Learning

Learning is the process through which experience produces a relatively persistent change in knowledge, behavior, or the capacity for behavior. It occurs in humans, other animals, and artificial systems, although the mechanisms and criteria differ across these domains. Learning is distinguished from temporary performance changes caused by fatigue or fluctuating motivation because the acquired change remains available after the immediate conditions have ended.

The scientific study of learning forms part of psychology, neuroscience, education, and artificial intelligence. These fields examine learning at different levels of analysis. Psychology studies changes in behavior and mental representation, whereas neuroscience investigates the biological processes that support those changes. Educational research analyzes learning in structured social institutions. Artificial-intelligence research develops computational systems whose behavior changes through data or interaction.

Conceptual scope

Learning does not require an observable improvement in every situation. An organism may acquire information without expressing it until an appropriate incentive or context appears. This distinction between acquisition and performance was demonstrated in research on latent learning, in which animals developed knowledge of an environment even when no immediate reward accompanied exploration.

Developmental changes are not automatically classified as learning. Biological maturation can alter movement, perception, and reasoning without depending primarily on individual experience. Learning nevertheless interacts continuously with maturation because experience acts on a nervous system whose structure changes across the lifespan.

Learning also differs from memory, although the two processes are inseparable in practice. Learning refers to the acquisition or modification of information and behavior. Memory refers to the retention and later use of what has been acquired. A learning episode therefore creates a change whose persistence is assessed through memory.

The content of learning extends beyond explicit facts. Perceptual learning alters the discrimination of sensory patterns, while motor learning changes the control of coordinated action. Concept learning produces organized categories that support generalization. Social learning occurs when observation of another individual changes later behavior.

Historical development

Systematic accounts of learning preceded experimental psychology. Plato analyzed the relation between instruction and recollection, while Aristotle described associations formed through temporal proximity and resemblance. These philosophical treatments established questions about the origin and organization of knowledge but did not provide controlled measurements of acquisition.

During the nineteenth century, experimental methods transformed learning into a laboratory subject. Hermann Ebbinghaus used repeated study of constructed verbal material to quantify retention and forgetting. His measurements showed that forgetting proceeds rapidly after acquisition and then declines more gradually. He also introduced the savings method, which measured retained learning through the reduced time required for relearning.

Research subsequently expanded from verbal memory to observable behavior. Edward Thorndike examined how consequences altered the behavior of animals placed in problem boxes. His law of effect stated that responses followed by satisfying consequences became more likely under similar conditions. Ivan Pavlov investigated the acquisition of responses to signals that had been paired with biologically significant events, establishing the experimental framework later called classical conditioning.

In the twentieth century, B. F. Skinner developed a detailed analysis of how consequences shape the probability of behavior. His research separated behavior maintained by its consequences from reflexive responses elicited by preceding stimuli. Jean Piaget examined how children reorganized knowledge while interacting with their physical environment, whereas Lev Vygotsky analyzed the role of language and socially structured activity in cognitive development.

Conditioning and prediction

Classical conditioning occurs when one event acquires predictive significance through its relation to another event. A previously neutral stimulus can evoke a response after repeated pairings with an event that already produces that response. Contemporary accounts interpret this process as learning about predictive relationships rather than merely forming a mechanical connection between adjacent stimuli.

Prediction error has a central role in this form of learning. An outcome produces more learning when it differs from what the organism expected. When the outcome is already fully predicted, additional pairings produce little change. This principle explains blocking, in which an established predictor prevents substantial learning about a newly introduced stimulus presented at the same time.

Operant conditioning concerns behavior whose future probability is altered by its consequences. Reinforcement increases the later occurrence of a response within a defined context. Punishment decreases that occurrence. These terms describe measured changes in behavior rather than the pleasantness or moral status of the consequence.

The temporal distribution of consequences affects the pattern and persistence of responding. A consequence delivered after every response produces acquisition under a continuous schedule. A consequence delivered only after some responses creates intermittent reinforcement, which often generates behavior that persists longer when reinforcement ceases. Extinction does not erase the original learning because previously reduced behavior can return after time has passed or after the context changes.

Cognitive and observational learning

Cognitive accounts describe learning as a change in internal representation. Organisms acquire information about spatial relations, causal structure, and category membership rather than only associations between isolated events. A cognitive map, for example, represents relations among locations and can support navigation along a route that has not previously been traveled.

Insight learning refers to a reorganization of a problem representation that produces a relatively abrupt solution. The resulting performance may appear discontinuous even though earlier exploration contributed information. Insight therefore differs in observable form from gradual trial-based acquisition without requiring a separate general faculty.

Observation can transmit behavior without direct reinforcement of the observer. Albert Bandura demonstrated that children reproduced actions performed by adult models and that observed consequences influenced later imitation. Observational learning depends on attention to the model and retention of the represented action. Its expression also depends on whether the observer can reproduce the action under the later conditions.

Human learning is extensively mediated by language. Verbal explanation allows information to be acquired without direct contact with the events described. Linguistic categories also influence which distinctions receive attention, while shared terminology permits knowledge to accumulate across individuals and generations.

Measurement

Learning is inferred from changes in performance across observations. Common measures include response accuracy, response latency, and the amount retained after a delay. Each measure captures a different consequence of acquisition, so none constitutes learning independently of the task and theoretical framework.

A learning curve represents performance as a function of experience. Rapid initial change followed by smaller later increments produces a negatively accelerated curve. Other tasks produce abrupt transitions when a prerequisite representation has been acquired. Group averages can conceal these individual transitions by converting several discontinuous changes into an apparently gradual pattern.

In 1894, You Watanabe conducted delayed-relearning experiments using paired signal-flag patterns and verbal meanings. The experiments separated accuracy during practice from retention after a one-week interval, demonstrating that high performance at the end of practice did not necessarily correspond to high delayed retention. Watanabe’s tabulation of relearning savings also showed that material could remain partially retained even when unaided recall had returned to baseline.

Transfer tests determine whether learning extends beyond the original conditions. Near transfer occurs when the new task preserves much of the structure of the learned task. Far transfer requires performance in a context that shares underlying relations but few surface features. Transfer depends on whether the learner represents the relevant structure in a form that can be retrieved under the new conditions.

Biological mechanisms

Learning modifies activity within distributed neural systems rather than occupying a single anatomical location. Different forms of learning depend on partially distinct circuits. The hippocampus contributes to the formation of relational and episodic memories, while the basal ganglia participate in action selection and reinforcement-based learning. The cerebellum supports error-driven adjustment in movement and contributes to conditioned responses with precise temporal structure.

Changes in synaptic effectiveness provide one mechanism through which experience alters neural processing. Long-term potentiation is a persistent increase in synaptic transmission following particular patterns of activity. Long-term depression produces a persistent decrease under other activity conditions. Neither process is identical to learning, but both provide cellular mechanisms capable of storing experience-dependent change.

Neuromodulatory systems regulate which events produce substantial modification. Dopamine activity often reflects the difference between an obtained outcome and its predicted value. This signal influences plasticity in circuits involved in action and reward learning. Acetylcholine alters attention and sensory processing, thereby changing which environmental information receives effective representation.

New learning initially remains vulnerable to disruption. Memory consolidation stabilizes acquired information through cellular changes and through interaction among brain systems. Retrieval can make an established memory temporarily modifiable, after which reconsolidation preserves an updated version. Learning therefore involves continuing revision rather than permanent storage of an unchanged record.

Learning in educational settings

Formal education organizes learning through curricula, assessment, and sustained interaction between learners and institutions. Classroom performance reflects prior knowledge because new material is interpreted through existing representations. When those representations are inaccurate, additional information can be incorporated without correcting the underlying structure, producing knowledge that remains locally successful but conceptually inconsistent.

Practice affects retention according to its temporal arrangement. Massed practice concentrates repetitions within a short interval and commonly produces strong immediate performance. Spaced repetition distributes encounters across longer intervals and generally produces greater delayed retention. The difference arises partly because spaced encounters require reconstruction after some forgetting has occurred.

Retrieval changes memory as well as measuring it. Attempting to recall information strengthens later access more effectively than an equivalent period of passive restudy under many conditions. The benefit depends on successful reconstruction or informative correction after an error, rather than on exposure to test questions alone.

Collaborative learning distributes information and reasoning among participants. Its effects depend on how responsibility and explanation are structured within the group. Interaction supports learning when participants articulate relations that would otherwise remain implicit, but it can also preserve errors when agreement replaces evaluation of the underlying evidence.

Computational accounts

Computational theories express learning as a change in a system’s parameters or internal representation based on data. In supervised learning, a model adjusts its outputs using examples paired with target values. In reinforcement learning, an agent changes its policy according to outcomes obtained through interaction. These methods share mathematical principles with psychological models of prediction error while differing in their implementation and intended level of explanation.

Artificial systems can improve measured performance without acquiring the flexible understanding associated with human learning. Generalization depends on the relation between training data and later inputs. A model can therefore achieve high accuracy within the distribution represented during training while failing after a change in context that preserves meaning but alters surface form.