What Does It Mean to Learn?

A New Scientific Definition - and an Important Question About What Comes Next

Scientists Are Still Asking a Surprisingly Basic Question

What does it actually mean to learn?

It sounds like a question science should have settled long ago.

Students learn. Animals learn. Neural networks learn. Organizations learn. Modern artificial-intelligence systems are routinely described as learning systems.

Yet the word learning has not always meant the same thing across psychology, neuroscience, education, biology, computer science, and other fields. Different disciplines have developed different definitions because they study different kinds of systems, mechanisms, and outcomes.

That creates a scientific problem.

If researchers use the same word to describe meaningfully different phenomena, communication becomes harder. Theories become more difficult to compare. Findings developed in one discipline may not translate easily into another.

In July 2026, Martin Lövdén, Isabelle Hansson, and Toms Voits addressed this issue directly in npj Science of Learning in a perspective article titled Learning: One Definition to Rule Them All? Their objective was ambitious: identify an umbrella definition broad enough to operate across disciplines and systems while remaining precise enough to support scientific discussion.

That effort matters because a definition does more than supply terminology. A useful definition establishes what phenomenon we are actually trying to explain.

What changes because learning occurred?

To answer that question, however, we first need to understand what makes learning so difficult to define.

Why Is Learning So Difficult to Define?

In everyday language, learning seems straightforward.

We might say someone has learned when that person acquires knowledge, develops a skill, understands an idea, remembers information, or changes behavior.

But each answer creates another question.

What counts as knowledge? What qualifies as understanding? How long must something be remembered before we call it learning? Does behavior have to change?

Would the same definition still work for a child learning multiplication, an animal adapting to its environment, a biological system responding to information, and an artificial neural network updating its internal state?

The difficulty is not simply that researchers disagree over wording.

The systems themselves can be profoundly different.

Human learning may involve language, memory, reasoning, emotion, social interaction, and prior experience. Machine learning may involve optimization of internal parameters. Biological systems may change in response to environmental information without anything resembling conscious thought.

If a definition becomes too narrow, it may apply only to one kind of system.

If it becomes too broad, it may lose scientific usefulness.

Lövdén and colleagues therefore propose a different strategy: establish a sufficiently general scientific umbrella while allowing individual disciplines to retain the more specific operational definitions required for their own research.

A shared definition does not require every field to explain learning through the same mechanism.

It gives researchers a common starting point for identifying the phenomenon.

The Proposed Definition

Environmental Information -> Processing -> System Change -> Changed Behavioral Potential

Lövdén, Hansson, and Voits describe learning as a process in which a system processes information from its environment in a way that changes properties of the system and, consequently, changes its behavioral potential and the way it may respond to that information.

Several parts of this formulation matter.

First, learning is a process. It is not simply the possession of information after the process is over.

Second, information from the environment must be processed in a way that changes the system. Exposure alone is therefore not sufficient. Something about the system must become different.

Third, that change has consequences. The system's potential responses are altered.

This allows the proposed definition to remain broad. The system might be a person, an animal, a cellular network, a group, or a machine. Their mechanisms can differ dramatically while the general relationship remains:

Information -> Change -> Changed Potential

That is the value of an umbrella definition: it identifies a common structure without pretending that every system learns in the same way.

But behavioral potential introduces an especially important distinction.

The consequences of learning do not always have to be immediately visible.

Learning Is Not Simply Behavior

Imagine that a student learns how to solve a particular type of equation.

The student does not solve equations every minute of the day. Nothing observable may happen for hours.

But the absence of visible behavior does not necessarily mean the learning disappeared. The student may still possess the capacity to solve the equation when the situation requires it.

This is why the proposed framework emphasizes behavioral potential rather than only observed behavior.

What a learner does now != everything the learner is capable of doing

That matters enormously for education.

A student who remains silent during a lesson may still have learned. A student who is not currently solving a problem may still possess the knowledge necessary to solve it later.

Conversely, observing a successful behavior does not automatically reveal everything about the underlying learning that made it possible.

Learning may therefore change what a system could do without producing an immediately visible behavior.

That distinction also changes how assessment should be interpreted.

A test score can provide evidence about learning.

It is not necessarily identical to learning itself.

Learning -> Changed Behavioral Potential

Observed Performance = One Possible Expression of That Potential

What can the changed learner now do?

Learning and Capability Are Not the Same Question

Consider two students who receive the same lesson.

Both encounter the same information. Both complete the same instruction.

Yet afterward, their capabilities may differ.

One student may remember the information but struggle to explain it. Another may explain it clearly but struggle to apply it to an unfamiliar problem. Another may recognize when the knowledge is useful, combine it with something learned previously, or adapt it when the original method no longer works.

These differences matter because learning and capability describe related but different aspects of development.

Learning asks: What process changed the system?

Capability asks: What can the changed system now do?

Learning Process -> System Change -> Changed Behavioral Potential -> Question of Capability

Capability is not being substituted for learning.

It is a new question made possible by learning.

That distinction is especially important in education. If our only question is whether information was encountered or whether a student produced one correct response, we may learn relatively little about the range of change that occurred.

What has learning made possible?

Three Questions That Should Not Be Confused

What is learning?

This is the definition question. It asks what phenomenon should count as learning.

How does learning occur?

This is the mechanism question. Different disciplines may answer it differently.

What changes because learning occurred?

This is the consequence question. It asks what becomes different about the learner or learning system afterward.

Definition != Mechanism != Consequence

A scientific definition can identify learning without specifying every mechanism by which learning occurs.

Likewise, establishing that learning occurred does not automatically tell us the magnitude, durability, range, or usefulness of the resulting change.

This is where the educational problem becomes particularly interesting.

Where the Science of Progressive Learning Enters

The Science of Progressive Learning should not be presented as a replacement for the definition proposed by Lövdén and colleagues. The two perform different jobs.

What phenomenon are we calling learning?

How can learning progressively increase capability and advance understanding in ways that give a learner greater ability to adapt and respond to the world?

A definition establishes the phenomenon.

A research program can then investigate how its consequences develop, how those consequences can be measured, and which instructional conditions improve them.

Definition of Learning -> Consequences of Learning -> Progressive Development

Keeping these constructs separate allows each to be investigated more carefully.

One way to investigate those consequences is through applicability: progressively changing what a learner must do with the same underlying knowledge.

What Can the Learner Do With What Was Learned?

Knowing something and being able to use it are related, but they are not identical.

A student may recognize a formula without knowing when to use it. A student may recite a definition without recognizing the concept when it appears in an unfamiliar setting.

Does the learner know it?

What can the learner do with it?

Applicability does not completely define understanding. Rather, it provides one possible source of evidence about capability.

We can investigate that evidence by changing the demands placed on the same underlying knowledge.

One Formula, Increasing Demands

Suppose a student learns the formula for the area of a circle: A = pi r^2.

Can the student recall the formula?

Can the student explain what r represents?

Can the student calculate an area when the radius is given?

Finally, change the appearance of the problem. Place the circle inside a larger geometric situation without explicitly telling the learner that the area formula is needed.

This is where that knowledge is useful.

The formula has not changed. The underlying concept has not changed. What changes is the demand placed on the learner.

Recall -> Explain -> Apply -> Recognize When to Apply

A correct response at one level does not automatically establish performance at every other level.

The goal is not simply to make the questions harder.

It is to vary what the learner must do with what was learned.

Stable Knowledge, Evolving Demands

Stable Knowledge Object + Evolving Task Demand -> Richer Evidence About Capability

If we simultaneously change the knowledge, context, difficulty, and required operation, determining what a learner's response actually tells us becomes harder.

But if the underlying knowledge remains relatively stable while its required use changes, we can ask increasingly informative questions about the capability associated with that knowledge.

This does not establish a universal hierarchy through which all learning must progress.

It gives us a structured way to ask different questions about the same learned material.

And that principle can travel beyond mathematics.

From a Formula to a Word

Consider the word stagnate.

Can the learner define stagnate?

Can the learner distinguish stagnate from a nearby but inappropriate word?

Can the learner use it correctly in a sentence?

Can the learner recognize when someone else has used it incorrectly?

Can the learner connect it to related concepts?

Can the learner recognize stagnation in an economy, a biological system, academic development, or technological progress?

Define -> Discriminate -> Use -> Evaluate -> Connect -> Apply Across Contexts

Vocabulary therefore gives us a relatively simple environment in which changing forms of use can be observed.

From Application to Transfer

Suppose a learner first encounters stagnate in a sentence about economic growth.

Later, the learner recognizes its relevance while discussing academic progress, biological development, or technological innovation.

The learner is no longer responding only to the original example.

The knowledge is being used across changing situations.

That makes transfer an important candidate dimension of capability.

But a successful transfer task does not tell us everything about understanding. Nor does failure on one transfer task prove that no learning occurred.

Learning != Capability != Any Single Measurement of Capability

Vocabulary now gives us something more than an illustration.

It gives us an environment in which hypotheses about capability can begin to be tested.

Does the Organization of Knowledge Matter?

Knowledge can be learned as isolated pieces, but it can also be related to other knowledge.

Consider progress and stagnate, or expand and contract. These pairs can help organize ideas around change, but they are not interchangeable synonyms or positions on one universal semantic scale. Their meanings and relationships depend on use.

This suggests a research question.

Suppose one student learns vocabulary primarily as isolated word-definition pairs while another learns the same words together with carefully selected relationships among them.

Would the second learner later show better discrimination, retention, application, or transfer?

Perhaps.

Plausible Educational Idea != Established Scientific Result

The next step is testing.

Turning the Idea Into a Test

Researchers could compare two groups learning the same vocabulary.

One could study the words primarily as individual word-definition pairs. The other could study the same words while also receiving carefully controlled semantic relationships among them.

Researchers could then compare outcomes such as immediate accuracy, discrimination, delayed retention, application, transfer, and the amount of learning required to reach proficiency.

But the comparison would only be meaningful if the instructional conditions were reasonably matched.

If one group simply receives more study time, examples, exposures, or retrieval opportunities, stronger performance might result from those additional learning opportunities rather than relational organization itself.

Different Outcome != Proposed Cause

What else could have produced the difference?

That question keeps the conclusion from becoming larger than the evidence.

Capability May Not Be One Number

Suppose one instructional group performs better.

What does better mean?

Perhaps students recall more immediately. Perhaps they retain more after a delay. Perhaps they discriminate meanings more accurately. Perhaps they transfer knowledge more successfully. Or perhaps they reach the same proficiency with fewer learning opportunities.

These outcomes are related, but they are not identical.

A student who learns something quickly but forgets it quickly presents a different learning profile from a student who requires more initial practice but later retains and transfers the knowledge effectively.

Speed != Capability

Capability != One Number

A more useful starting point may be a profile:

Capability Profile = {Accuracy, Retention, Application, Transfer, Adaptability}

This is an explanatory representation, not a validated measurement model.

Its purpose is to preserve potentially important distinctions until evidence tells us whether - and how - they should be combined.

From Definition to Investigation

The vocabulary experiment performs a limited but important function.

It shows that a conceptual question can become an empirical one.

What can the learner now do?

What observable differences can we measure?

Which instructional conditions contributed to those differences?

That progression prevents the discussion from remaining purely philosophical.

Claims about capability become scientifically useful when they generate observations that can be compared, challenged, replicated, and refined.

This brings us back to the Science of Progressive Learning.

Its purpose here is not to replace the scientific definition of learning. It is to investigate what becomes possible once learning has changed the learner.

The Next Scientific Problem

How much did the system change?

How should we measure the new capability?

Those questions are more difficult than asking whether a student produced a correct answer.

They require us to investigate learning across time, tasks, and conditions.

Perhaps some dimensions of capability will prove strongly related. Perhaps others will remain distinct.

Perhaps one instructional structure will improve retention without improving transfer. Another may improve speed while contributing little to adaptability.

Those possibilities must be determined by evidence.

The framework should generate questions before it generates conclusions.

What Can the Learner Now Do?

What does it mean to learn?

The question sounds the same.

But after following the argument, it no longer means quite the same thing to the reader.

A scientific definition helps us identify the phenomenon of learning.

For education, however, defining learning opens another question rather than closing the discussion.

Learning is not simply what information entered the learner. The deeper question is what the learner can now do because learning occurred.

That is where the Science of Progressive Learning begins its work.

Not by replacing the definition of learning.

Not by reducing learning to a grade.

Not by assuming that one successful performance reveals everything that changed.

Instead, it asks us to investigate the consequences:

What changed?

How durable is that change?

Where can it be used?

How far can it transfer?

What can the learner now do that could not be done before?

The scientific definition establishes the starting point.

The changed learner gives us the next frontier.