AI Should Not Think For Us

This is a polished English translation of an Icelandic opinion piece originally published in Visir.

By Helgi S. Karlsson, Behavior Analyst, Psychologist, and Teacher

It is understandable that people worry artificial intelligence may make us lazier in our thinking. If a machine can write the text, summarize the article, solve the problem, organize the task, and even answer on our behalf, then a natural question appears: do we stop thinking for ourselves?

The answer is yes, if we use it that way.

But that is not inevitable.

AI can easily become a crutch we rely on too much. But it can also become one of the most powerful tools we have ever known for strengthening our thinking, not weakening it. The difference does not lie first in the technology itself, but in how we use it.

If we say to AI, "Write this for me," there is a risk that we hand the thinking over to it. If we do that often enough, especially without reading carefully, doubting, revising, and taking a clear position ourselves, then we are no longer training our own thinking. We are letting another voice fill in the gaps for us.

That can be comfortable. But comfort is rarely the word we associate with growth.

Thought is behavior. It strengthens with use and weakens with disuse. We become better at analyzing, comparing, connecting, doubting, and placing things in context by doing those things again and again. If we stop practicing those skills, they fade. That applies to reading, writing, mathematics, argument, professional judgment, and even moral judgment.

That is why it matters so much that we learn to use AI in a way that increases flexibility and depth in our work, rather than outsourcing thought to it.

Good AI use does not necessarily begin by asking for an answer. It can begin by asking AI to help us think.

But getting an answer is not enough. We also need to learn to see the question more clearly.

There is a large difference between these two requests:

"Write an answer to this."

and:

"Help me understand this. What are the main perspectives? What might I be missing? What are the strongest counterarguments?"

In the first case, there is a risk that we learn nothing ourselves, deepen no position of our own, and strengthen no understanding. In the second, we receive powerful feedback on our own thinking, which both deepens and develops it.

That is one of the most important distinctions. AI should not only help us reach conclusions faster. It can help us see a wider context around the task: what assumptions we are making, which perspectives are missing, which consequences matter, and where our own thinking is still unclear.

Good AI use is therefore not only about getting better answers. It is about learning to see the question more clearly.

AI can be like a good conversation partner who asks us toward clearer thought. It can point out weaknesses in our arguments, show us other sides of an issue, help us put unclear thoughts into words, and remind us of details we forgot. It can also help us see more clearly when we are too certain, too quick to judge, or stuck inside our own tunnel vision.

But for that to happen, the user must be conscious of not taking the shortcut through the task.

A good way to train this is simple: think first, then use AI.

Write down your first idea. Try to answer the question. State your own judgment, even if it is imperfect. Then ask AI to criticize, improve, question, or compare. Then it does not become a replacement for thought, but a trainer of thought.

Another good rule is to ask not only for the conclusion, but for the process toward it.

Instead of asking, "What should I do?" you can ask, "What factors should I examine before I decide what to do?"

Instead of asking, "Is this right?" you can ask, "What assumptions would have to be true for this to be right?"

Instead of saying, "Write this better," you can ask, "What is unclear in this thought?"

These are all questions that deepen our own thinking and make AI a powerful collaborator, rather than the group partner who does everything while we do nothing, and then learn accordingly.

We also need to remember that AI can be wrong. It can sound confident when it is not. It can fill gaps, oversimplify, or accept a false premise if we frame it that way. That is why critical thinking is not less important in a world with AI. On the contrary, it becomes more important.

AI does not release us from responsibility for our own thinking. It makes that responsibility greater.

This does not apply only to children and young people. Adults also need to learn new rules. Employees, managers, specialists, parents, and teachers will all face the same temptation: to let the tool do the difficult parts of thinking for them. Sometimes that is all right. We do not always need to do everything from scratch. But if we always do it, then we pay for the convenience with our own capacity to think.

The right question is therefore not: "Will AI make us stupid?"

The right question is: "How do we use AI so it makes us sharper?"

In summary, the answer is to use it to increase thought, not reduce it. Use it to ask better questions. To see more perspectives. To test our own ideas. To learn faster. To write more clearly. To prepare conversations better. To receive counterarguments and feedback before we go into the world with a half-formed opinion.

Above all: use it to see more clearly what we ourselves are thinking.

AI can become a mirror that makes us more aware of our own thought. But a mirror does not think for us. It shows us something we must respond to ourselves.

That is where the responsibility lies.

We should not teach people to fear AI. We should teach people to work with it without working against themselves.

Because the future will probably not belong to those who use AI without thought. It will belong to those who learn to use it to strengthen their own thinking, deepen their position, and make better decisions.

After-note: cognitive load matters

After this article was submitted, a trusted outside mirror pointed out an important grace-layer: people do not always reach for AI as a shortcut because they are lazy. Often they are tired, rushed, overloaded, or working inside systems that demand more thought than they have bandwidth left to give.

That matters. But even when the reason is understandable, the effect can still be real. If we repeatedly let the machine carry the thinking for us, our own thinking gets less practice.

So the frame should not be shame. It should be care with a backbone: understand the pressure people are under, and still protect the human capacity to think.

I thank A. for deep conversation, critical review, and an important contribution to the structure of the frame.