, not before. Ask AI to question, challenge, quiz, critique, or provide feedback — not to produce the final answer.

The solution may not be comfortable, but it is increasingly necessary… and it is radical.

If we are serious about preserving the cognitive processes through which students actually learn, we must reintroduce assessment formats that require them to think without technological mediation.

• This means a return to paper-based, in-person examinations, where students must retrieve, structure, and articulate knowledge on their own.
• It also means placing much greater emphasis on oral examinations, viva-style defenses, and live discussions, where students must explain their reasoning, respond to questions, justify their choices, and demonstrate genuine understanding in real time.

4. When AI finishes the job social media started

In an age where written production can be outsourced almost instantly, education must urgently reconsider what it is truly assessing. If essays, summaries, reflections, and even sophisticated arguments can now be generated by AI in seconds, then part of the evaluative focus must shift back to what cannot be easily delegated: embodied presence, intellectual ownership, spontaneous reasoning, oral articulation, and the capacity to think aloud under human scrutiny.

This is not a minor adjustment. It goes to the heart of what education is meant to cultivate: independent minds. For several years, young people have already been exposed to digital environments that affect attention, emotional regulation, sleep, identity formation, and mental health. Social media platforms, infinite scrolling, short-form videos, and algorithmically optimized content have contributed to cognitive overstimulation, attentional fragmentation, social comparison, and dopamine-driven reward seeking.

Generative AI may add a new and deeper layer to this problem. The concern is no longer only distraction or overstimulation, but the possible bypassing of the very cognitive operations through which learning occurs: sustained attention, effortful retrieval, working memory engagement, reasoning under uncertainty, error correction, and metacognitive monitoring. If students repeatedly outsource these processes before they have internalized them, they may become skilled at producing outputs without developing the underlying capacity to think independently.

The risk is not simply that AI will make students “lazy.” The deeper risk is cognitive dependency: a weakening of intellectual autonomy, executive function, epistemic agency, and the ability to know, question, reason, and judge without machine assistance. The language of “brain atrophy” should be used carefully, but the principle remains clear: neural systems develop through use. What we repeatedly practice, we strengthen. What we systematically delegate, we may fail to develop.

This is why education must protect spaces where thinking remains effortful, embodied, and personally accountable. Paper-based examinations, oral defenses, viva-style assessments, live debates, and in-person problem-solving exercises may no longer be old-fashioned; they may become essential. They require students to retrieve knowledge, structure ideas, tolerate uncertainty, respond to challenge, and demonstrate genuine understanding in real time.

The point is not to reject AI. The point is to preserve the human capacities that make AI useful in the first place.

For information on learning processes

• National Academies of Sciences, Engineering, and Medicine. (2018). How People Learn II: Learners, Contexts, and Cultures. National Academies Press.
• Roediger, H. L., III, & Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science, 17(3), 249–255.
• Karpicke, J. D., & Roediger, H. L., III. (2007). Repeated retrieval during learning is the key to long-term retention. Journal of Memory and Language, 57(2), 151–162.
• Walker, M. P., & Stickgold, R. (2004). Sleep-dependent learning and memory consolidation. Neuron, 44(1), 121–133.
• Wenger, E., & Lövdén, M. (2016). The learning hippocampus: Education and experience-dependent plasticity. Mind, Brain, and Education, 10(3), 171–183.
• Kleim, J. A., & Jones, T. A. (2008). Principles of experience-dependent neural plasticity: Implications for rehabilitation after brain damage. Journal of Speech, Language, and Hearing Research, 51(1), S225–S239.
• Metcalfe, J. (2009). Metacognitive judgments and control of study. Current Directions in Psychological Science, 18(3), 159–163.
• Bjork, R. A., & Bjork, E. L. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In M. A. Gernsbacher, R. W. Pew, L. M. Hough, & J. R. Pomerantz (Eds.), Psychology and the Real World: Essays Illustrating Fundamental Contributions to Society (pp. 56–64). Worth Publishers.

This opinion paper was first published on LinkedIn: The Cognitive Cost of Convenience: Learning & the Use of GenAI.