Behavior and Behavior Mechanisms

Abstract

Behavior is the observable activity of an organism, and behavior mechanisms are the processes that produce, modify, and regulate it. For cognitive psychology the topic carries a double significance: behavior was the discipline's original object of study, the only thing a strict behaviorism would admit as evidence, and it is now treated as a rich signal about the internal computations that generate it. This article traces the principal mechanisms by which behavior is shaped: associative and operant learning, the biological constraints that make some associations easy and others nearly impossible, the purposive and social-cognitive accounts that reinstated internal representations and beliefs, and the reward-prediction-error system that gives reinforcement learning a neural substrate. Across these literatures behavior emerges not as a residue to be explained away but as the organizing fact that every level of analysis must ultimately answer to.

Keywords: behavior, operant conditioning, reinforcement learning, reward prediction error, biological preparedness

Behavior and behavior mechanisms, taken together, denote both what an organism does and the set of processes that determine what it does. The scientific indexing vocabulary places the pairing at the head of its classification of behavioral phenomena, as the root beneath which learning, motivation, emotion, personality, and social conduct are all arranged. That placement reflects the pairing's scope rather than a theory: behavior is the common currency into which every psychological process eventually converts, because an internal state is studied, in the end, by what it makes an organism do. Watson (1913) made this currency the whole of psychology, declaring that the discipline should concern itself exclusively with observable behavior and its environmental determinants. The century since has complicated that program without abandoning its premise, and the mechanisms surveyed here are the successive answers to a single question: what turns a stimulus, a history, and a nervous system into an action?

What Behavior and Behavior Mechanisms Is

Behavior is any activity of an organism that can be observed and, in principle, measured: a movement, a choice, an utterance, a physiological change recruited in the service of action. Behavior mechanisms are the processes that stand behind that activity and determine its form, its timing, and its change with experience. The distinction matters because the two can be studied at different levels. One can describe a behavior exhaustively, as the number of lever presses per minute, without yet knowing the mechanism, the reinforcement history or neural computation that produced that rate. Much of the history of psychology is a contest over how far one must go beyond the behavior itself to explain it.

The strict answer, behaviorism, was that one need not go beyond it at all. Watson (1913) argued that psychology should abandon introspection and the study of consciousness and restrict itself to the prediction and control of observable behavior, treating the organism as a system whose outputs are lawful functions of its inputs and its history. This program was austere but productive: it forced psychology to operationalize its claims as measurable behavior and produced the quantitative study of learning. Its limitation, which the rest of this article follows, is that the mechanisms turning input and history into behavior proved to require internal description, representations, goals, values, and computations that are not themselves behavior but are inferred from it. The modern view keeps behavior as the ground truth while treating it as evidence about the machinery, and that reversal, from behavior as the whole subject to behavior as the measured output of a mechanism, organizes everything below.

Types of Behavior and Behavior Mechanisms

The scientific indexing vocabulary places behavior and behavior mechanisms at the root of its tree of behavioral phenomena; it has no parent category, and the sixteen headings below are its direct subtypes. The grouping is an indexing classification, not a theory of how behavior is organized: it collects headings that range from broad processes such as motivation and emotion to narrow dispositions such as temperance, and their arrangement reflects the needs of literature retrieval rather than any claim that these are the natural joints of behavior. The subtypes are orthogonal in the sense that a single act can fall under several at once, a motivated social choice colored by emotion, and the classification does not partition behavior into mutually exclusive kinds. Only the subtypes that are themselves developed articles on this site are linked; the remainder are listed for completeness.

Because the category sits at the root of the behavioral tree, its subtypes are not variants of one process but the major divisions of the field itself. The cognitive analysis that follows cuts across them: operant learning touches motivation and emotion, social-cognitive theory touches attitude and personality, and the neural account of reward learning underlies choice in every one. The classification locates the topic; it does not explain it.

From Behaviorism to the Analysis of Behavior

The first systematic account of behavior mechanisms was associationist and environmental. Watson (1913) proposed that behavior is built from conditioned reflexes, stimulus-response bonds laid down by experience, and that the proper business of psychology is to establish the laws by which a given stimulus comes to evoke a given response. The environmental account had already acquired its central principle before behaviorism was named: Thorndike (1898), studying cats that learned to escape from puzzle boxes, found that responses followed by a satisfying consequence were gradually stamped in and emitted faster on later trials while ineffective responses dropped out, and he formalized this as the law of effect, the principle that a behavior's consequences govern its future probability. The approach reached its most powerful form in the experimental analysis of behavior, where the central mechanism is not the reflex but the operant: a class of responses defined by their effect on the environment and selected by their consequences. A response followed by reinforcement becomes more probable; one followed by punishment or extinction becomes less so. The organism's behavior is thus shaped over time by a contingency between what it does and what follows, much as variation and selection shape a population.

Skinner (1950) pressed the methodological point that this account needed no appeal to inner causes. In Are theories of learning necessary? he argued that explanations invoking unobservable mental or physiological events add nothing a functional analysis of the relation between behavior and its controlling variables does not already supply, and that psychology should describe those relations directly rather than postulate a hidden apparatus. This was a coherent and fruitful stance, and the orderly data of schedules of reinforcement vindicated its empirical core. But the restriction to external variables became the fault line of twentieth-century psychology. The sections that follow are, in effect, the accumulating evidence that the relation between behavior and its consequences is itself produced by internal mechanisms, biological, representational, and neural, that must be described if the behavior is to be understood rather than merely catalogued.

Biological Constraints on Learning

If behavior were shaped by a single general law of reinforcement, any response should be conditionable to any consequence with equal ease. It is not. Garcia and Koelling (1966) demonstrated the point decisively: rats readily associated a novel taste with later nausea but not with an immediate shock, and readily associated an audiovisual cue with shock but not with nausea. The ease of learning depended on the content of the association, not merely on the contiguity or contingency between events, as though the animal were biologically prepared to connect tastes with illness and external cues with pain. Seligman (1970) generalized this into the concept of preparedness: species come equipped with a dimension of readiness along which some associations are learned in a single trial, others only slowly, and still others not at all, so the laws of learning are relative to an organism's evolutionary history rather than general.

The same lesson arrived from applied animal training. Breland and Breland (1961), students of Skinner who conditioned animals commercially, reported that trained operants would drift back toward species-typical foraging behavior, a raccoon that would not release coins to earn food but rubbed them together as it would wash prey, a pig that rooted its tokens instead of depositing them. They called this instinctive drift and treated it as evidence that reinforcement operates not on a blank behavioral slate but on a repertoire already structured by evolution. The demonstration above makes the preparedness idea quantitative: a prepared association climbs toward its asymptote in a handful of trials while an unprepared one, governed by a smaller learning rate, lags far behind over the same experience, so the identical schedule of reinforcement yields very different behavior depending on what is being associated. Biological constraints do not overturn the law of effect; they specify the boundary conditions within which it holds.

Purposive and Social-Cognitive Behavior

The internal structure of behavior was pressed from another direction by evidence that animals behave as though they represent goals and layouts rather than merely emitting conditioned responses. Tolman (1948) argued for a purposive behaviorism in which rats learning a maze acquire a cognitive map, an internal representation of the spatial situation, rather than a chain of stimulus-response bonds. His evidence was behavioral but pointed inward: rats showed latent learning, mastering a maze they had explored without reward as soon as reward was introduced, and took novel shortcuts when the trained path was blocked. Such behavior is hard to explain as a fixed response chain and natural to explain as the use of a stored map, and it established that behavior can express knowledge the organism had no occasion to perform until a goal made it relevant.

Bandura (1977) extended internal mechanisms into the social domain and into belief about the self. His social-cognitive theory held that much human behavior is acquired by observing others rather than by direct reinforcement, and that whether a person then performs what they have learned depends on self-efficacy, the belief that one can successfully execute the behavior required to produce an outcome. Self-efficacy, on this account, is a mechanism in its own right: it determines which activities people attempt, how much effort they expend, and how long they persist in the face of difficulty, independent of their actual skill. With Tolman's cognitive map and Bandura's efficacy beliefs, the internal variables that strict behaviorism had excluded were readmitted as the very things that make the relation between situation and behavior what it is.

The Neural Mechanism of Reward Learning

The reconciliation of the behaviorist and cognitive traditions came from identifying the mechanism of reinforcement in the brain. Formal learning theory had already recast the law of effect as a computation: an organism maintains an estimate of the value of a situation or action and updates it in proportion to the prediction error, the difference between the reward it received and the reward it expected. Schultz, Dayan, and Montague (1997) showed that this abstract quantity has a physical signal. Recording from midbrain dopamine neurons, they found that the cells fired to an unexpected reward, fell silent when an expected reward was omitted, and, as a cue came to predict reward, transferred their response from the reward to the cue, exactly the signature of a reward-prediction error in a temporal-difference learning model. Dopamine was not a pleasure signal but a teaching signal, broadcasting how much better or worse an outcome was than expected.

Figure 1

The Dopamine Reward-Prediction-Error Signature

Niv (2009) reviewed how thoroughly this correspondence recast the study of behavior, treating reinforcement learning as a bridge between the normative question of how an agent should learn from reward and the mechanistic question of how the brain does. Rangel, Camerer, and Montague (2008) built the framework into a general account of value-based decision making, decomposing a choice into the representation of options, the valuation of each, the selection of an action, and the learning that updates values from outcomes, each stage mapped onto identifiable neural systems. The demonstration above runs the core update rule trial by trial: a value estimate climbs toward the reward as its prediction error shrinks geometrically, so that learning is fastest when expectations are most wrong and slows to nothing as prediction catches up with reality. This single equation, surveyed next in the Worked Example, connects the rat's lever press, Tolman's map, and the firing of a dopamine neuron to one computational idea.

Behavior as Explanandum

Even as behavior became a window onto computation, a parallel argument insisted that it remains the thing to be explained and cannot be bypassed. Tinbergen (1963), founding the discipline of ethology with Konrad Lorenz, set out four complementary questions any complete account of a behavior must answer: its causation (the immediate mechanism that triggers it), its development (how it arises over the individual's lifetime), its function (the survival or reproductive advantage it confers), and its evolution (its phylogenetic history across species). The four questions are not rival explanations but distinct and jointly necessary ones, two concerning proximate mechanism and two concerning ultimate, evolutionary, cause, and conflating them is a persistent source of confusion in the study of behavior.

The demonstration above sorts questions about a behavior into Tinbergen's fourfold scheme, making explicit that an answer about neural mechanism neither competes with nor substitutes for an answer about evolutionary function. This framework underwrites a recent methodological argument. Krakauer, Ghazanfar, Gomez-Marin, MacIver, and Poeppel (2017) warned that neuroscience's increasing power to record and manipulate neurons has encouraged a reductionist bias in which detailed behavior is treated as a mere readout, and argued that understanding the brain requires first characterizing behavior at its own level of description, because the computations a neural circuit implements are only interpretable against the behavior they serve. Juavinett, Erlich, and Churchland (2018) extended the point to experimental practice, examining the tradeoffs in choosing behavioral tasks for neuroscience and urging designs rich enough to capture the structure of natural decision making rather than impoverished tasks chosen only for ease of neural recording. Behavior, on this view, is not the surface beneath which the real explanation lies; it is the level of description that gives every other level its meaning.

Worked Example

Consider the reward-prediction-error rule used in the reward-learning demonstration. An agent holds a value estimate V of how much reward a cue predicts and updates it after each trial by V ← V + α(r − V), where r is the reward received, α is the learning rate, and the quantity δ = r − V is the prediction error. Take a learning rate α = 0.3, a reward of r = 1 delivered on every trial, and a starting estimate V₀ = 0, so the agent begins expecting nothing.

On trial 1 the prediction error is δ = 1 − 0 = 1.00, and the estimate becomes V₁ = 0 + 0.3 × 1.00 = 0.30. On trial 2, δ = 1 − 0.30 = 0.70, so V₂ = 0.30 + 0.3 × 0.70 = 0.51. On trial 3, δ = 1 − 0.51 = 0.49, so V₃ = 0.51 + 0.3 × 0.49 = 0.657. On trial 4, δ = 1 − 0.657 = 0.343, so V₄ = 0.657 + 0.3 × 0.343 = 0.7599. The pattern is exact: each prediction error is 0.7 times the one before, because the update closes 30 percent of the remaining gap every trial, so δ on trial t equals 0.7^(t−1) and the estimate equals Vₜ = 1 − 0.7^(t). After five trials V₅ = 1 − 0.7⁵ = 1 − 0.168 = 0.832, already five-sixths of the way to the reward. The estimate approaches r = 1 as the prediction error decays to zero; the behavior stabilizes precisely when the organism stops being surprised. This is the law of effect written as arithmetic: reinforcement changes behavior in proportion to how unexpected the reinforcement is, and once outcomes are fully predicted, learning, and the dopamine teaching signal that drives it, falls silent.

Discussion

The mechanisms surveyed here form a single historical argument rather than a list of competing schools. Behaviorism established behavior as the measurable object of psychology and discovered the law of effect; the biological-constraints literature showed that this law operates on a repertoire pre-structured by evolution; the purposive and social-cognitive traditions showed that the relation between situation and behavior is mediated by internal representations, goals, and beliefs; and the neuroscience of reward learning identified the physical signal that implements reinforcement, recasting the law of effect as a prediction-error computation carried by dopamine. Each development kept the data of the previous one while enlarging the description needed to explain them, so that the modern account treats behavior as the lawful output of a layered mechanism spanning ecology, representation, and neural computation.

This layering is why behavior and behavior mechanisms sits at the root of the field's classification. Learning, motivation, emotion, personality, and social conduct are not separate subjects that happen to be filed together but different faces of the one problem of how an organism's activity is produced and changed. The ethological argument gives the final discipline to the picture: a complete explanation of any behavior requires answers at the level of immediate mechanism and at the level of evolutionary function, and the two cannot be substituted for each other. Behavior is thus both the discipline's oldest object and its integrating one, the common measure to which accounts of mind must ultimately return.

Cognitive Implications

The trajectory from behavior-as-subject to behavior-as-evidence has a direct methodological consequence for cognitive psychology: behavior is the data, and internal structure is the inference. Every claim about a representation, a value, or a belief is ultimately licensed by a measured pattern of action, which is why the field's rigor depends on tasks whose behavior is diagnostic of the mechanism in question. The reward-prediction-error account is the clearest case: an abstract learning rule predicted a specific, counterintuitive neural signature, a transfer of response from reward to predictive cue, that was then observed, closing the loop between behavior, computation, and physiology.

The same logic warns against two failures. One is to treat behavior as a mere readout to be explained away once the neurons are understood; the ethological and task-design arguments show that behavior defines what the neurons are doing, so an impoverished behavioral description yields an uninterpretable neural one. The other is to multiply internal entities beyond what behavior can distinguish, the concern that motivated Skinner's skepticism and that remains a legitimate constraint. The discipline advances when a candidate mechanism makes a behavioral prediction precise enough to be wrong, so that the measured activity of the organism can adjudicate between accounts rather than merely illustrate one.

Current Directions

The most active current extends reinforcement learning from a single update rule to a theory of how the brain combines multiple learning systems. Gershman and Daw (2017) reviewed the integration of reinforcement learning with episodic memory, arguing that agents do not learn only by incrementally updating cached values but also by retrieving specific past episodes to evaluate options, a dual mechanism that better fits human choice than either alone. The open questions concern how the brain arbitrates between a fast, memory-based system and a slow, value-caching one, and when each is deployed.

A second current connects these ideas to artificial intelligence. Botvinick, Ritter, Wang, Kurth-Nelson, Blundell, and Hassabis (2019) examined why deep reinforcement learning agents, like animals early in training, learn slowly, and how mechanisms such as episodic memory and meta-learning can produce the fast, flexible learning that biological behavior displays, making the comparison between artificial and natural agents a two-way exchange. Running alongside both is the task-design debate opened by Juavinett, Erlich, and Churchland (2018), on how to build behavioral paradigms rich enough to engage the mechanisms of natural decision making while remaining tractable for neural measurement. Across these lines the direction is the same: to specify, with increasing computational precision, the mechanisms that turn experience into behavior.

Common Misconceptions

Behaviorism denied that mental states exist.

The methodological claim was that psychology should explain behavior through its controlling variables rather than postulate inner causes, not that inner states are unreal (Skinner, 1950). The modern account keeps behavior as the evidence while inferring the internal mechanisms behind it.

Any behavior can be conditioned to any consequence with enough training.

Learning is biologically constrained: some associations are acquired in a single trial and others scarcely at all, depending on the species' evolutionary preparedness (Garcia & Koelling, 1966; Seligman, 1970). The law of effect holds only within those boundaries.

Dopamine is the brain's pleasure chemical.

The phasic dopamine signal encodes a reward-prediction error, how much better or worse an outcome was than expected, not the hedonic experience of reward itself; it falls silent once rewards are fully predicted (Schultz, Dayan, & Montague, 1997). It is a teaching signal, not a pleasure signal.

Glossary

Behavior.

Any observable, measurable activity of an organism; the response side of the relation between an organism and its environment.

Behaviorism.

The program, founded by Watson, that restricts psychology to the prediction and control of observable behavior and its environmental determinants.

Cognitive map.

Tolman's term for an internal representation of a spatial situation that an organism can use to navigate flexibly, rather than a fixed chain of responses.

Instinctive drift.

The tendency of a conditioned operant to revert toward species-typical behavior, showing that reinforcement acts on an evolutionarily structured repertoire.

Law of effect.

The principle that responses followed by satisfying consequences become more probable and those followed by aversive consequences less so.

Operant conditioning.

The process by which the frequency of a response changes as a function of the reinforcing or punishing consequences that follow it.

Operant.

A class of responses defined by their effect on the environment and modifiable by their consequences, as distinct from a reflex elicited by a stimulus.

Prediction error.

The difference between a received reward and the reward that was expected; the quantity that drives value updating in reinforcement learning.

Preparedness.

Seligman's dimension describing how readily a species forms a given association, from one-trial learning to associations that cannot be formed at all.

Reinforcement learning.

A framework in which an agent learns the value of situations and actions by updating estimates in proportion to prediction error.

Reward-prediction error.

The specific prediction-error signal, encoded by phasic firing of midbrain dopamine neurons, that functions as a teaching signal for learning.

Self-efficacy.

Bandura's construct for a person's belief in their capacity to execute the behavior needed to produce a given outcome, governing effort and persistence.

Temporal-difference learning.

A reinforcement-learning method that updates value estimates from the moment-to-moment change in predicted reward, matching the dopamine signal's dynamics.

Tinbergen's four questions.

The four complementary levels, causation, development, function, and evolution, at which any behavior must be explained; two proximate and two ultimate.

Key Researchers

Albert Bandura

(1925-2021). Psychologist at Stanford University; developed social-cognitive theory and the concept of self-efficacy, showing that behavior is acquired by observation and regulated by beliefs about one's own capacity to act. Wikipedia - Wikidata

Yael Niv

. Professor of Psychology and Neuroscience at Princeton University; advanced the computational theory of reinforcement learning in the brain, linking dopaminergic prediction-error signals to how organisms learn the value of actions. Faculty Page

Wolfram Schultz

. Professor of Neuroscience at the University of Cambridge; discovered that midbrain dopamine neurons encode a reward-prediction error, giving reinforcement learning a direct neural substrate. ORCID - Wikipedia - Wikidata

Martin E. P. Seligman

(b. 1942). Zellerbach Family Professor of Psychology at the University of Pennsylvania; demonstrated biological preparedness in associative learning and learned helplessness, showing that the laws of learning are constrained by evolutionary history. Faculty Page - Google Scholar - Wikipedia - Wikidata

B. F. Skinner

(1904-1990). Psychologist at Harvard University; founded the experimental analysis of behavior, formalizing operant conditioning and arguing that behavior is shaped and maintained by its consequences. Wikipedia - Wikidata

Edward C. Tolman

(1886-1959). Psychologist at the University of California, Berkeley; introduced purposive behaviorism and the cognitive map, showing that behavior reflects internal representations and goals rather than stimulus-response chains. Wikipedia - Wikidata

John B. Watson

(1878-1958). Psychologist at Johns Hopkins University; launched behaviorism as a program for psychology, arguing that the discipline should study only observable behavior and its environmental determinants. Wikipedia - Wikidata

Frequently Asked Questions

What is behavior in psychology?

It is any observable, measurable activity of an organism, the response side of the relation between the organism and its environment; behaviorism made it the discipline's object of study (Watson, 1913).

What is the difference between behavior and a behavior mechanism?

Behavior is the activity itself; a behavior mechanism is the process, environmental, representational, or neural, that produces, modifies, and regulates that activity. Modern psychology treats behavior as the measured evidence for the underlying mechanism.

What is operant conditioning?

The process by which the frequency of a response changes according to its consequences: responses followed by reinforcement become more probable and those followed by punishment less so (Skinner, 1950).

Why can't any behavior be trained with equal ease?

Because learning is biologically constrained: a species is prepared by evolution to form some associations in a single trial and others scarcely at all, so the ease of conditioning depends on the content of the association (Garcia & Koelling, 1966; Seligman, 1970).

What did Tolman's cognitive map show?

That rats learning a maze acquire an internal representation of its layout rather than a fixed chain of responses, evidenced by latent learning and novel shortcuts, which reinstated internal variables in the explanation of behavior (Tolman, 1948).

What is self-efficacy?

Bandura's construct for a person's belief in their capacity to carry out the behavior needed to achieve an outcome; it governs which activities people attempt, how hard they try, and how long they persist (Bandura, 1977).

What does dopamine actually signal?

A reward-prediction error, how much better or worse an outcome was than expected, which serves as a teaching signal for learning; it is not a direct signal of pleasure (Schultz, Dayan, & Montague, 1997).

What are Tinbergen's four questions?

The four complementary levels at which any behavior must be explained, its immediate causation, its development over the lifespan, its adaptive function, and its evolutionary history, two proximate and two ultimate (Tinbergen, 1963).

References