Behavior

Abstract

Behavior is the observable, measurable activity of an organism: what it does, rather than what it feels or thinks. The Medical Subject Headings vocabulary classifies it under behavior and behavior mechanisms, the family that groups actions with the processes that produce them. The behaviorist program made it an experimental science and built a durable account of how actions are learned through association and consequence, which the cognitive revolution embedded rather than discarded by showing that conditioning depends on what an organism predicts. Modern work spans four levels: the behaviorist laws of learning, the ethological analysis of why a species acts as it does, the computational theory of reinforcement learning, and the neurobiology of reward and extinction. This article covers what behavior is, its MeSH subtypes, conditioning and extinction, and computational accounts, with a worked example and three interactive demonstrations.

Keywords: behavior, conditioning, reinforcement, extinction, reinforcement learning

Behavior is the oldest subject matter of psychology and, for a long stretch of the twentieth century, the only one the field would officially admit. It has the singular advantage of being public: two observers can agree on whether a rat pressed a lever or a child shared a toy in a way they cannot agree on a private sensation. That objectivity is what let the study of behavior become an experimental science, and it is why the laws of learning worked out on behaving animals remain among the most robust findings psychology has produced.

This article treats behavior as the common object of several converging sciences rather than the property of any one school. It sets out the behaviorist program and the conditioning laws that are its lasting contribution, traces how the cognitive revolution reinterpreted rather than overturned those laws, and then widens the lens to the ethological question of why behavior takes the species-typical forms it does and to the computational and neural accounts that now specify the mechanism of learning from reward.

What Behavior Is

Behavior is the set of an organism's responses to its internal states and external environment: movement, action, and the physiological and verbal activity that can be observed and recorded. The definition is deliberately austere. It says nothing about consciousness, intention, or experience, not because those do not exist but because behavior is defined by its observability, and it is precisely that restriction that gave the concept its scientific traction. When John Watson announced that psychology should be the science of behavior, his central move was to make the observable response, rather than the private report, the unit of analysis (Watson, 1913).

A useful account must keep three things distinct. There is the eliciting condition, the stimulus or state that occasions an action; there is the behavior itself, the response that can be counted or timed; and there is the consequence, the change in the environment that the behavior produces and that in turn shapes its future probability. Much of the science of behavior is the study of how these three are related: how a neutral stimulus comes to occasion a response, and how a consequence makes a response more or less likely. The Medical Subject Headings vocabulary reflects this breadth by placing an unusually large number of specific behaviors beneath the general descriptor, from feeding and reproduction to risk-taking and the use of tobacco.

Types of Behavior

Behavior is the parent category in the Medical Subject Headings tree for a large family of narrower descriptors, each a specific form or domain of action that the vocabulary indexes beneath it under behavior and behavior mechanisms. These subtypes are organized by the kind of activity, the agent, or the functional goal, and they cut across one another rather than partitioning behavior cleanly: a single act of smoking is at once a health behavior, a drug-use behavior, and a habit. The grouping is an indexing convenience for cataloguing the literature, not a psychological taxonomy, and most of these categories are themselves active research areas. The direct MeSH children of Behavior are:

Accident Proneness.

A putative individual disposition to incur accidents at a higher rate than chance would predict.

Adolescent Behavior.

The characteristic actions and conduct of individuals in the adolescent years.

Animal Behavior.

The observable activity of non-human animals, the subject matter of ethology and comparative psychology.

Behavioral Symptoms.

Observable actions that serve as signs of an underlying psychological or medical condition.

Child Behavior.

The characteristic actions and conduct of children.

Psychological Codependency.

A relational pattern of excessive reliance on, and preoccupation with, another person's needs.

Communication.

The exchange of information between organisms through signals, language, or display.

Consumer Behavior.

The actions and decision processes of people acquiring and using goods and services.

Criminal Behavior.

Action that violates the criminal law, studied for its psychological antecedents.

Dangerous Behavior.

Action that poses a risk of harm to the self or to others.

Drinking Behavior.

Patterns of fluid and, in particular, alcohol consumption.

Drug-Seeking Behavior.

Action directed at obtaining a drug, central to the study of addiction.

Escape Reaction.

A response that removes an organism from an aversive or threatening situation.

Exploratory Behavior.

Investigation of a novel environment in the absence of an immediate goal.

Feeding Behavior.

The actions by which an organism obtains and ingests food.

Habits.

Automatic, context-cued actions that recur with little conscious deliberation.

Harm Reduction.

Strategies that reduce the negative consequences of a risky behavior rather than eliminating it.

Health Behavior.

Action that bears on physical well-being, such as exercise, diet, and screening.

Hoarding.

The persistent acquisition of, and difficulty discarding, possessions.

Human-Animal Interaction.

The behavioral and psychological exchanges between people and other animals.

Illness Behavior.

The ways people monitor, interpret, and act on bodily symptoms.

Imitative Behavior.

The reproduction of an observed action, a basic route of social learning.

Impulsive Behavior.

Action taken without adequate forethought or regard for consequences.

Information Seeking Behavior.

The active pursuit of information to reduce uncertainty or guide decision.

Psychological Inhibition.

The suppression or restraint of a response, impulse, or interfering process.

Marijuana Use.

Patterns of cannabis consumption and their behavioral correlates.

Motor Activity.

The bodily movement of an organism, from locomotion to fine action.

Multitasking Behavior.

The attempt to carry out more than one task concurrently.

Personal Satisfaction.

The subjective sense of contentment with one's circumstances or achievements.

Recreational Drug Use.

The use of psychoactive substances for non-medical, pleasurable ends.

Reproductive Behavior.

Action related to mating, and the rearing of offspring.

Risk Reduction Behavior.

Action undertaken to lower the probability of an adverse outcome.

Risk-Taking.

The deliberate undertaking of action whose outcome is uncertain and possibly harmful.

Sedentary Behavior.

Waking activity marked by low energy expenditure, such as prolonged sitting.

Self Stimulation.

The delivery by an organism of reinforcing stimulation to itself, classically to brain reward sites.

Sexual Behavior.

Action related to sexual activity and its social expression.

Smoking.

The inhalation of combusted tobacco or other substances, a paradigm habit behavior.

Social Behavior.

Action directed toward, or occurring in the presence of, other members of a species.

Spatial Behavior.

Action organized with respect to location, orientation, and navigation through space.

Stereotyped Behavior.

Repetitive, invariant action with no obvious function, seen across species and conditions.

Sucking Behavior.

The rhythmic oral action of infants, an early organized and measurable response.

Tobacco Use.

Patterns of tobacco consumption across its various forms.

The Behaviorist Program

The experimental science of behavior grew from two discoveries made independently around the turn of the twentieth century. Edward Thorndike, studying cats escaping from puzzle boxes, found that actions followed by a satisfying consequence became more likely and actions followed by an annoying one less likely, a regularity he called the law of effect (Thorndike, 1898). The law was radical because it explained the selection of behavior without appeal to insight or reasoning: consequences alone, acting on a population of responses, could shape what an animal did. Watson then drew the methodological conclusion, arguing that psychology should abandon introspection entirely and confine itself to the lawful relations between stimuli and responses (Watson, 1913).

B. F. Skinner gave the program its mature form. He distinguished respondent behavior, elicited by a stimulus, from operant behavior, emitted by the organism and then selected by its consequences, and he made the rate of operant responding the fundamental datum of the analysis (Skinner, 1963). The schedule on which reinforcement is delivered, Skinner showed, controls the pattern of responding with great precision: fixed and variable ratios and intervals each produce their own characteristic and highly reliable response profile. Late in his career he generalized the logic into a claim about causation across three levels, arguing that selection by consequences operates in phylogeny through natural selection, in ontogeny through operant conditioning, and in culture through the transmission of practices, a single Darwinian principle acting on three timescales (Skinner, 1981). The first demonstration lets the reader compare the response patterns that the four basic reinforcement schedules generate.

Conditioning and Extinction

Two forms of learning do most of the work in shaping behavior. In classical conditioning a neutral stimulus that reliably precedes a biologically significant one comes to evoke a response on its own; in operant conditioning a response that reliably produces a consequence changes in frequency. The modern understanding of the first was transformed by Robert Rescorla, who showed that temporal pairing is neither necessary nor sufficient for conditioning: what matters is whether the conditioned stimulus provides information, whether it predicts the unconditioned stimulus better than the background does (Rescorla, 1988). Conditioning, on this view, is not the stamping-in of a reflex but the learning of a predictive relationship, and the organism behaves as a detector of contingency rather than contiguity.

The quantitative core of this account is the idea that learning is driven by prediction error, the discrepancy between the outcome expected and the outcome received, so that associative strength grows by increments that shrink as the prediction improves and the error approaches zero. Extinction, the decline of a conditioned response when the predictive relationship is broken, turns out not to erase the original learning. Mark Bouton's work established that extinction is new, context-dependent learning layered over the original association, which is why an extinguished response renews when the animal is returned to the acquisition context, recovers spontaneously with time, and reinstates after a reminder (Bouton, 2004). The persistence of the original memory beneath the extinction has direct clinical force, because it explains the return of fear and of relapse after exposure-based treatment. The second demonstration traces the growth and decline of associative strength under the prediction-error rule across acquisition and extinction.

From Behavior to Cognition

Even at the height of behaviorism, findings accumulated that pure stimulus-response theory could not accommodate. Edward Tolman showed that rats allowed to explore a maze without reward nonetheless learned its layout, revealing their knowledge only when a reward was later introduced, which implied that they had acquired a cognitive map, an internal representation of the environment, in the absence of any reinforced response (Tolman, 1948). Learning, this latent learning showed, could occur without performance, so the two had to be distinguished, and behavior could not be the whole of what an organism acquired.

A second limit came from the behavior of animals in applied settings. Keller and Marian Breland, former students of Skinner who trained animals commercially, documented systematic failures in which conditioned responses drifted back toward innate, species-typical patterns, a raccoon rubbing coins together rather than depositing them, a phenomenon they named instinctive drift (Breland & Breland, 1961). The lesson was that reinforcement does not operate on an arbitrary, infinitely plastic organism: evolution has prepared each species to learn some associations readily and others with difficulty or not at all. Together, latent learning and instinctive drift showed that behavior is governed jointly by consequence and by representation, and by a biology that constrains what can be learned, a recognition that helped launch the cognitive revolution without discarding the conditioning laws it built upon.

The Ethological and Biological View

Where the behaviorists asked how behavior is learned, the ethologists asked why it takes the particular forms a species shows. Niko Tinbergen crystallized the discipline by distinguishing four complementary questions that a full explanation of any behavior must answer: its immediate causation, its development over the lifespan, its survival function, and its evolutionary history (Tinbergen, 1963). The framework is still the standard organizing scheme of the biological study of behavior because it keeps separate the mechanistic question of how a behavior is produced from the functional question of what it is for, two questions that are easily and fruitlessly confused.

The biological view has since been extended downward to mechanism and outward to context. Robert Sapolsky's synthesis frames any behavior as the joint product of causes operating on nested timescales, the neural activity of the preceding second, the hormonal state of the preceding hours, the developmental and experiential history of preceding years, and the evolutionary pressures of preceding millennia, so that asking whether a behavior is biological or environmental is a category error (Sapolsky, 2017). This integrative stance does not compete with the conditioning laws or the cognitive account; it situates them, specifying the levels at which each operates and insisting that a behavior is explained only when the contributions of all of them are in view.

Computational Accounts of Learning

The prediction-error idea at the heart of modern conditioning theory turned out to have a precise formal expression in the mathematics of reinforcement learning, and the convergence has been one of the most productive in the science of behavior. Yael Niv's synthesis set out how reinforcement learning models, which cast behavior as the attempt to maximize cumulative reward by updating value estimates in proportion to the error between predicted and received reward, map onto the neural systems that implement learning, with the phasic activity of midbrain dopamine neurons encoding precisely the reward-prediction-error signal the models require (Niv, 2009). The formal theory thereby unifies the behavioral law of effect, the quantitative prediction-error account of conditioning, and an identified neural mechanism within a single framework.

The reach of the reinforcement-learning framework now extends well beyond the laboratory. In economics, Richard Thaler's account of how real choice departs systematically from the rational-agent ideal has absorbed the behavioral evidence that people learn and decide through exactly the kind of error-driven, context-sensitive processes the conditioning literature describes (Thaler, 2016). In the clinic, the move from protocol-based therapy toward process-based care, set out by Steven Hayes and Stefan Hofmann, treats a disorder as a network of learned behavioral and cognitive processes to be targeted individually, applying the science of behavior change rather than a diagnostic label (Hayes & Hofmann, 2017). The third demonstration lets the reader watch a simple reinforcement-learning agent update the value of two actions from the rewards they deliver.

Worked Example: Prediction Error Builds Associative Strength

Consider how a conditioned response grows over repeated pairings under the prediction-error rule that underlies modern conditioning theory. On each trial the associative strength `V` of a cue changes in proportion to the error between the maximum strength the outcome can support, the asymptote `λ`, and the strength already accrued, so that the increment is `ΔV = k(λ − V)`, the formal rule Rescorla and Wagner introduced to explain why a surprising outcome drives more learning than an expected one, where the rate `k` combines the salience of the cue and the outcome (Rescorla & Wagner, 1972). Take an asymptote of `λ = 1.0` and a rate of `k = 0.3`, and start from no prior learning, `V = 0`. Each trial adds a share of the remaining distance to the asymptote.

The increments form a shrinking geometric series: each trial closes the same fraction, `k = 0.3`, of whatever distance to the asymptote remains, so the prediction error on trial 5 is `(1 − 0.3)^4 = 0.240`, exactly the value in the table. The learning curve therefore rises steeply at first and then flattens, approaching but never quite reaching the asymptote, the negatively accelerated shape that conditioning data reliably show. Reverse the rule by setting the asymptote to `λ = 0`, as when the cue is now presented without the outcome, and the same equation drives the strength back down by the same shrinking fraction each trial: this is extinction as the prediction-error account models it. The arithmetic is deliberately simple, but it captures the central claim that behavior is learned in proportion to how much it surprises the organism, and that learning slows precisely as surprise is used up.

Discussion

The science of behavior has moved from a program that refused to look inside the organism to one that specifies, at several levels at once, the mechanisms that generate action. The behaviorists were right that behavior is lawful and that consequences select it, and the conditioning laws they established have survived every subsequent reinterpretation (Thorndike, 1898; Skinner, 1963). What they were wrong about was the sufficiency of the stimulus-response framework: latent learning showed that organisms acquire representations they do not immediately use, and instinctive drift showed that the slate is not blank (Tolman, 1948; Breland & Breland, 1961). The cognitive and biological accounts that followed did not refute conditioning but explained why it works, recasting it as the learning of predictive structure in a brain prepared by evolution to learn some things more readily than others.

The deepest integration of the field is the recognition that a single idea, learning driven by prediction error, connects the behavioral, the computational, and the neural (Rescorla, 1988; Niv, 2009). The same error signal that the Rescorla account inferred from the behavior of conditioned animals is computed by reinforcement-learning algorithms and carried by the phasic firing of dopamine neurons, a rare case in which a behavioral law, a formal model, and a physiological mechanism line up. The open problems lie in extending this unity: in explaining how extinction's context-dependence is implemented so that treatment relapse can be prevented (Bouton, 2004), and in scaling the reinforcement-learning account from two levers to the structured, long-horizon behavior that economics and the clinic must explain (Thaler, 2016; Hayes & Hofmann, 2017).

Cognitive Implications

Behavior occupies a peculiar place in cognitive psychology: it is at once the discipline's historical starting point and its ultimate dependent measure. Almost every cognitive theory is tested, in the end, by what people and animals do, so the conceptual care the behaviorists brought to defining and measuring responses remains indispensable even to researchers who study the representations the behaviorists denied. The lesson that performance and competence must be distinguished, forced by Tolman's latent learning, recurs throughout the field: a failure to act is not a failure to know, and inferring the second from the first is a standing hazard of behavioral measurement (Tolman, 1948).

The prediction-error principle has proven even more consequential for cognition than for behavior narrowly conceived. The idea that a system learns in proportion to how much an outcome violates its expectation has become a general template for theories of perception, attention, and decision, where the brain is modeled as continually generating predictions and updating them from error (Rescorla, 1988; Niv, 2009). What began as an account of why a dog salivates to a bell is now one of the organizing ideas of the cognitive and neural sciences, which is a measure of how much the rigorous study of behavior gave to the study of the mind it was once set against.

Current Directions

The most active current work pushes the reinforcement-learning account in two directions at once. One is downward, toward mechanism: having established that phasic dopamine encodes a reward-prediction error, researchers are now mapping how richer quantities, the distribution of possible rewards rather than their average, and the value of internally generated states, are represented and updated, extending the formal theory Niv synthesized toward the structured behavior of real organisms (Niv, 2009). The persistence problem that Bouton identified has become a target in its own right, with work synthesizing the behavioral and neurobiological mechanisms of Pavlovian and instrumental extinction aimed squarely at preventing the renewal and reinstatement of extinguished fear and drug-seeking after treatment (Bouton, 2004; Bouton et al., 2021).

The second direction is outward, into application. Behavioral economics has moved from cataloguing departures from rationality to designing the choice environments, the defaults and framings, that work with rather than against how people actually learn and decide, a program that treats behavior change as an engineering problem informed by the learning sciences (Thaler, 2016). In clinical science the process-based turn reframes psychotherapy as the targeted modification of specific learned processes, selected for the individual rather than dictated by a diagnosis, which brings the conditioning and cognitive accounts of behavior directly to bear on treatment (Hayes & Hofmann, 2017). Both lines share the ambition of carrying the laboratory science of behavior into the prediction and guidance of behavior in the world.

Common Misconceptions

Behaviorism claimed that thoughts and feelings do not exist.

No. The methodological claim was that psychology should study observable behavior because private events could not be measured intersubjectively, and Skinner's radical behaviorism explicitly treated private events as behavior too. The position was about what could be studied scientifically, not a denial that inner life exists (Watson, 1913; Skinner, 1963).

Classical conditioning is just the pairing of two stimuli in time.

No. Rescorla showed that mere temporal pairing neither guarantees nor is required for conditioning; what the organism learns is whether the cue predicts the outcome better than the background does. Conditioning is the detection of a contingency, not of contiguity (Rescorla, 1988).

Extinction erases what was learned.

No. The extinguished response returns with a change of context, the passage of time, or a reminder of the outcome, which shows the original association is still intact. Extinction is new learning layered over the old, not its deletion (Bouton, 2004).

Any behavior can be conditioned into any animal with enough reinforcement.

No. Species-typical dispositions constrain what is readily learned: conditioned responses drift back toward innate patterns, and some associations are acquired far more easily than others. Reinforcement acts on an organism prepared by evolution, not on a blank slate (Breland & Breland, 1961).

Glossary

Associative strength.

The learned strength of the link between a cue and an outcome, which grows with reinforced pairings toward an asymptote set by the outcome.

Behavior.

The observable, measurable activity of an organism in response to internal states and the environment; the subject matter of the science of behavior.

Behaviorism.

The program, announced by Watson, that psychology should be an objective science of observable behavior rather than of introspected experience.

Classical conditioning.

Learning in which a neutral stimulus that predicts a biologically significant one comes to evoke a response on its own.

Cognitive map.

An internal representation of the spatial layout of the environment, inferred by Tolman from latent learning in mazes.

Ethology.

The biological study of behavior in its natural context, organized around Tinbergen's four questions of causation, development, function, and evolution.

Extinction.

The decline of a conditioned response when the cue no longer predicts the outcome; new, context-dependent learning rather than erasure of the original association.

Instinctive drift.

The tendency of a conditioned response to drift back toward an innate, species-typical pattern, documented by the Brelands.

Law of effect.

Thorndike's principle that actions followed by a satisfying consequence become more likely and those followed by an annoying one less likely.

Operant conditioning.

Learning in which the frequency of an emitted response is changed by its consequences; Skinner's central paradigm.

Prediction error.

The discrepancy between the outcome expected and the outcome received; the quantity that drives learning in both conditioning theory and reinforcement learning.

Radical behaviorism.

Skinner's version of behaviorism, which treats private events as behavior subject to the same principles rather than denying their existence.

Reinforcement learning.

The computational framework in which behavior maximizes cumulative reward by updating value estimates in proportion to prediction error.

Reinforcement.

A consequence that increases the future probability of the response it follows; the engine of operant conditioning.

Renewal effect.

The return of an extinguished response when the organism is removed from the context in which extinction took place.

Rescorla-Wagner model.

The influential account in which associative strength changes by an increment proportional to the prediction error between the asymptote and current strength.

Shaping.

The operant technique of reinforcing successive approximations to a target response that the organism would rarely emit on its own.

Tinbergen's four questions.

The four complementary levels at which any behavior must be explained: its causation, development, function, and evolutionary history.

Key Researchers

Mark E. Bouton

(University of Vermont, University Distinguished Professor Emeritus). Experimental psychologist whose work on context, renewal, and spontaneous recovery established that extinction is new learning rather than the erasure of the original association. Faculty · Scholar

Steven C. Hayes

(University of Nevada, Reno, Foundation Professor Emeritus). Clinical psychologist who originated Acceptance and Commitment Therapy and the process-based approach to behavior change grounded in the contextual analysis of learned behavior. ORCID · Faculty · Wikipedia

Yael Niv

(Princeton University). Computational neuroscientist whose work on reinforcement learning links the formal theory of reward-guided behavior to the dopamine prediction-error signal in the brain. ORCID · Scholar · Wikipedia

Robert M. Sapolsky

(Stanford University). Neuroendocrinologist and primatologist whose integrative account explains behavior as the joint product of causes acting across neural, hormonal, developmental, and evolutionary timescales. Faculty · Wikipedia

B. F. Skinner

(Harvard University; 1904–1990). Founder of radical behaviorism and the experimental analysis of operant behavior, who framed selection by consequences as a principle acting across evolution, learning, and culture. Wikipedia · Wikidata

Edward L. Thorndike

(Teachers College, Columbia University; 1874–1949). Formulated the law of effect from puzzle-box studies of animal learning, the empirical root of instrumental conditioning. Wikipedia · Wikidata

Nikolaas Tinbergen

(University of Oxford; 1907–1988). Nobel-laureate ethologist whose four questions remain the standard framework for the biological explanation of behavior. Wikipedia · Wikidata

John B. Watson

(Johns Hopkins University; 1878–1958). Author of the behaviorist manifesto that redefined psychology as the prediction and control of observable behavior. Wikipedia · Wikidata

Frequently Asked Questions

What is behavior in psychology?

Behavior is the observable, measurable activity of an organism: its movements, actions, and recordable responses to internal states and the environment. Psychology defines it by its observability, which is what allowed the study of action to become an experimental science with lawful and replicable findings.

What is the difference between classical and operant conditioning?

Classical conditioning builds an association between two stimuli, so that a cue predicting an important event comes to evoke a response. Operant conditioning changes the frequency of an emitted response according to its consequences. The first is about what predicts what; the second is about what an action produces.

What is the law of effect?

It is Thorndike's principle that an action followed by a satisfying consequence becomes more likely to recur, while one followed by an annoying consequence becomes less likely. It explained the selection of behavior by its outcomes without appeal to reasoning or insight (Thorndike, 1898).

Does extinction erase a learned behavior?

No. An extinguished response returns when the context changes, time passes, or a reminder is given, which shows the original learning is still present. Extinction is new, context-dependent learning layered over the original association rather than its deletion (Bouton, 2004).

Can any animal be conditioned to do anything?

No. Species-typical, evolved dispositions constrain what each animal readily learns. Conditioned responses can drift back toward innate patterns, and some associations are acquired much more easily than others, so reinforcement acts on a prepared organism rather than a blank slate (Breland & Breland, 1961).

What is a prediction error in learning?

It is the gap between the outcome an organism expects and the outcome it receives. Learning is driven by this error: associative strength grows by increments that shrink as the prediction improves, which is why learning curves rise steeply and then flatten (Rescorla, 1988).

How is behavior related to reinforcement learning in the brain?

Reinforcement learning models cast behavior as the updating of value estimates in proportion to reward prediction error, and the phasic firing of midbrain dopamine neurons encodes precisely that error signal, uniting the behavioral law of effect, the formal model, and an identified neural mechanism (Niv, 2009).

Did behaviorists deny that the mind exists?

No. Their claim was methodological: that psychology should study observable behavior because private events could not be measured objectively. Skinner's radical behaviorism treated private events as behavior subject to the same principles, not as nonexistent (Watson, 1913; Skinner, 1963).

References