The Cost of Learning

September 3, 2026

This is the first article in a series exploring what personalization in education could mean in a world where access to knowledge is no longer scarce.

When knowledge is available almost everywhere, the challenge of education is becoming less about providing access to it and more about finding — and continuously adapting — a way for a particular person to interact with it.

Educational effectiveness increasingly depends not on how much knowledge a system provides, but on how well it adapts the way a person interacts with that knowledge: to their individual context and changing needs, and to a changing world that keeps reshaping the knowledge itself, as well as the person’s needs.

1. Access to Knowledge No Longer Explains the Value of a Course, Textbook, or Lecture

For a long time, educational institutions were one of the main organized ways to gain access to knowledge and to the people who could transmit it.

Today, digital technologies have dramatically expanded access to educational resources: books, scientific materials, lectures, online courses, videos, and other formats. UNESCO notes that digital technologies have significantly increased access to educational resources, while emphasizing that their value should not be judged by the mere use of technology, but by how well it helps address actual educational needs.

If expanding access to knowledge was once an important part of educational progress, then in a world of abundant knowledge, simply increasing the amount available is no longer enough to substantially improve the quality of education.

This does not mean educational institutions lose their value. It means their value can no longer be explained primarily by the fact that they provide access.

Access is no longer the only, and in many cases no longer the main, criterion of effective learning. Increasingly, the value of education depends on what happens after access is provided.

2. Access to Knowledge Is Not the Same as Learning

Acquiring knowledge is only one stage.

Between acquiring it and changing a person’s ability to act lies an entire process: understanding, connecting the new with what is already known, practice, feedback, and transferring what has been learned to real situations.

In this article, we treat learning as a process that changes a person’s understanding and ability to act. A fuller account would need to consider prior experience, motivation, cognitive load, and other factors. That is a task for a separate article.

Research on learning supports this: what a person already knows shapes how new information is encoded, understood, and connected to what they already know, and this can strongly affect learning outcomes.[1] Acquiring knowledge, in other words, is not simply a matter of accumulating information — people integrate it with what they already know, forming connections that allow them to use what they have learned in new situations.

If this is the case, the value of education depends not only on what knowledge it provides, but on how a particular person interacts with that knowledge.

3. Is the Same Learning Equally Effective for Different People?

People come to learning with different prior knowledge, experience, goals, and contexts. So the same educational path will not necessarily work equally well for everyone.

Imagine a public speaking course covering speech structure, voice, body language, storytelling, presentation design, improvisation, and Q&A.

Now imagine two participants.

The first has spoken in front of colleagues for years. They structure their thoughts well and are comfortable with slides, but struggle to answer unexpected questions and argue on their feet.

The second is the opposite: good at improvising, but unsure how to structure a speech, build an argument, or hold an audience’s attention through to a convincing conclusion.

Both join the same course and may even share the same broad goal — to become a more convincing speaker. But the skills they need to get there are different.

The first participant gains little from basic voice or slide training and would benefit more from practicing arguments under pressure.

The second needs little additional improvisation practice and would benefit more from learning to structure a speech.

Their goals may differ too. One wants to present convincingly to senior management. The other wants to speak at conferences.

The same course can give both of them access to all of this material. That does not mean they need to go through it the same way.

Personalizing a learning path does not have to mean building a separate program for every person. It can start by identifying which elements of an existing program someone actually needs, which they have already mastered, and which are simply irrelevant to their goal.

But even once we know which skills a person needs, we still do not know how they should best acquire each one.

And this raises the next question.

4. What If the Person Changes During Learning? What About the World Around Them?

Suppose we have determined which skills each participant needs and built different paths for them.

But learning changes the learner.

The first participant practices handling unexpected questions and, a few weeks later, can do it confidently. Their main limitation may now be adapting a speech to different audiences.

Over the same period, the second becomes better at structuring speeches. Their next obstacle might be voice control or holding an audience’s attention.

At every new stage, the person is no longer quite who they were when we designed their learning path.

If we keep following the original program, we risk recreating the very mismatch personalization was meant to solve.

The person is not the only thing that changes.

The world in which they apply their skills changes too. Their role may shift, a new task may appear, an audience may change, professional requirements may evolve, or new tools may become relevant. What was an important educational goal a few months ago may no longer be relevant.

The OECD notes that technological, economic, and social change is reshaping skill requirements and increasing the need for continuous development and updating of competencies.[2]

Therefore, a learning path that fits a person today may not fit them tomorrow.

It is worth separating two distinct problems:

• Knowledge can become outdated.

• But a learning path can stop fitting a person even when the knowledge itself remains relevant.

A person may have already mastered part of the material, changed their goal, encountered a new practical challenge, or discovered a different interest.

We design programs to change people, but people begin changing while going through them. As a result, a program that fit at the start may stop fitting before the learning is complete.

This means personalization cannot be limited to the initial configuration of a learning path.

If people and their world keep changing, the path has to change with them.

Which raises the next question: how do we determine the most appropriate next step for a particular person at a particular moment?

5. Can We Measure the Mismatch Between a Learning Path and a Person?

Suppose we have identified that our first speaker now needs to learn to adapt a speech to different audiences.

How should they learn it?

We could give them theory. We could ask them to analyze examples of effective speeches. We could ask them to prepare several versions of the same speech for different audiences. We could give them feedback after a real presentation. Or we could combine several of these approaches.

For someone else who needs the same skill, the best approach might be entirely different.

The same skill can be learned in different ways. And knowing what a person needs to learn is not enough to know how they should best learn it.

This is a further layer of personalization: choosing not only which skills to include in someone’s learning path, but how each one should be learned by that specific person.

To make that choice, we need some way to compare different learning approaches.

Suppose reading theory works well for one person with relatively little time and effort, while the same material leaves another person unable to connect it to practice. For the second person, reviewing their own speeches with feedback may work far better.

The same method, in other words, can have a different cost for different people.

We’ll call the cost of moving a particular person from their current state toward a particular learning outcome through a particular method the Learning Cost.

It depends not just on the difficulty of the material, but on the interaction between the person, the knowledge, the method, and the context.

This cost should not be understood simply as something to minimize. More effort is not automatically worse, and a more difficult learning experience is not necessarily less effective. Some forms of difficulty can improve long-term retention and transfer.[3]

The relevant question is therefore not “How can we make learning as cheap as possible?”

It is:

“Given a particular person, goal, context, and set of available methods, which learning path produces the desired outcome with an appropriate cost?”

A further distinction matters here.

If someone has already mastered a skill, spending substantial time teaching it again may create little additional value. If a method does not help them acquire what they actually need, the resources spent on it may likewise contribute little to their goal. If an educational activity does not match their current task, some of the resources invested in it may be unnecessary.

We propose calling the portion of Learning Cost that is attributable specifically to such a mismatch Mismatch Cost.

This is not yet a claim that Mismatch Cost is a precisely measurable quantity, or even that Learning Cost can ultimately be decomposed into clearly separable components. It is a working research framework.

The question is whether this framework can help us distinguish not simply between “expensive” and “cheap” learning, but between more and less suitable ways for a particular person to achieve a particular learning outcome.

With this framing, choosing the next step in a learning path becomes a comparison.

A person has a current state and a specific learning goal. Several possible next steps connect the two. Each has an expected outcome and requires certain resources.

Rather than defining an entire path in advance, a system could compare the available next steps at each stage and choose the one that offers an appropriate balance between expected outcome and Learning Cost, given the current context.

Under this approach, personalization means more than removing what a person does not need.

It means continuously choosing the most appropriate way for that particular person to acquire each skill they actually need.

And Learning Cost and Mismatch Cost, if they can be assessed reliably, could become some of the criteria used to make that choice.

6. Personalization Must Be Continuous

We began with the idea that, once knowledge is widely accessible, education’s value depends on more than access alone.

Learning depends on how a person interacts with knowledge. People start from different places, so the same path will not serve everyone equally. And personalization does not end there.

People change as they learn. Their knowledge, skills, goals, and contexts change. The world in which they will apply what they have learned changes as well.

A path that fit yesterday may be suboptimal today.

If that is true, personalization cannot be a one-time choice of the right program.

Personalization may mean not choosing the best program for a person once, but continuously choosing the next action based on their current state, the expected learning outcome, and the cost of that action.

A learning path, on this view, is not a predetermined sequence of lessons but a continuously updated sequence of decisions.

For each next step, we would need to determine:

• which skill or outcome the person actually needs right now;

• which ways of acquiring it are available;

• what outcome each could produce;

• and what Learning Cost each approach would create for this particular person, in their current state and context.

If Learning Cost or Mismatch Cost can be estimated reliably enough, they could become tools for making that decision.

But before building such a system, two more basic questions need answers — questions that later articles in this series will take up:

1. Can we distinguish the necessary costs of learning from the costs that arise specifically from a mismatch between the person, the knowledge, the method, and the context?

2. Can we estimate the cost of different learning methods for a particular person, and use that estimate to choose their next step?

We do not have the answers yet.

But finding them may help clarify what education should look like in a world where access to knowledge is no longer its main value.

Sources

[1] Prior knowledge and learning

W. S. Dochy et al. / research on the mechanisms through which prior knowledge affects learning, including comprehension, encoding, and transfer.

“How does prior knowledge affect learning? A review of 16 mechanisms and a framework for future research.” Learning and Instruction, 2025.

https://www.sciencedirect.com/science/article/pii/S1041608025001207

[2] Changing skill requirements

OECD. “Adult skills and work.”

https://www.oecd.org/en/topics/policy-issues/adult-skills-and-work.html

See also: OECD, Skills Outlook 2025, on how technological, economic, and social change is reshaping skill requirements and increasing the need for continuous skill development.

[3] Desirable difficulties

Robert A. Bjork and Elizabeth L. Bjork. Research on “desirable difficulties” in learning, showing that conditions that make learning or retrieval more difficult in the short term can sometimes improve long-term retention and transfer.

See, for example: Bjork & Bjork, research on desirable difficulties in learning and vocabulary acquisition.

https://bjorklab.psych.ucla.edu/research/