When thinking is no longer compulsory

Until recently, writing a text, comparing several options or resolving a question required going through a process: seeking information, organising ideas, ruling out possibilities and forming a conclusion.

Now we can ask an artificial intelligence to do much of that work for us. Within seconds, we get a clear, structured answer that is often more comprehensive than the one we would have produced on our own.

The problem is not necessarily the quality of the result. The answer may be useful, correct and help us move forward. The issue arises when we begin to confuse immediate access to a conclusion with having actually thought.

Artificial intelligence does not only speed up thinking. It can also allow us to avoid some of the tasks that build it: formulating questions, weighing up arguments, spotting contradictions or living with uncertainty for a while. Research by Microsoft Research into the use of generative AI among knowledge workers found that greater confidence in these tools could be associated with less critical-thinking effort during certain tasks.

What happens to our ability to understand when we need to take less and less part in constructing answers?

Getting an answer is not the same as understanding it

Having the right answer does not necessarily mean understanding why it is right. We can repeat a conclusion, present an analysis or defend a recommendation without clearly knowing which data supports it, which assumptions it contains, which alternatives were ruled out or in what circumstances it would cease to be valid.

Artificial intelligence makes this difference less visible because it can produce answers that are well structured, confident and convincing. Its fluency can lead us to interpret ease of reading as a sign of truth or depth. Research into processing fluency shows that we tend to regard statements our brains process more easily as more reliable.

That is why an answer can seem clear to us without our having truly understood the problem. We recognise the argument, follow its logic and may even agree with it, but we would not always be able to explain it in our own words, identify its limits or reconstruct the reasoning that leads to it.

Understanding does not mean recognising that an answer sounds good. It means being able to explain it, question it and know what would have to change for it to no longer be valid.

Thinking is not simply about reaching a conclusion

Thinking is not just about finding an answer. It also involves framing the problem well, connecting ideas, spotting contradictions, reviewing our own assumptions and accepting that, sometimes, we still do not know enough.

Much of this process does not produce immediate results. Doubting, changing our minds or remaining with a question for a while may seem unproductive, but that is precisely where judgement is built. We do not only discover what we think, but also why we think it and what might make us reconsider it.

Artificial intelligence tends to offer us an exit: an explanation, a summary or a proposal that appears to bring things to a close. This can be very useful, but it can also lead us to abandon too soon questions that still need time, context or discussion.

Human thought does not always move in a straight line. Sometimes it needs to pause, go back and recognise that an initial answer was insufficient.

The value of thinking lies not only in the conclusion reached, but also in the transformation that takes place as we try to reach it.

The risk of confusing speed with intelligence

We live in a culture where responding quickly is often interpreted as a sign of competence. The person who writes first, analyses more data or proposes an immediate solution seems to have a better grasp of the problem.

Artificial intelligence reinforces this expectation. It makes it possible to generate more texts, compare more options and produce more analyses in far less time. In many tasks, this acceleration is clearly positive: it removes repetitive work, makes information easier to access and helps us move forward.

But working faster does not necessarily mean judging better. Research into the speed–accuracy trade-off shows that making decisions more quickly usually means having less time to gather and assess information.

AI can reduce the time needed to produce an answer, but it does not always reduce the time needed to understand a problem. Some issues still require context, comparison and reflection that cannot be accelerated without losing important nuances.

The real opportunity lies not only in doing the same thing faster, but in deciding what to do with the time we have gained.

Are we using it to think better, or simply to produce more?

Intellectual comfort and the weakening of judgement

When a tool consistently offers a reasonable answer, it is easy to become accustomed to accepting the first suggestion, to stop comparing alternatives or to ask for a summary before we have read and thought things through for ourselves.

This delegation is not always negative. So-called cognitive offloading allows us to transfer certain tasks to external tools and reserve mental resources for other matters. The problem arises when we stop distinguishing between what is worth automating and what we need to keep practising.

If we turn to AI before forming a position of our own, we may begin to use its answers as an endpoint rather than a starting point. Little by little, we stop asking ourselves what is missing, what alternatives exist or which part of the reasoning does not quite convince us.

The risk is not only that we receive incorrect information. We can also lose the ability to recognise when an answer is superficial, insufficient or simply too convenient to be challenged.

Judgement is not weakened because a machine makes mistakes. It is weakened when we stop exercising our ability to assess what it proposes to us.

The power of plausible answers

One of artificial intelligence's most persuasive features is its ability to produce answers that seem reasonable. They are well written, maintain an internal logic and tend to adopt a confident tone. This formal quality can cause us to let our guard down.

A plausible answer can conceal a weak premise, oversimplify a complex issue or present an interpretation as though it were a fact. It can also reinforce our existing preferences, especially when it confirms what we were already expecting to find.

This is not an entirely new phenomenon. Research into processing fluency shows that we tend to regard statements that are easy to process as more reliable. When something is quickly understood, it seems more familiar, coherent and, at times, truer.

With AI, this effect takes on a new dimension. An answer does not need to be openly false to mislead us. It need only omit important nuances, close an open question too soon or present a debatable conclusion with a confidence it does not warrant.

The problem, therefore, lies not only in the errors the system may make, but in our willingness to accept what fits, sounds good and allows us to move on without stopping to check it.

When an answer seems intelligent, we tend to demand less evidence that it truly is.

Using AI to think better, not to stop thinking

The answer is not to reject artificial intelligence, but to use it in a more demanding way. The question is not how much we can delegate, but what kind of involvement we want to retain within the process.

We can start by formulating an initial answer before consulting the tool, ask it for arguments against our position or request that it identify the assumptions and weak points in a conclusion. We can also compare interpretations, verify important claims and explain in our own words what we have obtained.

One of the most useful ways of working with AI is to ask it for questions, not only answers. A good question can reveal what we had not considered, broaden the problem or help us identify a premature simplification.

It is also worth deciding in advance which parts of the process we do not want to delegate. Defining the problem, assessing the consequences or making the final decision may require a human involvement that no automated answer should replace.

AI adds more value when it expands the thinking we are already doing than when it completely replaces our participation. Used in this way, it does not eliminate intellectual effort: it directs it, tests it and, at best, deepens it.

The process is also part of the outcome

In certain areas, the journey is not an obstacle we should eliminate. It is precisely what allows us to learn.

Writing, studying, researching, designing or solving a problem does not only lead to a visible outcome. During that process, we learn to organise ideas, recognise errors, connect concepts and make better-grounded decisions. Even the attempts we discard help build a more solid understanding.

Artificial intelligence can shorten many of these tasks and free up valuable time. However, if we always delegate the whole journey, we may achieve better immediate results while at the same time developing less ability to tackle the next problem without help.

This is especially important in activities where the aim is not only to deliver something, but to learn how to do it: forming an opinion of our own, understanding a subject, writing in a recognisable voice or making a difficult decision.

There are processes that do not merely produce an answer. They also produce the person capable of understanding it.

That is why not every effort should be interpreted as inefficiency. Sometimes, what seems like a delay is precisely what transforms information into knowledge and experience into judgement.

Conclusion

Artificial intelligence can help us search for, organise, compare, expand and reframe ideas. It can show us paths we had not considered, speed up tasks that once took a great deal of time and offer us an initial structure from which to begin working.

We do not need to defend an artificial separation between human thought and these tools. We will increasingly think with them, just as we already think with books, search engines, maps, calculators or conversations with other people.

The question is not whether AI will take part in our processes, but what role we want to retain within them. We can use it as support to explore a question more thoroughly, or allow it to progressively replace our involvement in the search for an answer.

That difference will not always be visible in the final result. Two texts may seem equally correct, even though one is the product of genuine reflection and the other has been accepted with barely any scrutiny. But the difference will become clear when we need to justify a decision, respond to an objection, spot an error or face a new problem.

I believe the challenge is not to prevent artificial intelligence from thinking with us, but to stop it from always thinking in our place. Because an answer may save us time, but only the effort of understanding can turn it into judgement.