The Flat-Brain Society: When AI Answers Before We Learn to Think

Image generated with AI under licence.

Artificial intelligence allows us to do more, faster and, in many cases, with better results.

It can help us research, write, analyse information, code, create images, prepare presentations and solve problems. For an experienced professional with sound judgement and a good understanding of the context, AI can be an extraordinary productivity tool.

But there is a risk that still receives relatively little attention:

What happens when AI stops complementing our thinking and starts replacing it?

What happens when we get the answer before we've even learned how to ask the question?

We risk becoming a flat-brain society.

The Invisible Impact of AI

Much of the public debate around artificial intelligence focuses on its most visible consequences:

• The potential disappearance of jobs.

• Algorithmic bias.

• Privacy and intellectual property.

• The concentration of power in a handful of companies.

• Energy and resource consumption.

• The risk of losing control over increasingly autonomous systems.

All these issues matter. But there is another, far less visible impact: the gradual outsourcing of our cognitive abilities.

Thinking takes effort. So does learning.

As human beings, we need to confront uncertainty, explore alternatives, make mistakes, identify contradictions, connect ideas and question our own conclusions. These aren't signs of human inefficiency. They're an essential part of learning.

AI, however, can remove much of that friction. And while this is one of its greatest strengths, it's also one of its biggest risks.

If we always take the easiest route, we may end up achieving "better results" in the short term while gradually losing the ability to produce those results ourselves — or to improve on them by working alongside AI in the future.

We could call this "cognitive debt."

We gain speed today at the cost of becoming more dependent tomorrow.

Being Able to Produce an Answer Doesn't Mean We've Learned Anything

It's essential to distinguish between performance and learning. At the same time, we need to recognise that we live in a society that clearly rewards performance.

Someone can deliver an excellent piece of work with AI assistance without truly understanding the subject, remembering the arguments or being able to defend the conclusions.

The result may be correct. But the underlying skills may never have developed.

The results of an experiment conducted in Turkey involving almost a thousand mathematics students using GPT-4-based tools were recently published in the Proceedings of the National Academy of Sciences. Students with access to a ChatGPT-like interface performed better on exercises while they were allowed to use it. However, when they were tested without the tool, they performed worse than students who had never had access to it.

Similarly, the University of Hong kong published the findings of another comparable study, this time conducted in China with 26,000 students. Students who used AI to complete their homework reduced the time spent on it by 30% and, on average, achieved grades 18% higher than those who didn't use it. However, when exam time came, these AI-assisted students saw their average grades fall by 20%.

The Turkish study, however, went a step further. Researchers designed an experiment in which students interacted with an AI solution that acted as a tutor. Instead of giving them the answers directly, it provided hints and guided them through the reasoning process.

In this case, students continued to perform better on exercises, but they also retained what they had learned: when tested without AI assistance, their grades did not deteriorate.

This suggests that AI doesn't necessarily harm learning. The reality is more complex:

The impact depends directly on whether AI replaces cognitive effort or helps guide it.

Using AI to obtain an answer is not the same as using it to understand how that answer is reached.

Other studies in professional environments point in the same direction. Microsoft conducted research involving 319 professionals and concluded that greater confidence in AI was associated with lower self-reported levels of critical thinking. At the same time, AI shifted cognitive effort towards other activities, such as verifying information, integrating responses and supervising the final output.

So we're not necessarily witnessing the disappearance of thinking. We're seeing it redistributed. The relevant question is:

Are we abandoning skills that we will later need to supervise what machines produce?

The Flat Brain

The real danger isn't forgetting a fact that we can look up in seconds.

Humanity has been outsourcing memory and calculation for centuries. We've done it through writing, books, calculators, search engines and navigation systems.

The problem arises when we stop building the knowledge and mental models that allow us to interpret, question and validate answers.

To recognise that a conclusion is absurd, we first need to understand the problem.

Many years ago, I had a boss who used to tell me: "The problem isn't that what you're presenting is wrong. It's that you don't understand why it's wrong."

That lack of awareness is what makes the situation truly dangerous for us as human beings.

To challenge a recommendation, we need to have our own judgement.

To identify bias, we need to be capable of imagining a different perspective.

To reject an answer that sounds convincing, we need to retain the ability to question it.

A "flat brain" isn't a brain that stores less information. It's a brain that accepts answers without going through the process needed to evaluate them.

Ultimately, it's about losing our curiosity, critical thinking, ability to connect ideas and willingness to challenge established assumptions.

And that creates another risk: a world in which answers become increasingly similar.

Generative models produce outputs based on patterns found in enormous amounts of information. They can generate original alternatives, but they can also pull us towards the statistical centre of what has already been said, particularly when we ask generic questions and accept the first answer.

If millions of people use the same tools, ask similar questions and never challenge the results, we may achieve enormous collective efficiency at the expense of diversity of thought.

We've never produced so many answers. But perhaps we're asking fewer questions.

The Future of Society: Students

This risk is particularly significant for children and young people.

An experienced professional can use AI to speed up a task because they already have the knowledge, references and judgement needed to assess the result.

A student, on the other hand, is still developing precisely those abilities.

When students use AI to bypass the reasoning process, they're not simply outsourcing a task. They may also be outsourcing the learning that the task was supposed to provide.

From their perspective, the incentive is obvious: less effort, greater speed and probably a better assignment.

The problem emerges in the medium term, when they have to explain what they've submitted, apply their knowledge in a different context or deal with a situation where AI gets something wrong.

The ease of producing a result can create an illusion of competence.

This phenomenon isn't limited to education. It can occur in virtually any profession. An analyst who stops developing hypotheses, a consultant who automatically accepts a summary, a programmer who no longer understands the code being generated, or a manager who delegates decisions to an algorithm may all remain productive for a while.

Until a new, ambiguous or critical situation arises.

That's when we'll discover whether AI was enhancing their capabilities or replacing them.

We see this phenomenon every day in the training sessions we deliver, both at universities and business schools, where students increasingly resort to copy-and-paste without analysing the problem they need to solve, defining its scope or even assessing the answer.

We also encounter it in many companies that have embraced AI implementation driven purely by hype or the promise of rapid, supposedly greater cost efficiency, without establishing a cultural change plan or internal governance protocols.

What Do the AI Gurus Think?

This concept of "cognitive debt" took on a new dimension a few days ago when Nvidia CEO Jensen Huang was interviewed by Ezra Klein, a well-known journalist associated with The Washington Post and Newsweek.

Klein raised the findings of the University of Hong Kong study with Huang.

Huang acknowledged that some basic mathematical skills are being lost and questioned whether that necessarily matters. His argument was that humanity has always abandoned certain skills when adopting new technologies, and that young people will develop other capabilities, such as systems thinking and collaboration with AI systems.

That's where the controversy began. Some interpreted his comments as an acceptance of cognitive decline among children. That wasn't exactly what he said, but there was certainly a tendency to downplay an underlying problem that is genuinely critical.

Why? It's difficult to know. But coming from someone with a clear interest in the rapid development of AI, his comments inevitably raise questions.

The fundamental question, however much some AI gurus may downplay it, remains extraordinarily important:

Which capabilities can we outsource without losing our ability to understand and control the tools performing them?

Not every skill made unnecessary by technology loses its cognitive value.

Learning mathematics isn't just about performing calculations. It also develops abstraction, logic, sequential thinking and problem-solving skills.

Writing isn't simply about producing text: it forces us to organise our thoughts.

Memorisation isn't just about storing information: it provides the foundation of knowledge upon which we build new connections.

Some operations may eventually become unnecessary in everyday life. But before abandoning a skill, we should understand what other abilities we were developing by learning it.

Technological efficiency alone shouldn't determine which skills are worth preserving.

Applying the Three-Bucket Method to Learning

When we approach an AI-related change management process, we use the three-bucket metoth to help us reflect on how we coexist with this technology.

Bucket 1 – What AI Can Do Autonomously

This includes activities that technology can perform as well as, or better than, we can.

In an educational context, handing these tasks directly to AI may improve the immediate result, but it can also eliminate an opportunity to learn.

Not everything that can be automated should be automated during the learning process.

Bucket 2 – What We Do Better by Collaborating with AI

This is where the greatest potential lies.

AI can explain a concept in different ways, adapt exercises to a student's level, ask questions, offer counterarguments, identify mistakes and help explore new perspectives.

Instead of simply providing the solution, it can act as a tutor, a discussion partner or an intellectual challenger.

Here, AI doesn't eliminate effort. It guides that effort and makes it more productive.

Bucket 3 – What Fundamentally Depends on Human Beings

This bucket includes purpose, judgement, responsibility, contextual understanding, values, curiosity and the final decision.

But we must avoid one mistake: assuming these abilities will remain intact simply because we currently consider them uniquely human.

The third bucket isn't naturally protected. It's a muscle.

If we stop exercising curiosity, reasoning, scepticism and decision-making, we will also weaken the very qualities that make us valuable alongside technology.

Paradoxically, while we believe we're strengthening the first bucket — letting AI do everything — we may actually be undermining our ability to develop the third.

Banning AI Isn't the Same as Educating

Institutional responses vary considerably.

Some education systems have chosen to ignore the phenomenon. Others have tried to block certain tools.

Intermediate approaches are also emerging: France restricts independent use of generative AI according to age and places it within an educational framework; Australia has a national framework for responsible AI use; and UNESCO recommends an age-appropriate, human-centred approach supported by data protection and training.

Restrictions may be necessary in certain situations and environments. Just as we don't allow students to use calculators when the goal is to learn basic arithmetic, limiting AI can make sense when the objective is to develop a specific skill.

But blocking a tool doesn't teach anyone how to use it.

Moreover, a policy based exclusively on prohibition can increase inequality.

Young people whose families, schools or mentors can teach them to use AI critically will develop new advantages.

Everyone else will still gain access to AI, but they'll probably end up using it as a free answer-generating machine.

The divide won't simply be between those who have access to AI and those who don't.

It will be between those who learn to think with AI and those who learn to let AI think for them.

Digital Literacy vs. Cognitive Literacy

The key is that we don't need to choose between adopting AI and rejecting it.

We need to learn how to use it without surrendering our autonomy.

That means introducing a few simple rules:

• Make an initial effort before asking AI for the answer.

• Develop your own hypothesis.

• Ask for hints, questions and counterarguments rather than complete solutions.

• Check claims against independent sources.

• Explain the result in your own words.

• Identify which parts of the work were carried out by AI.

• Always ensure that a person remains responsible for the final decision.

• Regularly practise certain skills without technological assistance.

Additionally, in professional environments, we should stop measuring AI success solely in terms of hours saved.

We should be asking ourselves:

• Does the professional understand the result?

• Can they recognise when it's wrong?

• Can they defend the decision?

• Could they solve a variation of the same problem?

• Are they using the time saved to strengthen their contribution in the third bucket?

Immediate productivity cannot be our only measure of progress.

What Should We Be Asking Ourselves?

Artificial intelligence has the potential to become one of the most powerful tools ever created for expanding human knowledge.

But it could also produce a generation of people who are extraordinarily efficient at obtaining answers and surprisingly dependent on technology to interpret them.

The difference won't lie solely in the technology. It will lie in how we choose to use it, teach it and govern it.

The question isn't whether our children, students or professionals should use AI. They will!

The real question is something else entirely…

Are we teaching them to use AI to think better, or simply to think less?

Legislation can establish boundaries. Institutions can define frameworks. Companies can implement controls.

But no regulation can replace the need to develop sound judgement, responsibility and critical thinking.

Because the greatest risk isn't that artificial intelligence will eventually learn to think like a human being. It's that human beings will gradually give up thinking for themselves.