Can Talking with Artificial Intelligence Make You Smarter?

For a long time, having a question was only the beginning of the problem.

Imagine wanting to understand why the sky turns red at sunset. Depending on the era, finding an answer might mean locating someone who knew, looking through an encyclopedia, going to a library, finding the right book, and figuring out where the explanation was hiding.

The question took seconds to occur to you. The answer could take days.

And a good answer often created another question. If the explanation involved light in the atmosphere, you might wonder about waves. Then frequency. Then human vision. A question that began with a glance at the sky could end up crossing physics, chemistry, and biology.

Every change of subject meant another search.

The knowledge was available. Getting to it was the problem.

From the Library to the Conversation

Libraries were one of humanity’s great solutions to that problem. Bringing books together in one place meant bringing together experiences, discoveries, and ideas from other places and other times.

But you still had to find the book, get access to it, and read enough to discover whether the answer was actually there.

Catalogs and classification systems made that easier. Computers, digital databases, and the internet made it easier still.

Then came search engines.

In the 1990s, Larry Page and Sergey Brin developed at Stanford the search engine that would become Google. The breakthrough was not producing all the knowledge on the internet. It was making that knowledge much easier to find.

The distance between a question and information shrank.

Still, a search engine found documents. You searched, chose a result, opened a page, read, went back, revised the search, and continued.

Conversational artificial intelligence introduced another change.

It did more than reduce the distance between a question and an answer. It reduced the distance between an answer and the next question.

That changes more than it might seem.

When Conversation Is Already Practice

When a conversation requires the user to formulate and reformulate questions and follow the answers, something so simple it almost goes unnoticed is happening.

To converse with artificial intelligence, you have to communicate.

If you type, you have to put what you are thinking into words. If you send a voice message, you have to explain the problem so that another intelligence—even an artificial one—can understand it.

Sometimes it gets you wrong. You notice, try again, and rephrase:

— That’s not what I meant.

Then you reorganize the explanation, add something you forgot, and realize the problem wasn’t entirely clear to you either.

The more complex the conversation, the more you are likely to need to explain precisely what you want.

Then the movement reverses. The AI responds. And you have to understand the response.

Maybe it is three paragraphs. Maybe fifteen. You read, interpret, identify what you did not understand, and formulate your next question.

There is no need to claim that talking with AI will automatically turn someone into a great writer or public speaker. But something elementary is happening: communication and reading are being practiced.

And practice matters.

Someone who spends hours conversing this way may end up reading thousands of words without formally deciding, “Now I’m going to practice reading.” You might come in to ask how to fix a faucet and, a while later, find yourself reading about water pressure, materials, corrosion, and somehow the history of sanitation.

The tool also has a feature that encourages you to keep going: it answers.

A book can be excellent, but it doesn’t notice that you closed it on page 37. A conversation responds to your next question.

That is where the interaction starts to get more interesting.

When Not Understanding Stops Being Everyone’s Problem

Imagine someone who left school and returned years later to a classroom with thirty or forty students.

The teacher explains a topic. The student doesn’t understand. Asks a question. The teacher explains again. It is still unclear.

A good teacher can try another approach, but there is a limit that has nothing to do with teaching ability: an entire class is sitting there.

Some students understood. Others have questions too. Some need more examples. Others would like the lesson to move on.

The teacher has to manage all of that at once.

An AI does not face that constraint when it is talking with a single user.

— I don’t understand.

It explains again.

— Still don’t.

It changes the example.

— Explain it without the formula.

It tries another route.

— Now I understand that part, but not the part before it.

It goes back.

This can happen as many times as necessary without making thirty-nine other students wait—and without the small human embarrassment of raising your hand for the fifth time.

More significantly, the explanation can change.

You can say you understand concrete examples better, ask for simpler language, request exercises, or ask the AI to slow down.

You can even reverse the relationship:

— Don’t give me the answer. Ask questions until I can get there myself.

At that point, the AI stops being merely an answer machine and begins to function as a tutor.

That possibility is already being studied under controlled conditions. In 2025, Harvard researchers published an experiment with college physics students comparing active learning in class with an AI tutor carefully designed around pedagogical principles. In that setting, students who used the tutor achieved greater learning gains in less time and reported more engagement and motivation.

The important detail is “carefully designed.”

Putting a text box in front of a student is not enough to make pedagogy happen spontaneously.

But the experiment shows the scale of the possibility: individual, adaptable tutoring, continuously available to people who have access to the tool.

The Question That Leads to the Next Question

Personalization is only part of the change. The other part is speed.

Ask the AI why the sky turns red. The answer mentions light scattering. You ask why some frequencies scatter more. The response raises a question about the human eye. From there comes a question about the evolution of vision.

Within minutes, you have crossed several fields without interrupting the investigation to locate a new source at every turn.

That does not mean you have become an expert in all of them.

Speed of exploration is not depth of knowledge.

There is also a practical problem: beginners may be the people least equipped to recognize when an AI explanation is wrong. A false answer can sound perfectly convincing to someone who does not yet know enough to question it. Learning with AI therefore requires a kind of skepticism: asking for sources, checking them when necessary, and distinguishing a plausible explanation from verified information.

Despite those limits, speed matters because it keeps the investigation alive.

Previously, a connection to another field might mean locating a book, getting the book, reading part of it, and returning to the question days later. Now you can explore the next hypothesis while the previous one is still in memory.

One answer can open three paths. You follow one. Go back. Try the second. Notice that it connects to something discussed forty minutes earlier. The investigation starts to feel less like a line of books and more like a network.

And here is a question worth saving for another article: what happens to the way we think when we compress the interval between associations so dramatically?

For now, it is enough to notice that the interval has changed.

AI Can Think for You. That Is Precisely the Problem.

So far, we seem to have found a remarkably convenient machine: it talks, explains, adapts, crosses subjects, and never complains that it has already answered that three times.

Naturally, there is a complication.

It can also do the work for you. Write the text. Solve the exercise. Summarize the book. Organize the argument.

Hand you an answer so complete that all you have to figure out is where the copy button is.

The concern about outsourcing cognitive work is therefore real.

A study presented at CHI 2025 by researchers from Microsoft Research and Carnegie Mellon University examined 319 professionals and 936 reported examples of generative AI use at work. Greater confidence in AI was associated with less self-reported exercise of critical thinking.

But another part of the same study is interesting: cognitive work did not simply disappear. Some of it moved. Users shifted toward checking information, integrating responses, and supervising what the AI had produced.

That brings us to a more useful distinction than simply asking whether AI is good or bad for intelligence.

And there is an important asymmetry: AI makes both learning and avoiding learning easier. The same system that lets you insist on an explanation twenty times can deliver the finished result in seconds.

It is not reasonable to imagine that every user will always choose the more demanding path—especially when the shortcut is one sentence away.

The tool can do it for you, explain it to you, or think with you.

In the first case, you ask for a result; in the second, for understanding; in the third, you present a hypothesis, challenge the response, ask for evidence, request an opposing perspective, try explaining in your own words, and use the tool to find weaknesses.

The technology is practically the same. The cognitive exercise is not.

A systematic review published in 2026, bringing together 88 empirical studies of large language models in education, found precisely this mixed picture: benefits in performance, engagement, and accessibility, alongside concerns about overreliance, reliability, privacy, and cognitive effects that vary by context.

Perhaps the main risk is not the existence of a tool that can think for us. Perhaps it is forgetting that we can choose not to ask it to.

So far, we have mainly followed the exploratory mode: one question opens another and the path branches. But not all learning works best that way. Sometimes we want to explore; at other times, we need to follow a sequence. One important feature of AI is the possibility of switching between those two modes.

And What About Books?

Now we can return to the book.

A good book has an extraordinary advantage: someone built the path before you arrived.

The author chose what to present first, what to save for later, which concepts require foundations, and where to take the reader.

That creates something a spontaneous conversation does not necessarily have: a deliberate learning progression.

A book can also take you somewhere you would never choose on your own. The next chapter does not change simply because something else caught your interest.

That limitation can also be a virtue.

But the opposition between books and AI begins to dissolve when we realize that AI itself can build a structure before the conversation starts.

Imagine wanting to study mathematics in depth.

Instead of asking about random topics, you can begin by explaining where you are and where you want to go. You can ask the AI to consult official curriculum documents, a university’s course requirements, or a set of reliable references and build a progression of study.

First comes the map.

Foundations.

Modules.

The sequence of topics.

Books or units.

Chapters.

Exercises.

The student can review that structure before starting and demand the sources used. If you want to follow your state education department’s standards or a particular university’s curriculum, ask the AI to consult the official documents—simply saying “follow the state standards” does not magically turn the machine into a representative of the state education department.

Then the studying begins.

You ask for the first chapter. Read, ask questions, do the exercises, and move on.

In the next chapter, a difficulty appears.

— I don’t understand.

The explanation changes.

— I think I’m missing some earlier foundation.

The AI can suggest a review.

After a few modules:

— Analyze where I’m struggling and see whether we need to adjust the rest of the program.

Now we have something different from both a traditional book and an improvised conversation.

There is structure, but it can adapt.

There is a sequence, but the reader can talk with it.

The material can be planned beforehand and recalibrated along the way.

The book has not disappeared.

In a sense, it has become something you can converse with.

So Can AI Make Someone Smarter?

It would be pleasant to end with a “yes.” It would also be premature.

Here, “smarter” does not mean demonstrating an increase in a psychometric measure of intelligence. What this article can support is more specific: AI can expand and make more frequent the exercise of capacities involved in learning—formulating, reading, interpreting, comparing, explaining, checking, connecting, and revising your own reasoning.

Those gains are not automatic. The same tool can be used to avoid almost all of them. It reduces the friction involved in learning and, at the same time, the friction involved in outsourcing the task.

That is why the effect depends neither solely on the technology nor solely on the user. It also depends on how the tool is designed and what kind of interaction it encourages—and we do not yet know the effects of decades of everyday life with systems like these.

Perhaps the next question is incomplete.

It is not enough to ask:

— What is artificial intelligence doing to our intelligence?

There is another question:

— What are we doing with our intelligence while we use artificial intelligence?

Perhaps much of the answer lies in that difference.