Using AI and not becoming an idiot - Xavier Denoel - Conscious Change & Human Growth

Using AI and Not Becoming an Idiot — Xavier Denoël

Using AI without losing the ability to think — Artificial intelligence is often discussed through extremes: either as an existential threat or as a miraculous productivity booster. Much less attention is given to a quieter, more structural question: what does systematic reliance on AI do to our capacity to learn?

From Tool to Cognitive Shortcut

Most generative AI tools are designed to reduce effort. They draft, summarise, code, rephrase and decide faster than humans. From a performance perspective, this is undeniably attractive. From a learning perspective, it is ambiguous.

Many users progressively delegate writing, reasoning, searching, structuring ideas — and even decision-making. What starts as assistance becomes substitution. The risk is not dependence on technology as such — humans have always externalised effort — but systematic avoidance of the mental operations that enable learning.

What Happens in the Brain When We Learn

Learning is not the accumulation of information. It is a biological process, involving distinct but interdependent brain systems. Research in neuroscience broadly identifies three essential phases:

Encoding — Theory & Attention

New information is processed through the prefrontal cortex and encoded by the hippocampus. This phase is cognitively demanding — but necessary. It is where understanding begins.

Practice — Retrieval & Application

Repeated effort to retrieve and apply knowledge strengthens neural connections. This is where learning becomes durable. The effort itself is the mechanism.

Metacognition — Error Processing & Adjustment

Feedback and correction generate prediction error signals associated with dopaminergic activity. These signals guide the brain in reinforcing effective strategies. Together, the three stages allow learning to move toward automatisation via the basal ganglia.

Short-circuiting any of these phases weakens the entire process.

AI and the Bypass of Practice

When an essay, a solution or a block of code is generated externally, retrieval effort disappears, error detection is reduced, and metacognitive adjustment is minimal. The result is output without internalisation.

This explains a recurring observation: individuals who produce sophisticated work with AI support may struggle to explain the concepts they use. The knowledge has not been integrated — it has been outsourced.

The "Google Effect", Amplified

This phenomenon is not new. Studies on digital memory have shown that when information is easily retrievable online, individuals are less likely to store it internally — a mechanism known as the Google Effect (Sparrow, Liu & Wegner, 2011). Generative AI intensifies this effect: it does not only retrieve information, it structures reasoning, resolves uncertainty, and removes cognitive friction.

From a neurocognitive perspective, this risks atrophying the very systems needed for complex thinking — because the hippocampus and prefrontal cortex are under-stimulated when AI pre-digests every cognitive challenge before it reaches you.

Empirical Evidence: Cognitive Debt

A 2025 experimental study examined brain activity during writing tasks under different conditions, using EEG measurements:

AI-assisted Lowest Neural activity — output produced, learning minimal

Internet-assisted Intermediate Some retrieval effort required

Unassisted Highest Full cognitive engagement — deepest encoding

The authors describe this as cognitive debt: short-term efficiency gained at the cost of long-term learning capacity. This aligns with established findings: learning requires effortful processing (Bjork & Bjork, 2011).

Why Experts Benefit and Novices Struggle

An important nuance: experienced professionals tend to use AI differently. Senior developers, for example, verify, debug, reinterpret, and integrate AI output into existing mental models. Their learning cycle was completed before AI entered the picture. As a result, AI extends their capabilities rather than replacing them.

Novices, by contrast, risk skipping foundational stages entirely — placed in a position of "orchestration" without having learned the instruments. This distinction is crucial for training, onboarding and education policies.

AI as Tutor, Not as Substitute

The most constructive response is not the warning, but the alternative. AI can support learning if it is deliberately constrained to promote active engagement rather than passive consumption.

1
Propose exercises — ask AI to challenge you, not to answer for you.
2
Guide reflection — use AI to surface questions, not to close them.
3
Delay answers — attempt the task yourself first, then compare with AI output.
4
Encourage error analysis — when AI is wrong, the correction teaches more than the answer.

Left unstructured, learners tend to exploit AI for immediate answers. Learning requires friction — and AI does not impose it naturally.

Choosing Effort Over Ease

For the first time in history, we have tools capable of removing almost all cognitive friction. Whether this becomes an opportunity or a liability depends entirely on how consciously we choose to use them.

"It is effort that shapes the brain, not results. AI will not make us less intelligent — but delegating thinking systematically might."

Xavier Denoël — AI, Learning & Human Cognition

The question is no longer whether AI is intelligent. It is whether we are willing to remain so.

Sources & References

Bjork, R. A., & Bjork, E. L. (2011). Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning
Kandel, E. R. et al. (2021). Principles of Neural Science
Sparrow, B., Liu, J., & Wegner, D. M. (2011). Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips. Science
Sweller, J. (2011). Cognitive Load Theory
Dweck, C. (2006). Mindset: The New Psychology of Success
Micode — La Fabrique à Idiots (video reference)