What Is Learning?

A person practices a task and gets better at it.

We naturally say that they learned.

Later, they are exhausted, distracted or working under different conditions, and their performance falls. Did the learning disappear?

Now consider the reverse. Someone performs better because the instructions become clearer, stronger cues are available or the task becomes easier. Did they necessarily learn more?

These cases expose one of the foundational problems in learning science:

Observable behaviour is evidence about learning, but it is not identical to learning.

Learning concerns change attributable to experience. Performance concerns what an organism does under the conditions present at a particular moment.

The two are inseparable in practice, but they are not interchangeable.

What scientists mean by learning

There is no single definition of learning that every scientific tradition uses in exactly the same way.

Some researchers define learning primarily through systematic changes in behaviour resulting from experience. Others emphasize changes in the organism that alter later capacities, knowledge or performance. These approaches place the explanatory boundary differently, but they converge on an important problem: a change in behaviour at one moment does not necessarily tell us whether learning occurred.

For the purposes of this Library, learning will refer broadly to experience-dependent change that alters an organism's later capacities, knowledge, expectations or patterns of responding beyond a merely transient fluctuation in performance.

Several distinctions follow from this.

Learning must be related in some meaningful way to prior experience. It need not appear immediately in behaviour. It need not consist of explicit knowledge. And it need not last forever.

At the same time, not every behavioural change is learning.

Performance can change because of fatigue, motivation, temporary physiological state, injury, medication, maturation, changing opportunity or differences in the environment. A person may behave differently even when little or no new learning has occurred.

The scientific problem is therefore not simply to detect change.

It is to determine what changed, what contributed to that change, and what the observed behaviour actually tells us about learning.

Learning and performance are not the same thing

Prior experience can produce changes that affect what an organism can later discriminate, expect, know or do.

But those changes are only part of what determines current behaviour.

Performance also depends on the conditions under which behaviour occurs.

Conceptually:

PRIOR EXPERIENCE

↓

EXPERIENCE-DEPENDENT CHANGE

↓

ALTERED CAPACITY / KNOWLEDGE / EXPECTATION

↘ OBSERVED PERFORMANCE ↗

CURRENT CONDITIONS

state · motivation · opportunity · context · task demands

This is a conceptual distinction, not a complete causal model. Learning and current performance can themselves reflect multiple interacting processes.

The important point is simpler:

Performance is influenced by prior learning, but it is not uniquely determined by prior learning.

A person can retain a learned capability and still perform poorly when exhausted. A learned response may not occur because the relevant opportunity is absent. A change in context can alter which cues are available. Motivation can affect whether and how an existing capability is expressed.

This means that current behaviour cannot be interpreted solely as a readout of what has been learned.

Performance can improve or deteriorate without equivalent changes in learning

Suppose someone performs exceptionally well during practice.

That performance tells us something real: it tells us what they were able to do under those conditions.

It does not automatically tell us how much learning occurred.

Performance can improve because assistance is available, cues are unusually informative, task demands are reduced or motivation temporarily increases. Remove those conditions and some of the improvement may disappear.

Research demonstrates the problem from the opposite direction as well. Under some experimental conditions, practice can look worse while later retention is stronger. Conditions that improve immediate performance do not always produce stronger evidence of retained learning.

This does not mean that difficult practice is inherently better.

It means that immediate performance and learning can diverge.

Better performance does not necessarily mean more learning, and worse performance does not necessarily mean learning has disappeared.

A decline in performance establishes that behaviour changed under those conditions. By itself, it does not tell us why.

Fatigue may have increased. Context may have changed. Relevant cues may be missing. Competing responses may have become stronger.

Sometimes prior learning really has been modified or lost in ways that matter. But a single performance cannot distinguish that possibility from every alternative.

This is why the learning–performance distinction matters.

Learning can precede its expression

The distinction becomes especially clear when experience produces changes whose behavioural consequences appear only later.

Classic research on latent learning helped establish this possibility. Experience can produce changes whose consequences become evident when later conditions make them relevant, even when the behavioural pattern that eventually reveals those changes was not apparent during initial experience.

The historical experiments do not establish one exclusive theory of what was learned or how it was represented. Their broader importance is more modest and more durable:

Learning can precede the conditions that reveal it.

That conclusion blocks a common inference.

If behaviour does not change immediately, we cannot automatically conclude that no learning occurred.

But the opposite inference would be equally mistaken.

We cannot simply declare that invisible learning must have occurred whenever behaviour fails to change.

Learning still requires evidence

Separating learning from performance does not make learning immune to evidence.

Failure to observe a behavioural change in one situation may be inconclusive. It does not prove that no learning occurred.

But it also does not prove that hidden learning occurred.

Evidence may emerge through later retention, changed behaviour under different conditions, transfer, altered choices or expectations, or convergence across multiple measures.

None of these provides a universal test.

Retention can strengthen the case that a change extends beyond the temporary conditions present during acquisition. Transfer can show that the consequences of experience extend beyond one narrow task arrangement. But learning can also be context-specific, so failure to transfer does not automatically demonstrate absence of learning.

The broader principle is what matters:

Claims about learning depend on patterns of evidence that support experience-dependent change while distinguishing it, where possible, from temporary influences on performance.

A single performance is evidence about learning, not learning itself.

This distinction does not weaken behavioural evidence. It tells us how carefully that evidence must be interpreted.

Learning is change attributable to experience

Calling something learning does more than describe a difference between two moments.

It attributes some part of that difference to prior experience.

Consider two statements:

Experience occurred. Later, behaviour changed.

And:

Experience contributed to the later change.

They are not equivalent.

Behaviour may change because of maturation, temporary physiological shifts, injury, medication, altered motivation, environmental constraints or other causes. The fact that an experience happened before a behavioural change does not establish that it produced the change.

As established in Correlation, Prediction, Causation and Mechanism, temporal sequence alone is not enough to establish causation.

A learning claim therefore carries an explanatory commitment: the experience contributed to the change.

This also clarifies why exposure and repetition should not be confused with learning.

Encountering information creates an opportunity for experience-dependent change. Repetition can create repeated opportunities for such change.

But their occurrence alone does not establish that learning happened.

Someone can encounter the same information repeatedly without acquiring what we would expect them to know or do afterward.

Learning concerns the change produced through experience, not merely the presence of experience itself.

“Relatively enduring” does not mean permanent

Learning has often been described as a relatively permanent change in behaviour or behavioural potential resulting from experience.

The formulation captures an important intuition: learning should not simply be another name for every fleeting fluctuation in performance.

But permanence is too strong if interpreted literally.

Learned changes can weaken in expression. They can depend on context. Later experience can modify earlier learning. Retrieval can become difficult. New learning can interfere with what was acquired before.

A useful distinction is therefore whether the evidence supports something more than a merely transient performance fluctuation.

Relatively enduring does not mean permanent.

There is also no universal amount of time after which a change suddenly qualifies as learning. Different learning processes operate across different timescales, and the evidence required to establish them depends partly on the phenomenon being studied.

Nor does short duration automatically place a process outside learning. Habituation and sensitization, for example, are widely recognized forms of non-associative learning and can vary substantially in duration.

Persistence matters as evidence.

It is not a universal stopwatch for deciding what counts as learning.

Learning is broader than conscious knowledge

Learning is often imagined as conscious acquisition:

I understand the rule.

I know the answer.

I can explain what changed.

But experience can alter discrimination, expectations, response tendencies or skilled performance without producing a complete conscious account of what has been acquired.

Research on implicit and statistical learning demonstrates that explicit verbal awareness is not required for every form of learning, although determining exactly what someone is or is not aware of can itself be methodologically difficult.

The appropriate conclusion is therefore limited:

Learning does not always require explicit awareness or the ability to verbalize what has been learned.

That does not make consciousness irrelevant.

It means that explicit report is one possible expression of learning, not its universal definition.

Learning is not one universal mechanism

Once learning is defined broadly, another mistake becomes possible: treating everything called learning as the product of one underlying process.

Learning science instead investigates multiple forms of experience-dependent change.

Associative learning examines how relationships among events, actions and consequences can be acquired. Habituation and sensitization concern forms of non-associative learning. Perceptual learning examines experience-dependent changes in discrimination and perception. Motor and skill learning concern changes in coordinated perception and action. Other research traditions investigate statistical regularities, explicit knowledge and additional forms of acquired change.

These should not be assumed to reduce to one mechanism simply because they share the word learning.

Learning is a broad category of experience-dependent change, not one universal mechanism.

Nor should learning be reduced to one level of explanation.

It is not identical to observable behaviour, because behaviour depends on more than learning.

It is not identical to explicit cognition, because learning need not be consciously reportable.

And it is not identical to neural plasticity.

Learning involves biological change, but neural plasticity and learning are not synonymous. Biological systems can change for reasons other than learning, and detecting a neural change does not by itself tell us what was learned or what behavioural consequence followed.

Different questions about learning therefore require different levels of explanation.

How skill learning changes perception and action is developed separately in Skill Learning Changes Perception and Action.

Learning, memory and performance are related but distinct

Learning also needs to be distinguished from memory.

The concepts overlap because the consequences of experience must in some sense remain available if they are to influence later behaviour. But they do not describe exactly the same problem.

Learning concerns change attributable to experience.

Performance concerns what is expressed under particular conditions.

Memory concerns processes through which information or effects of experience persist, become available or are reconstructed later.

These are broad distinctions, not complete theories.

A failure to retrieve something does not necessarily mean that no learning ever occurred. Successfully performing a learned action is not the same as acquiring it. And later remembering involves processes that cannot simply be identified with the original learning event.

Those questions are developed downstream in Remembering Reconstructs the Past and Consolidation Stabilises Learning, while Sleep and the Preservation of Adaptation examines one of the biological conditions relevant to preserving acquired change.

Development introduces another layer of complexity because the meaning and causes of change differ across developmental timescales. That relationship is developed in Development Changes the Meaning of Change.

What is learning?

We can now return to the observation with which we began:

A person's behaviour changed.

Was that learning?

The behavioural change matters, but it does not answer the question by itself.

We need to know whether the evidence supports change attributable to experience rather than only a difference in current performance conditions.

Someone performs exceptionally well immediately after instruction. That may provide evidence of learning, but it does not by itself reveal how the change will persist or appear elsewhere.

Someone performs poorly later. That may reflect modification of prior learning, but it may instead reflect fatigue, context, motivation, opportunity or other conditions affecting expression.

Someone shows no immediate change. That does not prove that no learning occurred. But neither does it permit us to assume that learning occurred invisibly.

This is the central distinction:

A single performance is evidence about learning, not learning itself.

Learning will therefore refer broadly to experience-dependent change that alters what an organism can know, expect or do, while its observable expression remains dependent on the conditions under which behaviour occurs.

That definition gives us a foundation.

It does not tell us which learning process produced the change.

And that is the next question.

Some experiences teach organisms about relationships between events, cues, actions and consequences.

Understanding that class of processes requires a more specific concept: Associative Learning.