Humanity has learned to create knowledge, but has not learned to transmit understanding

Humanity has never had such access to information as it does today.

Practically anyone can find a textbook on quantum physics, lectures from the world's best universities, programming documentation, scientific articles on genetics, or courses in economics, history, engineering, or medicine in just a few seconds.

And with the advent of artificial intelligence, even the need to manually search for most of this information disappears. You can simply ask a question and get an answer.

It would seem we have solved the problem of access to knowledge.

But we have not solved a significantly more important problem.

Access to information is not access to understanding.

You can give a person an absolutely accurate scientific text containing extremely valuable information and transmit practically nothing to them.

Because to understand information, a person must already understand something.

And this is precisely where one of the fundamental problems of the modern system of knowledge transfer lies.

The Concept — The Elementary Unit of Understanding

We are accustomed to measuring information in words, pages, books, lectures, hours of study, gigabytes of data.

But human thinking works differently.

We do not store books inside ourselves.

We build concepts.

A concept is a complete semantic unit that a person is able to recognize, mentally reconstruct, and use to understand something next.

For example, the word "pressure" by itself is not knowledge.

For one person, it represents a school definition they have almost forgotten.

For an engineer, it is an entire system of physical models.

For a plumber, it is also practical experience: how water behaves, what happens when pipe diameter changes, why the system makes noise, where problems arise.

The word is the same.

The concepts behind it are different.

And vice versa: a person can perfectly well understand a certain idea without knowing the accepted scientific term at all.

For example, they understand that a system can automatically reduce its impact when the result becomes too large, and increase it when the result becomes too small.

They already understand the mechanism.

They simply might not know that the term "negative feedback" exists.

Therefore, a term and a concept are not the same thing.

A term is a pointer.

A concept is what unfolds behind that pointer in a person's head.

This distinction seems small, but it completely changes the perspective on education.

Humans Do Not Accumulate Knowledge Linearly

The view of learning as the sequential filling of the head with information is too primitive.

Something else happens.

A person masters a new concept through existing ones.

At first, a complex idea has to be explained at length — through examples, analogies, and familiar phenomena.

Then the person understands it.

After that, the entire construct can be packaged into a short name.

Now it is enough to say:

"negative feedback"

— and a model already familiar to the person unfolds in their head.

Several paragraphs of information have turned into two words.

But something even more important has happened.

The new concept can now be used to understand subsequent concepts.

Therefore, learning does not look like this:

knowledge + knowledge + knowledge + knowledge.

Rather:

understood → compressed → reused → understood something more complex through this → compressed again.

A person is constantly restructuring their own language of thought.

The richer this system is, the more complex information can be transmitted to them in a short message.

A phrase completely meaningless to one person can transmit a huge amount of knowledge to another in a few seconds.

This is precisely why two people can read the same scientific article and effectively extract completely different amounts of information from it.

Transmitting Information Does Not Mean Transmitting Understanding

Let us imagine the sentence:

Modern systems use knowledge tracing in conjunction with knowledge graphs to build adaptive learning paths, taking into account the learner's cognitive load and prerequisites.

For an expert, this may be a completely normal and compact formulation.

For a vast number of people, this is informational noise.

Not because they are incapable of understanding the idea.

But because the message is encoded through concepts that do not yet exist in their head.

One could express roughly the same thought differently:

A computer can gradually determine what a person already understands, what knowledge they still need, and in what sequence it is best to provide it, so as not to force them to study too much new material at once.

Now the message has become accessible to a completely different audience.

The core information has barely changed.

The way it is encoded has changed.

And this allows us to formulate a most important principle:

A good explanation is determined not only by what the transmitter knows. It is determined by what the receiver already knows.

A universally good explanation may not exist at all.

Scientific Knowledge is Compressed More and More Tightly

As any field develops, it generates its own language.

This is necessary.

It is pointless for specialists to explain basic ideas from scratch to each other every time. Complex constructs receive names and turn into new building blocks of thought.

A single term begins to replace a page of explanations.

This is extremely efficient information compression.

But it comes at a price.

To unpack such a term, a person must already possess the corresponding concept in advance.

Therefore, every developed field gradually creates its own decoder.

Physicists understand physicists.

Biologists understand biologists.

Economists understand economists.

Programmers understand programmers.

The problem becomes especially noticeable at the boundaries of disciplines.

A highly qualified specialist can open a paper from a neighboring field and find that a significant portion of the text is practically incomprehensible to them.

They know how to read every individual word.

But they do not know how to unpack the concepts behind them.

A paradox arises.

Humanity knows more and more, but it becomes harder and harder for an individual to gain access to all of human knowledge.

Modern Education Tries to Solve the Wrong Problem

Traditional education usually starts with the subject matter.

Here is the syllabus.

Here is the sequence of topics.

Here is the first semester.

Here is the second.

Here is the reading list.

And then thousands of completely different people are pushed along this route.

But these people have different heads.

One already knows half of the required concepts.

Another lacks a few basic ones.

A third perfectly understands the idea itself, but does not know the professional term.

A fourth knows the term and is able to repeat the definition, but actually has no concept behind it.

A fifth came solely for one specific capability, for which 80% of the course will never be needed.

We try to adapt the person to the curriculum.

It is more logical to do the opposite.

You Must Not Start with the Subject. You Must Start with the Goal

Suppose a person wants to acquire a certain capability.

Not "learn programming."

But, for example:

learn to independently create simple digital tools for your business.

This is a completely different problem statement.

Now we can ask:

What does this specific person actually need to understand in order to acquire this capability?

What do they already understand?

What concepts are missing?

Which traditionally studied things has modern technology already made optional?

What is enough for them to know superficially?

What is necessary to understand deeply?

Which intermediate concepts can be skipped entirely?

And suddenly it turns out that a traditional path of a thousand conventional units of knowledge can be shortened to two hundred.

Not at the expense of a worse result.

But by removing information that is not needed for the set goal by this specific person in the existing technological world.

Technologies Constantly Destroy Old Mandatory Knowledge

At the dawn of personal computers, people really had to be taught to launch the graphical shell with the win command.

It was necessary to explain what a window, an icon, and a double click were.

Today, a vast number of people will never need this knowledge.

Technology has hidden the corresponding level of complexity.

But education possesses enormous inertia.

We continue to build curricula based on historically established chains:

to understand D, first study A, then B, then C.

Although the modern world may already allow:

A → D.

Or even:

D.

This is happening especially fast now thanks to AI.

What was a necessary professional competency five years ago may tomorrow become an internal technical detail of an intelligent system.

Consequently, the chain of mandatory concepts cannot be fixed.

It must be continuously revised.

Artificial Intelligence Makes a Different Learning Architecture Possible

Before the advent of modern AI systems, personalizing knowledge transfer in this way was practically impossible.

A textbook author is forced to write a single version of the text.

An instructor can slightly adapt the explanation to the audience, but is physically incapable of constantly maintaining an accurate model of the knowledge of each of hundreds of students.

AI is capable of working differently.

The system can gradually build a living model of a specific person's concepts.

Not simply:

knows mathematics — 72%.

But:

understands this concept well; uses this one, but does not understand the mechanism; knows this one under a different name; lacks the necessary causal connection here; this term is unfamiliar to the person, although the idea itself is already clear to them.

After that, the explanation task becomes computable.

We have:

the person's goal;

their current system of concepts;

the vast graph of human knowledge;

modern technological capabilities.

We need to find the minimal chain of changes that will lead the person from their current state to the required capability.

What Needs Minimization is Not Learning Time

Usually we ask:

How can we explain this course faster?

This is a secondary question.

The first should be:

Why should a person study this entire course in the first place?

If the curriculum contains a thousand concepts, and a hundred are necessary for a specific person's goal, speeding up the study of the thousand is poor optimization.

First you need to throw out nine hundred.

Of the remaining hundred, the person already knows a portion.

Some can be replaced by a more general model.

Some are enough to simply denote: "know that such a field exists and when to turn to it."

Some truly need to be understood deeply.

Only after this does it make sense to optimize the explanation.

We cannot force the human brain to learn twenty times faster.

But sometimes we can reduce the amount of what they actually need to study by a factor of twenty.

This is a fundamentally different way of accelerating human development.

At the Same Time, the Minimal Path is Unique to Everyone

Suppose it is necessary to explain a certain control system to someone.

It is more convenient to explain it mathematically to an engineer.

To a programmer — through program state and a feedback loop.

To an entrepreneur — through prices and demand fluctuations.

To a plumber — through pressure and a valve.

These are not four different concepts.

These are four different ways to arrive at a sufficiently similar internal model.

Once understanding has emerged, one can say:

In general terms, this is called such-and-such.

Now the professional term has ceased to be an unknown word.

It has become a compressed pointer to an already existing concept.

The next explanation can already be built through it.

Therefore, the learning system must constantly change along with the person.

The explanation they need today will become annoyingly primitive in six months.

Their internal decoder has changed.

This means the system's language must change as well.

First, People Must Be Taught How to Learn

But a problem exists even deeper than any technology.

A vast number of people do not understand at all what happens at the moment of learning.

A person comes to a course.

Listens to a lecture.

At some point, they stop understanding what is happening.

They keep listening.

Subsequent explanations are built on top of the misunderstood concept.

Twenty minutes later, three concepts are not understood.

An hour later, half the material is not understood.

And the person concludes:

"I'm not smart enough."

Although the technical description of the problem may be completely different:

a single node is missing in the chain of understanding.

Perhaps the instructor used a term, assuming it was already known.

Perhaps they skipped a causal link.

Perhaps they used an abstraction that would have been easier to show to this specific person through an example first.

Perhaps the entire intermediate material is not needed for their task at all.

Therefore, one of the first skills a person must be taught is diagnosing their own lack of understanding.

Not:

I don't understand physics.

But:

In this explanation, I understand A and B, but I don't understand why C follows from B.

Or:

I don't know what the word X means.

Or:

The mechanism itself is clear to me, but I don't understand why it is applied here.

Non-understanding ceases to be an evaluation of the person.

It becomes a specific technical malfunction in the chain of concepts that can be found and fixed.

This is what learning how to learn actually means.

AI Can Both Increase and Decrease the Intellectual Gap

Yet another problem arises here.

We often assume that AI available to everyone will intellectually level people out.

This is not necessarily true.

Let us imagine two people with identical access to the same model.

The first uses it like this:

Do this.

AI does it.

The person gets the result.

The second:

Why did it turn out this way? What alternatives exist here? What am I not understanding? What more general principle is this similar to? Where else is it applied? What assumptions did we tacitly make just now?

AI answers.

The answer gives birth to new questions.

New answers create new concepts.

New concepts allow the person to ask questions that could not even have occurred to them before.

A positive feedback loop arises:

understanding → stronger questions → stronger use of AI → new understanding → even stronger questions.

Therefore, identical access to AI can lead to completely different results.

For one person, productivity grows primarily.

For another, the person themselves grows simultaneously.

And the gap between them may not shrink, but accelerate its growth.

In such a case, AI becomes an amplifier of intellectual dispersion.

The Complexity of the Result No Longer Reflects the Complexity of the Person

This is a new state for humanity.

Previously, a relatively strong connection existed:

to create more complex things, a person had to become more competent themselves.

AI is gradually breaking this connection.

A person can create a software product with practically no understanding of programming.

They can obtain a complex analytical report without mastering the corresponding analytical methods.

They can manage processes whose internal workings they barely understand.

This is not necessarily bad.

But a fundamental distinction appears:

the complexity of the result produced by a person ≠ the intellectual complexity of the person themselves.

It is possible to produce a huge volume of intellectually complex output while importing almost all of the complexity embedded within it from AI.

Therefore, productivity can no longer be considered a sufficient indicator of human development.

We must ask a different question:

What changed in the person's head after completing this work?

But We Must Not Turn Every Task into a Lesson

It is easy to draw the wrong conclusion from this:

This means AI should force a person to think independently every time and explain everything it does.

This will destroy the very meaning of intellectual automation.

Human attention is limited.

You cannot force a person to study the internal workings of every tool they use.

Precisely for this reason, the idea of the minimally necessary concept becomes central.

The system must be able to distinguish:

the person needs to understand this;

it is enough to know that this exists regarding that;

this will only be needed when a specific situation arises;

this can be completely handed over to the machine.

The goal is not to maximize a person's knowledge.

The goal is to maximize their ability to understand and act at a minimal cognitive cost.

The Concept Becomes the Ultimate Information Unit

An architectural principle follows from this.

Not an article.

Not a book.

Not a lesson.

Not a course.

Not a discipline.

A concept.

Large informational constructs are only needed because sometimes a long sequence of explanations is required for a person to build a single concept.

But the ultimate result is not a read page.

The result is a semantic construct that has emerged in the head, which the person is able to use further.

Therefore, a concept can be viewed simultaneously as the elementary and the ultimate unit of knowledge.

Elementary — because further learning is built out of already mastered concepts.

Ultimate — because the task of information transfer ends not when a person has read the text, but when the required concept has been successfully reconstructed in their model of the world.

Text is the transport.

A term is the address.

An explanation is the decoding method.

A concept is the result.

Perhaps We Need a New Theory of Knowledge Transfer

Classical information theory answers questions of signal transmission magnificently.

But for human learning, it is not enough to know how much information was transmitted.

You can transmit a gigabyte of text and change practically nothing in the receiver's head.

And you can utter a single phrase — and restructure their view of an entire field.

This means we are interested in a different magnitude.

Not:

how much information was transmitted?

But:

what useful change occurred in the receiver's concept system, and at what cognitive cost?

This allows us to define learning effectiveness differently.

A good explanation is not the most complete one.

Not the most scientific one.

Not the shortest one.

And not even necessarily the most accurate in the maximum number of details.

A good explanation is one that creates a sufficiently accurate concept for the next necessary action or understanding with minimal expenditure of human attention.

And then this concept becomes the building block for the next one.

From Universal Education to a Living System of Concepts

As a result, a completely different architecture emerges.

There is no single correct course.

There is no fixed set of mandatory prerequisites.

There is no single optimal explanation.

There is a person in state H₀.

There is a capability they want to acquire — T.

There is the current state of the world and technology — W.

And there is a certain minimal transformation:

H₀ → H₁ → H₂ → … → T

Each transition is the emergence or restructuring of a concept.

After each transition, the route can be recalculated.

Perhaps a new concept made it possible to skip the next three.

Perhaps an old gap was discovered.

Perhaps the person has already built the necessary connection independently.

Perhaps technology has changed and an entire branch of knowledge is no longer needed.

Therefore, the graph must be alive.

And the person's model must be alive.

The Main Task — Increase the Throughput Capacity of Human Development

Human physiology changes slowly.

We cannot simply increase human memory tenfold or make the brain absorb complex models at the speed of a computer.

But a huge portion of today's cost of learning is not determined by physiology at all.

It is determined by poor information routing.

We teach what is already known.

We explain through unknown terms.

We transmit details that will never be needed.

We force the study of historically accumulated prerequisites.

We use identical explanations for different people.

We confuse the recognition of terms with the understanding of concepts.

We fail to notice small gaps that block entire subsequent chains.

And we force a person to independently search for a path through the gigantic space of human knowledge.

For the first time, AI allows us to attack almost all of these losses simultaneously.

Not to make the brain faster.

To make the path shorter.

Not to give a person more information.

To transmit less — but precisely that which they are able to turn into understanding.

Not to force everyone to know everything.

To give everyone a minimally sufficient system of concepts that allows them to move forward.

And perhaps this very thing will become one of the main changes in education in the era of artificial intelligence.

Because humanity's problem has long not been a shortage of information.

The problem is how much of a human life it takes today to turn accessible information into understanding.