Feedback Guides Learning

A learner attempts a new movement and misses the target. A coach points out that the release happened too late. On the next attempt, the learner adjusts the timing and performs much better.

It looks as though the feedback worked. But suppose the same correction is needed on every attempt, or the improvement disappears when the coach stops speaking. The feedback clearly changed what the learner could do while it was available. Whether that change became independently available is another question.

The same question appears across many learning settings. A correct answer can be shown before someone understands why their answer was wrong. A live display can guide a movement without preparing the person to perform when the display disappears. A detailed comment can contain information that its recipient cannot yet interpret or use.

Feedback can guide learning by making consequences, errors and task-relevant discrepancies informative for subsequent adjustment. But receiving information is not the same as learning from it.

Feedback makes consequences informative

As explored in Repetition Is Not Practice, repeated action does not guarantee useful practice. Experience becomes more instructive when the learner can detect relevant relations among what was done, what happened and what might be changed.

Feedback can make some of those relations more available for evaluation.

Some feedback arises through the task itself. A person can see where a ball lands, hear whether a note is in tune, feel whether a load is balanced or observe whether a chosen strategy produced the intended result. This is often called intrinsic feedback: information available through acting and through the learner's own perceptual systems.

Other feedback is added by an external source. A teacher marks an answer, a coach describes a movement, a device displays speed or force, or a colleague explains how a response affected a conversation. This augmented feedback may supply information that was otherwise unavailable or make existing information easier to notice.

Neither form acts on the learner automatically.

Information can be available without being detected. It can be detected without being understood. It can be understood without revealing an effective adjustment. An appropriate adjustment can be identified without being successfully performed. Even a successful adjustment may remain tied to the immediate situation rather than becoming a durable change.

The sequence is conditional:

Information becomes available → it may be detected → it may be interpreted → it may support adjustment → the adjustment may contribute to learning.

No transition guarantees the next.

When it is detected and used, feedback changes the informational conditions under which subsequent actions are evaluated and selected. It does not insert a correct response into the learner. It can make one attempt more informative for the next.

Different feedback answers different questions

Feedback can contain several kinds of information.

Knowledge of results concerns the outcome achieved:

The ball landed left of the target.

The answer was incorrect.

The task took forty seconds.

The response received seven points out of ten.

This answers some version of: What happened?

Knowledge of performance concerns how the response was produced:

The release occurred too late.

The second movement began before the first was complete.

The explanation omitted the link between two claims.

Force increased too quickly.

This answers: How was the response performed?

Corrective information proposes what might be changed:

Begin the release earlier.

Complete the first movement before initiating the second.

Explain why the evidence supports the conclusion.

Increase force more gradually.

This answers: What could be attempted next?

These functions are related, but they are not interchangeable.

Knowing that a ball missed the target does not reveal which feature of the movement produced the miss. Knowing that the release occurred late identifies a performance characteristic, but it does not prove that changing this characteristic will solve the problem. A proposed correction adds direction, but the proposal may still be incomplete, inappropriate or beyond what the learner can currently enact.

The most useful information therefore depends on what is already available and what remains uncertain. Outcome information may add little when the result is obvious. Performance information may be valuable when a relevant movement cannot be seen or felt clearly. Corrective guidance may help when the learner cannot generate a plausible alternative.

No category is inherently superior. Its value depends on the informational problem it helps solve.

An error does not contain its correction

Error information indicates that something differs from a target, expectation or criterion. It does not necessarily tell us why.

A missed throw could result from direction, force, timing, stance, coordination or a changing environment. An incorrect answer could reflect missing knowledge, a misunderstood question, a calculation mistake or an unsuitable strategy. The same outcome can have several possible causes, and different causes can require different responses.

Several operations are hidden inside what is casually called correcting an error:

A discrepancy is detected.

The relevant part of the response is located.

A possible cause is identified.

An alternative is selected.

The learner attempts the alternative.

The result of that adjustment is evaluated.

Feedback may contribute at any one of these stages. It may verify that an answer was wrong, identify where a problem occurred, explain why the response failed, offer a hint, demonstrate an alternative or prescribe what to try next.

More information can help, but usefulness does not rise automatically with quantity. A detailed explanation may exceed what the learner can process. Several simultaneous corrections can make it unclear which adjustment matters most. Precise information can also be precisely focused on the wrong variable.

The relevant question is whether the information is understandable, task-relevant and usable by this learner at this point in learning.

This also distinguishes feedback from prediction error. As explored in Prediction Error Drives Updating, a prediction error is a discrepancy between what was expected and what occurred or was newly signalled. Feedback may provide information from which such a discrepancy is detected, but the concepts are not identical. A demonstration, an explanation and a social evaluation can all contain information that is not captured by a simple difference between prediction and outcome. Prediction errors can also arise through the consequences of action without anyone externally supplying feedback.

In formal learning models, prediction error is a model variable. It should not automatically be equated with a consciously felt error or with one biological mechanism.

Feedback can therefore reveal that adjustment is needed without revealing which adjustment will work.

When guidance helps—and when it creates reliance

One way feedback can affect performance is by guiding action.

A real-time signal can help a learner coordinate an unfamiliar movement. A teacher's correction can interrupt repeated use of an ineffective strategy. A measurement device can reveal a variable that the learner cannot perceive accurately. Immediate information can be especially valuable when errors are unsafe, costly or difficult to reverse.

Guidance is not inherently opposed to learning. Augmented information may be essential when task-generated information is unavailable, unreliable or too difficult for the learner to interpret.

The difficulty appears when performance becomes organized around a signal that will later be absent.

Imagine learning a balance task while watching a display that continuously shows every deviation from the target. The display may enable rapid corrections and excellent practice performance. But the learner may become skilled at following the display without becoming equally skilled at detecting and correcting deviations through information that remains available once the display is removed.

Performance may then deteriorate when the display disappears.

That decline does not show that nothing was learned. The person may perform better than before practice while also losing the temporary support that made practice performance especially strong. Removing the feedback also changes the performance environment. Dependence cannot be inferred from any decline alone; it must be evaluated against baseline performance, appropriate comparison conditions and later unsupported performance.

Guidance and learning can therefore coexist. Feedback may support the current response while also contributing to a capability that persists beyond it. The proportions cannot be read directly from practice performance.

Reliance on external information is not necessarily a defect. Many tasks are properly performed with continuing feedback. Drivers use speedometers, clinicians use monitoring equipment and pilots use flight instruments. Skilled performance in these settings includes using information that remains part of the target environment.

The independence question becomes important when the eventual task requires performance without that information, with different information or under conditions in which the signal may become unreliable.

As skills develop, what a learner can perceive and control may change. Skill Learning Changes Perception and Action explains this broader reorganization, while Automaticity Develops Gradually shows why the support needed early in learning may differ from what is needed later. Feedback is useful when it helps build capabilities suited to the eventual performance conditions—not merely when it makes the present attempt look successful.

Why feedback findings differ

Timing, frequency, specificity and relation to ongoing action all change the conditions under which feedback is used. Their effects also interact with the learner, the task and the outcome being measured. This is why research does not yield one universally superior schedule.

Immediate feedback can help someone correct an unfamiliar response before the same error is repeated. It may reduce uncertainty and prevent the learner from becoming stuck. A delay, however, can create time to evaluate the attempt, predict the result or generate a possible correction before receiving additional information.

The value of that delay depends on what the learner can do during it. If the learner has relevant knowledge and perceptual information, the interval may support self-evaluation. If those resources are missing, waiting may add difficulty without making the attempt more informative.

Frequency creates a similar interaction.

Frequent feedback can support complex performance and make subtle discrepancies visible. For a beginner, information after every attempt may be more useful than repeated action without understanding what is going wrong.

In other conditions, feedback after every attempt can reduce opportunities to evaluate task-generated information or generate independent corrections. A lower frequency may provide opportunities for self-assessment when the learner has enough information to evaluate the attempt. It can also withhold information that the learner still needs.

The evidence therefore supports neither "more feedback is better" nor "less feedback produces deeper learning."

Specificity also has two sides. "That was wrong" may be too vague to support adjustment. "Your timing was late" directs attention toward a more relevant feature. But a long list of every observable imperfection can exceed the learner's capacity to prioritize and act. Feedback is useful when it reduces relevant uncertainty, not when it maximizes the amount of information delivered.

Feedback can also be concurrent, arriving while an action is being performed, or terminal, arriving after the attempt has ended. Concurrent feedback can support rapid adjustment, but it may become part of the control system for the action. Terminal feedback allows the learner to complete the attempt and experience its consequences before receiving additional information. Their effects depend partly on what must remain possible when the added information is no longer present.

Some research also gives learners control over when they receive feedback. In motor-learning studies, learner-controlled feedback has sometimes benefited later retention and transfer. Choosing to request information can connect feedback to an active uncertainty, prediction or self-evaluation. But control alone does not guarantee effective use. Learners may request feedback too frequently, avoid it after poor attempts or seek confirmation rather than diagnostic information.

These variables are not independent levers with fixed effects. They help explain why a feedback condition that supports one learner, task or outcome may be unhelpful in another. What matters is what information the learner can use at the present stage and what capability the eventual task requires.

Feedback is not reward, reinforcement or prediction error

Feedback often overlaps with reward, reinforcement, pleasure and prediction error. A single event can perform several functions, but those functions still require separate evidence.

Feedback concerns information about a response, its performance, its consequences or its relation to a criterion.

A reward is a valued or wanted outcome. A reward can provide feedback, but it need not explain what produced success or what should be repeated.

Reinforcement refers to a process through which a consequence increases or strengthens relevant future behavior or learned responding under specified conditions. It cannot be established merely because praise, money or another desirable event followed an action.

Pleasure is an experience. Feedback can be useful without being pleasant, and a pleasant response does not by itself demonstrate reinforcement.

Praise after a successful attempt might indicate that a criterion was met, provide a valued social outcome and affect later behavior. Whether it actually performed each of these functions must be evaluated rather than assumed. Reward, Reinforcement and Pleasure develops these distinctions further.

Feedback from another person can also carry social meaning alongside its task content. A correction may affect confidence, perceived status or the learner's sense of being evaluated. Attention may shift from the task toward approval, embarrassment, defensiveness or comparison with others. These effects matter here because they can influence whether task information is noticed, trusted and used.

Perceived credibility can increase the likelihood that feedback will be accepted, but credibility does not guarantee accuracy. A confident or trusted source can still misidentify the cause of an error.

This is why "positive" and "negative" are insufficient descriptions of feedback quality. Praise can be pleasant but uninformative. Correction can be uncomfortable but useful. Either can direct attention toward the task or away from it.

The relevant issue is not only how the message feels, but what information it contains, what it makes salient and how it affects subsequent action.

How to tell whether feedback guided learning

One improved attempt cannot by itself establish durable learning. Different measurements answer different questions.

Acquisition concerns change during practice. Did performance become more accurate, efficient or reliable while practice continued?

Testing after augmented feedback is removed asks whether relevant performance remains possible without that support.

Retention asks whether some improvement remains after time has passed.

Transfer asks whether the developed capability supports performance under changed conditions.

These outcomes must be separated. Practice performance may combine developing capability with temporary guidance. Retention provides evidence that change persisted, but not that it will generalize. Transfer demonstrates some generalization, but its meaning depends on how the new condition differs from practice.

Retention also does not identify the process through which a change persisted. A later retention test provides behavioral evidence, not a complete account of Consolidation.

Research must then distinguish association from causation. Learners who receive more feedback may also receive more practice, better instruction or easier tasks. They may begin with greater skill, motivation or access to resources. If they later perform better, feedback is only one possible explanation.

As Correlation Is Not Causation explains, relevant comparisons are required. Researchers might compare feedback with no feedback, one feedback type with another, or learner-controlled feedback with a matched or yoked schedule imposed externally. Such designs can isolate particular differences, although matched schedules are not necessarily psychologically identical: chosen feedback may relate differently to the learner's uncertainty than the same schedule received passively.

The measured outcome also matters. An intervention can improve immediate accuracy without improving retention, or improve retention without transferring to a new task. Researchers must specify what changed, when it was measured and whether feedback remained available during the test.

A large meta-analysis illustrates this heterogeneity. Feedback interventions improved performance on average, yet more than one-third of the observed effects were negative. That result does not describe every kind of feedback or prove the same pattern for durable learning. It shows why the category feedback cannot be assigned one dependable effect.

Even when an intervention produces change, the mechanism remains a separate question. Feedback might alter attention, effort, strategy, confidence, error detection or several processes at once. A process changing alongside performance does not by itself show that it carried the effect.

The most defensible conclusion is therefore specific: particular feedback conditions can improve particular outcomes for particular learners and tasks. Other conditions can have little effect, support performance only while present or make performance worse.

Information creates an opportunity for updating

Return to the learner whose movement improved after a correction.

The improvement mattered. It showed that the feedback changed what the learner could do under those conditions. But it did not yet reveal whether the learner could detect the problem independently, reproduce the adjustment later or adapt it to a different situation.

For feedback to guide learning, relevant information must be detected, interpreted and connected to an action the learner can attempt. The consequences of that adjustment must then become informative in turn. If independent or generalized performance matters, persistence and transfer must be tested rather than assumed.

Feedback is therefore neither a message that automatically becomes knowledge nor a correction that automatically becomes skill.

It creates an opportunity for experience to become more informative—and for the consequences of one attempt to support a better-informed next attempt.