AI vs. the Human Teacher in the Classroom

The most obvious argument made by technology enthusiasts today goes something like this: if AI really can do a better job of individualized instruction, why not bring that advantage into the mass classroom?

Let the teacher conduct the lesson for the whole class while AI simultaneously adapts the learning process to each individual child: explain things more slowly to one student, offer additional examples to another, give a third more challenging problems, and repeat whatever a fourth student does not understand as many times as necessary.

Individualized and Comfortable Learning

At first glance, this looks almost like an ideal model. At last, we would have the kind of individualized instruction that a classroom of thirty students has never been able to provide. But there is a subtle substitution of concepts here: the fact that AI can individualize instruction does not in itself mean that such individualization will make education better.

First, we have to decide what exactly should be individualized, and in which direction.

Suppose a child does not understand fractions. AI can give the student easier problems, explain the concept in different words, provide visual examples, go back to division, or offer additional practice. But the actual pedagogical problem may be something else entirely. Perhaps the child does not need yet another explanation of fractions because the difficulty lies deeper: the student has a weak intuitive grasp of numbers, does not understand the relationship between part and whole, or has failed to master the material that came before.

Perhaps, on the contrary, the student needs to stop receiving hints and struggle with the problem independently for a while. Or perhaps the teacher deliberately gives the student a problem beyond their current level because the goal is to provoke precisely the intellectual leap after which previously disconnected pieces of knowledge finally come together into a system.

All of these are forms of individualized teaching as well.

The ability of AI to adapt to a student therefore answers only the question, “How should we adapt the learning process?”

It does not answer the much harder question: “What, exactly, should we be adapting the student toward at this particular moment, and what do we want to change in the child’s mind?”

There is a second problem: individualized and comfortable learning is not necessarily the best learning.

Comfort is a rather strange criterion for educational quality, although contemporary educational rhetoric often treats it as if a child should be able to make the entire journey of discovery without ever encountering frustration, difficulty, or the experience of not being able to do something.

But learning, by definition, requires cognitive effort. Sometimes a person needs not to receive an immediate hint. Sometimes it is useful not to understand something for a while. Sometimes one needs to take the wrong path, follow it all the way to the end, and discover the mistake independently.

If an adaptive system constantly responds according to the formula — “struggling? give a hint; still struggling? simplify; losing interest? entertain; upset? provide encouragement” — we may end up with a perfectly comfortable passage through the material that gives the student a sense of success without necessarily developing independent thought.

A good teacher sometimes deliberately chooses not to individualize instruction. The teacher may require the entire class to master a particular method of solving a problem, even while knowing five other methods. The teacher may refuse to give a particular child a hint precisely because they understand that the child needs to overcome the difficulty independently. The teacher may ask students to listen to someone else’s solution, compare two approaches, or explain their own reasoning to someone who does not understand it.

In theory, AI could also be programmed to employ such strategies. But then we return to the original question: who decided, and on the basis of what educational model, that at this particular moment a child needs help — while at another moment the child should instead be left in a state of intellectual discomfort?

Personalization itself cannot answer that question.

Indeed, a system that is too well adapted to a student may quietly construct an intellectual environment around that student that constantly accommodates their particular traits, weaknesses, pace, and preferences.

Yet one of the most important purposes of education is precisely to teach a person, gradually, how to function in a world that is not always going to accommodate them.

Intellectual Environment

Finally, there is another factor that often gets lost in discussions about individualized learning: a child at school learns not only from the teacher and not only from textbooks and other instructional materials.

The child learns within a shared intellectual environment.

A student hears a classmate give the wrong answer and tries to understand exactly where the mistake was made. They see a method of solving a problem that never occurred to them. They discover that the same problem can be understood in radically different ways. They try to explain an idea to someone who does not understand it, argue, clarify, revise their arguments, observe other people’s mistakes and discoveries.

None of this is incidental social noise interfering with “efficient individualized learning.” It is an essential part of learning how to think.

A model in which every student is given a perfectly calibrated personal AI tutor could therefore, in theory, strengthen individualized instruction while simultaneously weakening one of the most important mechanisms of education: the collision of one person’s way of thinking with another’s.

The question is not whether AI can make education more individualized. Of course it can. That is precisely where its enormous potential lies.

The real question is: why have we decided that maximum individualization should be the ultimate goal of good education?

Education should not merely adapt the environment to the individual. It should also gradually teach the individual how to exist in an environment that does not adapt itself to them. It should not merely help a student follow the most convenient path; it should also force them to encounter the unfamiliar, the difficult, and the unfamiliar ways of thinking of others.

AI may therefore become an extraordinarily valuable tool for individualized support. But treating individualization itself as proof of improved educational quality is a fundamental mistake.

A comment from AI itself: Humanity has a remarkable habit of declaring whatever it can measure to be good simply because it has finally learned how to scale it.