Connectionism and the Law of Effect – Edward Thorndike

At the close of the nineteenth century, the nascent discipline of scientific psychology stood precariously at an intellectual crossroads. Having broken away from speculative Cartesian philosophy and British mentalism, early experimentalists sought to establish an empirical science of mind. However, much of the fledgling field remained mired in the subjective method of analytic introspection championed by Wilhelm Wundt and Edward Bradford Titchener, or the anthropomorphic anecdotalism that dominated early comparative animal studies. It was within this fragmented intellectual terrain that Edward Lee Thorndike published his groundbreaking 1898 doctoral dissertation, Animal Intelligence: An Experimental Study of the Associative Processes in Animals. Thorndike fundamentally transformed the study of mind by replacing introspection with objective measurement, and anecdotal romance with the cold mechanics of the laboratory puzzle box.

Thorndike’s foundational paradigm, known as Connectionism, posited that all learning, regardless of phylogenetic complexity, could be reduced to the formation and modification of specific associative bonds between sensory impressions (stimuli) and motor impulses (responses). Rather than attributing human-like ratiocination, abstract deliberation, or sudden flashes of conceptual insight to non-human organisms, Thorndike demonstrated that learning is inherently incremental, mechanical, and governed by deterministic laws. At the core of this theoretical edifice was the Law of Effect, a principle asserting that actions followed by satisfying states of affairs are stamped into the neural architecture of the organism, while those followed by annoying states are systematically eradicated. This mechanistic formulation transformed functionalist psychology and provided the empirical cornerstone upon which the behavioral revolution was erected.

Far from remaining an obsolete relic of pre-Skinnerian behaviorism, Thorndike’s theoretical architecture continues to exert a pervasive influence across contemporary cognitive science, neurobiology, educational technology, and artificial intelligence. From the synaptic plasticity rules formulated by Donald Hebb to the distributed weight adjustments of artificial neural networks and the mathematical algorithms of modern reinforcement learning, the core tenets of connectionism have demonstrated profound longevity. This treatise provides an exhaustive analytical investigation into Thorndike’s connectionism, tracing its historical epistemology, dissecting the experimental mechanics of the puzzle box studies, charting the conceptual evolution and revisions of the primary and subordinate laws of learning, examining its pedagogical revolutions, and detailing its mathematical resurgence in contemporary computational neuroscience.

1. Introduction to Edward Thorndike and the Foundations of Connectionism

1.1 Biographical Trajectory and Academic Milieu

Edward Lee Thorndike was born in 1874 in Williamsburg, Massachusetts, the son of a Methodist minister whose itinerant pastoral assignments imbued the young scholar with an ascetic work ethic and an abiding respect for disciplined inquiry. Thorndike matriculated at Wesleyan University, where he initially pursued classical literature, before being intellectually captivated by William James’s seminal 1890 masterwork, The Principles of Psychology. Moving to Harvard University in 1895 to pursue graduate studies under James, Thorndike began his experimental investigations into animal learning. Denied attic laboratory space by the Harvard administration, Thorndike famously incubated and ran conditioned choice experiments with chicks in the cellar of James’s Cambridge home, a vivid testament to the improvisational and disruptive nature of early comparative psychology.

Frustrated by Harvard’s limited institutional resources for comparative animal laboratories, Thorndike accepted a fellowship at Columbia University in 1897 under the mentorship of James McKeen Cattell, one of the pioneers of quantitative psychometrics. Cattell’s rigorous focus on mental measurement, statistical variance, and objective instrumentation provided the ideal academic crucible for Thorndike’s empirical inclinations. Thorndike brought with him to New York a radical disillusionment with speculative mentalism. He rejected the introspective paradigm that sought to describe the ephemeral contents of consciousness through subjective self-observation, arguing that psychology could only attain the status of a natural science if it anchored its findings to publicly verifiable physical stimuli and quantifiable muscular movements.

The culmination of this transition was the 1898 publication of his doctoral monograph, Animal Intelligence: An Experimental Study of the Associative Processes in Animals. This work delivered a decisive methodological shock to the academic establishment. Thorndike replaced the romanticized, armchair philosophizing of British associationism with rigorous chronometric curves derived from custom-built apparatuses. His subsequent transition to Teachers College at Columbia University in 1899 marked a permanent pivot toward the practical application of his laboratory principles. Over a prolific career spanning five decades, during which he authored over five hundred books and articles, Thorndike transformed comparative psychology into an applied science of human engineering, setting the stage for twentieth-century instructional design and psychometric assessment.

1.2 Conceptual Definition of Connectionism

Connectionism, within the Thorndikian taxonomy, denotes an original Stimulus-Response (S-R) associative framework that conceptualizes the mind as an intricate, physical, and functionally organized network of bonds connecting sensory input to motor output. In stark opposition to the structuralism of Wilhelm Wundt and Edward Titchener, which sought to dissect conscious experience into elemental sensations, images, and affective qualities, Thorndike completely bypassed the introspective analysis of inner states. He asserted that the primary data of psychology must be the functional correlations observed between the presentation of environmental events and the subsequent kinetic output of the organism.

Central to this connectionist architecture is the ontological assertion that the fundamental unit of learning is the neural bond. When an organism interacts with its surroundings, sensory receptors are activated by an objective pattern of physical energy. This incoming excitation does not prompt an autonomous, immaterial consciousness to deliberate upon courses of action. Rather, it is mechanically shunted through a network of neuromuscular conduction pathways, culminating in a specific somatic or visceral movement. Learning, therefore, is nothing more and nothing less than the physiological modification of these pathways, altering the statistical probability that a given stimulus constellation ($S$) will evoke a specific behavioral response ($R$).

By defining the mind as an aggregate of concrete associations, Thorndike introduced a thoroughly quantitative paradigm. He contended that individual differences in intellectual capacity do not stem from qualitative differences in the essence of an indivisible soul or a transcendental rational ego. Instead, intellect is a direct mathematical function of the sheer quantity, strength, and complexity of the S-R bonds an individual has established. In his 1905 work, The Elements of Psychology, Thorndike articulated this radical reductionism, asserting that human mental life, with all its poetic, scientific, and philosophical triumphs, is structurally continuous with the basic connective networks operating in lower animals, differing solely in the vast number and hyper-refined granularity of its associative bonds.

1.3 Epistemological Shift in Comparative Psychology

To appreciate the disruptive force of Thorndike’s connectionism, one must examine the epistemological landscape of late nineteenth-century comparative psychology. Following the publication of Charles Darwin’s The Descent of Man in 1871, evolutionary thinkers sought to demonstrate mental continuity between animals and humans. However, this pursuit was largely conducted via the “anecdotal method,” most vividly exemplified by George John Romanes in his 1882 volume Animal Intelligence. Romanes compiled second-hand, anthropomorphic accounts from pet owners and amateur naturalists, attributing complex cognitive faculties, such as grief, deception, abstract reasoning, and intentionality, to dogs, cats, and birds based on isolated, unverified observations.

Thorndike mounted a devastating critique against this anthropomorphic tradition. He argued that the anecdotal method suffered from catastrophic confirmation bias, systematically cataloging unique, seemingly miraculous feats of animal cunning while ignoring the millions of instances where animals failed to solve simple problems. Moreover, by interpreting animal actions through the lens of human introspective consciousness, Romanes and his contemporaries committed what Thorndike viewed as an unpardonable scientific error: attributing high-level rational deduction to actions that could be fully explained by lower-order physiological mechanisms.

In response, Thorndike rigorously applied the methodological principle known as Lloyd Morgan’s Canon, formulated by British psychologist C. Lloyd Morgan in 1894: “In no case may we interpret an action as the outcome of the exercise of a higher psychical faculty, if it can be interpreted as the outcome of the exercise of one which stands lower in the psychological scale.” Thorndike operationalized this canon by confining non-human animals to standardized, replicable laboratory puzzle boxes where their every movement could be precisely recorded, timed, and analyzed. Through this experimental approach, Thorndike stripped comparative psychology of its anthropomorphic myths, demonstrating that animal problem-solving was not the result of sudden deductive insight or rational contemplation, but the product of blind, deterministic associative processes unfolding under the pressure of biological drive.

2. Historical Epistemology: From Associationism to Functionalist Behaviorism

2.1 Lineage of British Associationism

While Thorndike was an arch-empiricist, his conceptual framework did not emerge in a vacuum; it represented a radical materialization of classical British associationism. Beginning with John Locke’s 1690 An Essay Concerning Human Understanding, associationist philosophy argued against the Cartesian notion of innate ideas, proposing instead that the mind begins as a tabula rasa upon which sensory experience inscribes simple ideas. These simple ideas, through repeated contiguity in space and time, coalesce into complex conceptual structures. David Hume expanded this doctrine in 1739 by identifying resemblance, contiguity, and cause-and-effect as the universal connective tissues of human mental operations, while David Hartley provided an early physiological grounding in 1749 by hypothesizing that mental associations correspond to miniature physical vibrations within the nervous system.

However, classical associationism remained an idealistic and introspective enterprise: it dealt almost exclusively with associations formed between mental ideas within a disembodied theater of consciousness. Thorndike’s historical breakthrough lay in taking these abstract laws of mental association and externalizing them into the physical domain of somatic behavior. He broke with the tradition of Locke, Berkeley, and Hume by arguing that the primary association is not between two passive cognitive representations (Idea A summoning Idea B), but between an objective environmental condition (Stimulus) and an overt muscular action (Response).

In executing this pivot, Thorndike was heavily indebted to the transitional insights of Scottish philosopher Alexander Bain. In his 1855 treatise The Senses and the Intellect and his 1859 companion work The Emotions and the Will, Bain introduced the concept of spontaneous activity. Bain noted that animals often engage in random, exploratory kinetic movements in the absence of explicit environmental stimulation. If one of these spontaneous movements accidentally results in an agreeable sensation or the alleviation of physical discomfort, the animal experiences a physiological pleasure that, through association, increases the likelihood of that specific movement being repeated. Thorndike seized upon Bain’s philosophical hypothesis and transformed it into a rigorous, experimentally verified law of learning.

2.2 Functionalism and Evolutionary Adaptation

Thorndike’s connectionism was equally shaped by the intellectual current of American functionalism, an evolutionary school of thought nurtured by William James, John Dewey, and James Rowland Angell. Functionalism discarded the static, structural taxonomic categorization of conscious elements in favor of an active, teleological inquiry: What is the adaptive utility of the mind in the struggle for survival? Rooted in Charles Darwin’s paradigm of natural selection, functionalism viewed psychological phenomena as biological instruments evolved to mediate between the urgent internal drives of the organism and the complex, fluctuating demands of its ecological niche.

Thorndike recognized that phylogenetic natural selection, operating over generational timescales through differential reproductive success, was insufficient to account for an individual organism’s immediate, ontogenetic adjustments to dynamic environments. If an animal’s behavioral repertoire were entirely hardwired through genetic selection (instinct), any minor variation in the ecology of its habitat—such as a shifted predator path or a novel obstacle blocking a food source—would lead to extinction. There had to be an intra-organismic, ontogenetic analogue to natural selection: a rapid, flexible behavioral mechanism capable of calibrating the organism’s kinetic outputs in real time.

This ontogenetic adaptation was precisely what the connectionist framework provided. S-R bond formation functioned as an evolutionary trial-and-error engine operating within the lifespan of a single individual. Random behavioral variants were generated (akin to genetic mutations), and environmental outcomes—satisfaction or annoyance—acted as selective filters, preserving adaptive motor acts and discarding maladaptive ones. Under this functionalist lens, the mind was not an ornate, introspective chamber of contemplation, but an operational toolkit for achieving physical equilibrium, energy conservation, and environmental mastery.

2.3 The Precursor to Watsonian Behaviorism

The historical positioning of Thorndike has been a subject of enduring debate among historians of psychology: was he the final and most sophisticated of the evolutionary functionalists, or the foundational architect of early behaviorism? When John B. Watson issued his explosive 1913 manifesto, “Psychology as the Behaviorist Views It,” he explicitly repudiated the introspective method, redefined psychology as a purely objective experimental branch of natural science, and proclaimed behavior prediction and control as its sole aims. In executing this revolution, Watson built directly upon the methodological foundation Thorndike had engineered fifteen years prior.

Methodologically, Thorndike was undeniably proto-behaviorist. He banned introspective protocols from his comparative laboratories, operationalized all dependent variables as observable escape latencies, and treated the organism as an input-output mechanism responding deterministically to physical environmental triggers. Watson’s rejection of mentalistic concepts in favor of the strict Stimulus-Response schema was a direct appropriation of the connectionist bond. Thorndike had already demonstrated that an exhaustive, rigorous account of animal learning could be articulated without appealing to subjective consciousness.

Yet, Watson ultimately broke with Thorndike over the latter’s retention of affective terminology. Thorndike’s central thesis rested upon the concepts of “satisfaction” and “discomfort” (or “annoyance”), constructs that Watson attacked as unscientific, subjective, and teleological remnants of introspective mentalism. Watson championed a radical peripheralism that recognized only physical contiguity, frequency, and recency of movement, flatly denying that affective consequences could retroactively alter neuromuscular pathways. Thorndike resisted Watson’s absolute peripheralism, maintaining that consequence was the ultimate arbiter of learning. This critical divergence positioned Thorndike not as a radical Watsonian, but as the direct conceptual grandfather of B.F. Skinner’s operant conditioning, serving as the historical bridge connecting nineteenth-century functionalism to twentieth-century neobehaviorism.

3. The Experimental Apparatus: Puzzle Boxes and Animal Intelligence Studies

3.1 Engineering and Mechanics of the Puzzle Box

The empirical engine of Thorndike’s connectionism was the puzzle box, an ingenious family of experimental testing chambers designed and hand-built by Thorndike using discarded wooden packing crates, wire mesh, and simple hardware. These apparatuses were constructed to house individual animal subjects—primarily domestic cats, but also dogs and chicks—under strictly controlled, repeatable physical conditions. The primary design objective was to present the animal with an unmistakable environmental problem: the subject was enclosed within a confined space while a desirable biological incentive (typically a piece of fresh fish or meat) was placed directly outside the wire slats, fully visible and olfactory-accessible, yet physically unreachable without opening a release hatch.

The release mechanisms integrated into these boxes exhibited varying degrees of mechanical complexity. Box “A,” the most iconic apparatus, measured approximately twenty inches long, fifteen inches wide, and twelve inches high. Its spring-loaded escape door was held shut by a latch connected via a system of cords and pulleys to a wire loop suspended inside the center of the cage, as well as to a small wooden pedal or “treadle” mounted on the floor. Depressing the treadle with a paw or pulling downward on the loop introduced mechanical tension that pulled the bolt, allowing the door to swing open instantly under spring tension.

Thorndike designed multiple iterations of the puzzle box to systematically vary the mechanical demands placed on the animals. Simpler boxes required a single, direct motor action:

• Box A: Pulling a wire loop or depressing a floor treadle.
• Box C: Turning a wooden button that barred the door from the inside.
• Box K: Stepping on a triple-jointed floor lever.

More advanced configurations introduced multi-step, sequential contingencies:

• Box 100: Required an animal to depress a treadle, pull a string hanging from the roof, and slide back a horizontal deadbolt before the door would release.

Crucially, Thorndike standardized the internal state of his animal subjects by implementing rigorous food deprivation protocols. The cats were tested prior to their daily feeding, ensuring a heightened, operationalized state of hunger-driven motivation that catalyzed immediate, persistent kinetic output.

3.2 Empirical Observations of Trial-and-Error Learning

When an uninitiated cat was deposited inside a puzzle box, its immediate behavioral output was characterized by frantic, uncoordinated, and indiscriminate kinetic activity. In the vivid descriptions recorded in his 1898 monograph, Thorndike documented that the animal did not sit quietly to visually survey the mechanical components of the cage, nor did it deduce the relationship between the string, the pulley, and the door latch. Instead, propelled by hunger and confinement, the animal engaged in instinctual escape reactions: it clawed violently at the wire mesh, squeezed its body through narrow slats, thrust its paws through every aperture, bit at the wooden framing, and struck out wildly in all physical directions.

Inevitably, amid this chaotic burst of motor activity, the animal would accidentally trigger the release mechanism. In Box A, a thrashing paw would inadvertently snag the wire loop or land upon the floor treadle. The latch lifted, the spring door flew open, and the cat bounded out to consume the fish morsel waiting on the exterior platform. When Thorndike placed the same animal back into the box for a second trial, the cat did not immediately repeat the successful action. Instead, it resumed its generalized, frantic clawing and thrashing, though the total duration of these extraneous movements was slightly reduced. Accidental success occurred once more, followed by external reward.

Over a succession of dozens of sequential trials, a distinct behavioral pattern emerged: the progressive, gradual elimination of all useless, extraneous movements. The non-functional behaviors—clawing at the ceiling, biting the side slats, attempting to burrow under the frame—gradually diminished in frequency and vigor. Simultaneously, the single motor act that reliably produced the latch release—depressing the treadle—was executed with increasing rapidity and spatial economy. Eventually, by the twentieth or thirtieth trial, the cat, upon being placed into the box, would walk directly to the treadle, press it downward without hesitation, and exit through the opened door. This mechanical progression led Thorndike to coin the term trial-and-error learning, which he later reformulated as the more precise conceptual phrase “selecting and connecting.”

3.3 Quantitative Analysis of Learning Curves

Thorndike’s definitive empirical refutation of animal rationality rested upon his mathematical analysis of learning curves. He meticulously recorded the latency of escape—the precise number of seconds elapsed from the moment the animal was locked in the box until the instant the door latch was tripped—and plotted these temporal latencies on the vertical axis against the successive ordinal trial numbers on the horizontal axis. These graphs represented some of the very first quantitative learning curves in the history of psychology.

Thorndike argued that if animals possessed the capacity for rational deduction, abstract conceptualization, or what the later Gestalt school would designate as “insight,” the mathematical topography of these curves would exhibit a distinctive discontinuous shape. The latency curve would remain elevated during the initial phase of perplexity, and then, at the precise moment the animal mentally grasped the mechanical principle of the lever, the curve would undergo an abrupt, near-vertical plunge to near-zero latency, remaining permanently flat thereafter. Insight would manifest as an unmistakable step-function: an instantaneous transformation from total ignorance to absolute mastery.

The empirical data, however, demonstrated no such discontinuity. Across hundreds of experimental runs with dozens of cats, dogs, and chicks, the learning curves universally presented as gradual, jagged, highly irregular downward slopes. The latency dropped incrementally, frequently backsliding on subsequent trials when the animal temporarily reverted to useless movements before stabilizing into rapid execution. Thorndike asserted that this gradual, continuous downward progression mathematically disproved the existence of rational insight. The jagged slope demonstrated that learning was not a cognitive realization, but a slow, mechanical stamping in of an S-R connection, accompanied by the gradual stamping out of competing non-functional bonds. Replications across distinct phylogenetic classes confirmed that this quantitative slope was an invariant biological characteristic of associative learning.

4. The Core Architecture of Connectionism: Stimulus-Response (S-R) Bonds

4.1 The Nature and Topography of S-R Bonds

The theoretical bedrock of Thorndike’s connectionism is the concept of the Stimulus-Response bond (frequently denoted as the S-R connection). To understand Thorndike’s architecture, one must clarify the ontological status of this bond. In its functional definition, an S-R bond is not a reified material entity, but a measurable shift in behavioral probability: given the presentation of sensory stimulus complex $S$, the probability that the organism will execute motor response $R$ approaches unity. However, in its physical definition, Thorndike conceived of the bond as an actual, material modification within the physiological conduits of the central and peripheral nervous systems.

Stimulus specificity in connectionist theory is highly granular. The “Stimulus” is not an abstract mental perception, but the complete physical array of sensory activations impinging on the animal at any given moment: the visual angles of the wooden box slats, the tension in the animal’s musculature, the olfactory signature of the testing chamber, and the visual geometry of the hanging wire loop. The “Response” is the precise, localized pattern of efferent motor impulses directed to specific somatic muscle groups: lifting the right forelimb, extending the claws, curling the digits around the wire, and applying downward kinetic force.

Crucially, an organism does not enter a learning situation as a neutral, unstructured system. Rather, the organism possesses an innate or previously learned bond hierarchy. Prior to experimental exposure, the stimulus conditions of confinement elicit a hierarchy of dominant response tendencies:

1. $R_1$: Clawing at the nearest opening.
2. $R_2$: Biting the wooden framing.
3. $R_3$: Meowing and vocalizing distress.
4. $R_{24}$: Lightly pressing a low wooden pedal with a paw.

In this initial hierarchy, the functional response ($R_{24}$) sits near the absolute bottom, with an exceptionally low probability of execution. The entire mechanics of connectionist learning consists of restructuring this hierarchy: demoting the unsuccessful dominant responses and elevating the successful subordinate response to the absolute apex of the associative matrix.

4.2 The Concept of Neural Plasticity in Early Connectionism

Operating decades before the development of electron microscopy, patch-clamp electrophysiology, or modern molecular neurobiology, Thorndike boldly advanced a speculative physiological substrate for his connectionist bonds. He drew heavily from the emerging neuron doctrine articulated by Santiago Ramón y Cajal and the conceptualization of the synapse pioneered by Charles Sherrington. Thorndike postulated that the brain and spinal cord functioned as a vast, highly integrated biological telephone switchboard, where incoming sensory lines were routed to outgoing motor lines through variable, plastic junctions.

Thorndike hypothesized that learning corresponds directly to changes in physiological conduction resistance at these synaptic interfaces. In his early texts, he spoke of “conduction units”—pathways of neurons linking sensory receptors to effector muscles. When an S-R bond is unformed or weak, the synaptic junction between the relevant afferent and efferent neurons possesses high physiological resistance, preventing the sensory excitation from traversing the path to trigger the motor response. Learning, according to Thorndike, is the physical reduction of this resistance. The synaptic junction undergoes a structural modification—which he variously conceptualized as terminal bouton swelling, dendritic growth, or chemical alterations within the cleft—that facilitates the passage of the nerve impulse.

While Thorndike recognized that early twentieth-century neurology lacked the empirical instruments to verify his micro-physiological hypotheses, he insisted that connectionism was fundamentally an epiphenomenon of nervous system mechanics. He anticipated Donald Hebb’s 1949 neurocomputational postulate by five decades, suggesting that the repeated passage of a neural impulse through a specific conduction unit, followed by a favorable metabolic state induced by satisfaction, physically altered the conductivity of the synapse. Thorndike anchored the mind directly to the biophysical laws of the central nervous system, rejecting any dualistic appeals to non-physical mental forces.

4.3 Trial-and-Error as ‘Selecting and Connecting’

As Thorndike’s theoretical framework matured, he grew increasingly dissatisfied with the popular nomenclature of “trial-and-error,” arguing that it obscured the precise mechanics of learning. In his 1931 work, Human Learning, he formally advanced the phrase “selecting and connecting” as the definitive scientific description of associative adaptation. The process of learning does not involve random flailing followed by deliberate cognitive choice; rather, it is a dual-stage, non-cognitive mechanical sorting process governed entirely by environmental contingencies.

The first component of this mechanism, selecting, is driven by the behavioral variability of the organism when confronted with an environmental obstacle. Because existing high-probability S-R bonds fail to alleviate the biological drive (such as hunger or confinement), the animal moves through its repertoire of response tendencies. The environmental context acts as a merciless selective filter, entirely analogous to the mechanism of natural selection in evolutionary biology. Movements that fail to produce a physiological transition toward equilibrium are mechanically discarded, while movements that coincidentally alter the physical status quo in an adaptive direction are captured by the environment.

The second component, connecting, represents the automated stamping in of the selected motor response to the concurrent sensory context. There is no intermediate cognitive deliberation, no internal voice declaring, “This pedal opens that door; therefore, I shall press it.” The connection is stamped in passively, below the threshold of conscious intentionality. Thorndike conceptualized this as a deterministic probability shift: each reinforcement mathematically increases the probability vector of that specific S-R coordinate within the organism’s total behavioral matrix, while decreasing the vectors of competing paths. The organism does not select the connection; the environment selects the response and stamps in the connection.

5. The Law of Effect: Theoretical Mechanics, Evolution, and Revision

5.1 Original Formulation of the Law of Effect (1898-1911)

The Law of Effect represents the theoretical core of Edward Thorndike’s connectionism and stands as one of the most consequential conceptual formulations in the history of behavioral science. In his 1911 seminal volume, Animal Intelligence, Thorndike articulated the classic, verbatim formulation of the law that would govern psychological discourse for decades:

“Of several responses made to the same situation, those which are accompanied or closely followed by satisfaction to the animal will, other things being equal, be more firmly connected with the situation, so that, when it recurs, they will be more likely to recur; those which are accompanied or closely followed by discomfort to the animal will, other things being equal, have their connections with that situation weakened, so that, when it recurs, they will be less likely to recur. The greater the satisfaction or discomfort, the greater the strengthening or weakening of the bond.”

To shield this formulation from charges of subjective mentalism, Thorndike provided strict, operational definitions of his affective terms. A satisfying state of affairs was defined not as an ephemeral, introspective feeling of joy or spiritual contentment, but as an objective, behavioral condition: “one which the animal does nothing to avoid, often doing such things as attain and preserve it.” Conversely, an annoying state of affairs (or discomfort) was operationally defined as “one which the animal avoids or changes.” By anchoring these definitions to observable approach and avoidance vectors, Thorndike sought to retain the explanatory power of reinforcement while preserving physicalistic rigor.

In this original 1911 iteration, the Law of Effect posited an absolute, elegant symmetry between reward and punishment. Satisfaction acted as a positive mechanical force that actively “stamped in” the neural connection, increasing its conductivity. Annoyance acted as an equal and opposite negative force that actively “stamped out” the connection, decreasing its conductivity or severing the synaptic bond. Furthermore, Thorndike had to confront the philosophical objection of retroaction: how could an event that occurs in the future (the reward received after pressing the lever) reach backward in time to alter a bond that had already transpired? Thorndike resolved this by asserting that the sensory-motor conduction trace persists within the nervous system for several seconds, allowing the systemic neurochemical state of satisfaction to act directly upon the physically lingering neural trace.

5.2 The Experimental Shift: Revision of the Law (1930s)

For more than three decades, the symmetrical formulation of the Law of Effect stood as an accepted pillar of psychological orthodoxy. However, in the late 1920s and early 1930s, Thorndike embarked on a massive empirical research program involving human subjects, primarily utilizing complex verbal learning, multi-choice associative tasks, and the acquisition of foreign vocabulary. The results of these meticulous investigations, published in The Fundamentals of Learning (1932) and The Psychology of Wants, Interests, and Attitudes (1935), forced Thorndike into a profound and courageous theoretical retreat: he radically truncated the Law of Effect, destroying the symmetry he had championed for thirty years.

The experimental paradigms were elegantly designed. Human participants were presented with thousands of stimulus cards containing rare Spanish words, each accompanied by four or five English options. The subject made a choice; if correct, the experimenter immediately declared “Right!” (a satisfier); if incorrect, the experimenter declared “Wrong!” (an annoyer). If the original Law of Effect were correct, the declaration of “Right” should stamp in the correct association, while the declaration of “Wrong” should actively stamp out the incorrect association, reducing the probability of that erroneous response being repeated on subsequent passes below the baseline level of chance.

The empirical data yielded a striking discovery: while the reward (“Right”) consistently and robustly stamped in the correct S-R bond, the punishment (“Wrong”) completely failed to weaken the incorrect bond. In many cases, an error punished by the declaration of “Wrong” was actually more likely to be repeated on the next trial than if the subject had received no feedback at all, simply because the mere occurrence of the response had exercised the connection (in accordance with the Law of Use). Punishment did not actively unseat, sever, or stamp out the neural connection. Thorndike was forced to concede that satisfaction and annoyance are fundamentally asymmetrical. He formally excised the punitive half of the Law of Effect, concluding that reward is the primary, direct driver of associative bond strengthening, whereas punishment exerts only an indirect, erratic, and often negligible effect on bond dissolution.

5.3 The Spread of Effect Phenomenon

During the same series of human verbal learning experiments that dismantled the symmetry of the Law of Effect, Thorndike uncovered an extraordinary, unanticipated empirical anomaly that he termed the Spread of Effect. He observed that when a specific S-R bond was rewarded by the experimenter (e.g., the declaration of “Right”), the reinforcing potency of that satisfaction did not restrict its influence exclusively to the target connection. Rather, the reinforcing energy appeared to diffuse or “spill over” across the temporal and spatial continuum of the learning task, inadvertently strengthening adjacent connections that had been explicitly punished or had occurred entirely by error.

When Thorndike plotted the retention curves of serial lists, he discovered a precise mathematical gradient of reinforcement. If response number $N$ was rewarded, the erroneous responses immediately preceding it ($N-1$, $N-2$) and immediately succeeding it ($N+1$, $N+2$) exhibited an elevated probability of repetition on subsequent test trials, directly proportional to their temporal proximity to the rewarded response:

• Connection $N-1$ showed the highest rate of accidental strengthening.
• Connection $N-2$ showed a weaker, but statistically significant boost.
• Connections $N+1$ and $N+2$ showed corresponding forward-spreading increases in bond strength.

The reinforcing effect radiated outward from the point of satisfaction like a physical wave, diminishing with temporal and ordinal distance.

For Thorndike, the Spread of Effect was monumental theoretical evidence for the purely mechanical, non-cognitive nature of reinforcement. If learning were governed by rational insight or conscious understanding, a learner who was told “Right” for item $N$ would consciously note that specific pairing, leaving the adjacent incorrect items ($N-1$ and $N+1$, which had been greeted with an explicit “Wrong”) to be expunged. The fact that the adjacent erroneous bonds were demonstrably strengthened proved that satisfaction operated automatically and physiologically, bypassing conscious discrimination. While later researchers, such as Leo Postman and Joseph Nuttin, argued that the phenomenon was an artifact of guessing biases and serial clustering strategies, Thorndike viewed the Spread of Effect as the ultimate proof that the Law of Effect operated as an automatic, blind, and somatic force of nature.

6. Primary Laws of Learning: Law of Exercise and Law of Readiness

6.1 The Law of Exercise: Use and Disuse

Alongside the Law of Effect, Thorndike originally formulated a second primary pillar of connectionism: the Law of Exercise. This law comprised two complementary sub-principles: the Law of Use and the Law of Disuse. The Law of Use stated that, other things being equal, the more frequently a connection between a situation and a response is exercised, the stronger that connection becomes. The Law of Disuse asserted that when a connection between a situation and a response is not exercised over a period of time, the strength of that connection progressively decreases, its synaptic conductance decaying according to a predictable mathematical curve of forgetting.

For over three decades, the Law of Exercise served as the primary scientific justification for rote learning, mechanical drills, and repetitive practice regimens across American public education. It was assumed that the raw, mechanical repetition of an S-R pairing was sufficient to deepen the neural groove, carving the association permanently into the brain’s conduction pathways. Millions of school children were subjected to endless, uncontextualized drills under the assumption that sheer frequency of execution was the engine of intellectual acquisition.

However, in his transformative 1932 volume The Fundamentals of Learning, Thorndike overturned this principle. Subjecting the Law of Exercise to rigorous experimental isolation, he devised experiments in which subjects repeated an action hundreds or thousands of times without receiving any feedback or knowledge of results. In one classic study, blindfolded subjects were instructed to draw a line exactly four inches long on a blank sheet of paper. They repeated this action over three thousand consecutive times. When the lines were measured, Thorndike discovered that the three-thousandth line was no closer to four inches than the first; the subjects had demonstrated zero learning. Sheer repetition had not strengthened the correct bond. Thorndike boldly struck down the Law of Exercise, concluding that mere repetition without consequence does not cause learning. The Law of Exercise was relegated to a subordinate status: exercise does not establish a bond; it merely provides the temporal opportunity for the Law of Effect to operate.

6.2 The Law of Readiness: Neuromuscular Preparations

The third primary law in Thorndike’s original triadic architecture was the Law of Readiness. Unlike modern developmental concepts of “school readiness” or cognitive maturation, Thorndike’s Law of Readiness was an uncompromisingly physiological formulation dealing with the immediate, micro-level neuromuscular status of the organism’s conduction units. It was designed to explain why an identical physical outcome could, depending on the internal somatic state of the organism, be experienced as intensely satisfying on one occasion and profoundly annoying on another.

Thorndike formulated this physiological principle through three interlocking, axiomatic conditions:

1. When a conduction unit is in a state of readiness to conduct, conduction by it is satisfying to the animal, and the animal will do things to maintain or facilitate that conduction.
2. When a conduction unit is in a state of readiness to conduct, for it not to conduct is annoying, and the animal will engage in behaviors designed to overcome the blockage or express behavioral distress.
3. When a conduction unit is not in a state of readiness to conduct, being forced to conduct is annoying, and the animal will resist or exhibit avoidance behaviors.

Under this formulation, readiness is the physiological substrate of motivation. A conduction unit is “ready” when its constituent neural pathways are chemically sensitized and primed with action potentials, typically driven by metabolic deficits or biological drives.

This law directly integrated the organism’s affective and motivational states into the mechanical execution of motor programs. If a feline subject has been deprived of food for twenty-four hours, the visceral and neural conduction units governing food search, mastication, and ingestion are in an acute state of readiness; under these conditions, reaching the food platform and consuming the fish is intensely satisfying, thereby stamping in the preceding escape bond. If the same animal is sated, those identical conduction units are not in a state of readiness; being forced to interact with the food is either neutral or annoying, and the escape connection fails to be stamped in. Readiness served as the physiological gatekeeper determining whether an environmental event possessed the functional properties of a satisfier or an annoyer.

6.3 Dynamic Interaction Among the Primary Laws

In Thorndike’s comprehensive theoretical matrix, the primary laws did not function as isolated, independent mechanisms; they operated as a dynamic, interlocking triad that governed the total trajectory of behavioral adaptation. The Law of Readiness operated as the foundational boundary condition: it set the physiological parameters of the organism, establishing the motivational topography and dictating precisely which sensory-motor outcomes could serve as effective satisfiers or annoyers in a given operational context.

Once readiness established the motivational potential, the Law of Exercise provided the necessary kinetic canvas. Through behavioral variability and repeated exposure to the environmental situation, the organism generated the requisite volume of motor trials. Exercise brought the organism into direct physical friction with the environment, producing the accidental successes that allowed the crucial S-R coordinates to be instantiated in real time and space.

Finally, the Law of Effect acted as the supreme and decisive selective filter. It intervened at the moment of behavioral execution, examining the outcome generated by exercise through the biological lens calibrated by readiness. If the response culminated in an outcome aligned with physiological readiness, the Law of Effect stamped the connection into the nervous system; if the response clashed with that readiness, the bond languished. In mathematical terms, Thorndike envisioned learning as a compounding function wherein readiness determined the sign and magnitude of the affective scalar, exercise determined the number of iterations, and the Law of Effect executed the permanent adjustment of the connection’s synaptic weight. Without readiness, effect was blind; without exercise, effect had no material to shape; but without effect, readiness and exercise were completely powerless to establish a single enduring habit.

7. Subordinate Laws and Auxiliary Principles of Connectionism

7.1 Multiple Response and Set/Attitude

Beyond the primary triad, Thorndike recognized that real-world problem-solving required auxiliary mechanisms to explain how an animal initially arrives at an adaptive response when no pre-existing bond is available. The first of these auxiliary principles is the Law of Multiple Response (or varied reaction). Thorndike argued that if an organism, upon encountering an environmental obstacle, were capable of only a single, stereotypical reflex, it would inevitably perish if that reflex failed to remove the obstacle. Learning is fundamentally contingent upon behavioral plasticity: when a prepotent response fails to produce satisfaction, the organism systematically shifts through an alternative repertoire of motor acts until an effective action is discovered.

This dynamic variability is modulated by the Law of Set, Attitude, or Disposition. Thorndike recognized that an organism does not face the external world as an undifferentiated, passive automaton; it is continuously governed by an internal “set”—a temporary or enduring neuropsychological orientation. This internal set functions as a powerful associative filter, determining not only what the animal will do, but what will satisfy or annoy it. A cat dominated by an intense “fear set” will display an entirely different array of multiple responses (freezing, cowering, defensive hissing) than a cat dominated by a “hunger set” (prowling, clawing, sniffing).

Furthermore, Thorndike expanded the Law of Set to account for enduring individual differences and cultural conditioning in human populations. A person’s comprehensive life history, vocational training, moral beliefs, and emotional temperament coalesce into an overarching mental attitude that dictates how complex environmental configurations are interpreted. The set alters the perceptual salience of incoming stimuli, prioritizing certain associative channels while dampening others, thereby ensuring that identical physical environments elicit radically different S-R sequences from individuals possessing divergent psychological dispositions.

7.2 Prepotency of Elements and Response by Analogy

As organisms ascend the phylogenetic scale, environmental stimuli cease to be simple, isolated physical events; they present as massive, multidimensional sensory arrays. To explain how an organism avoids becoming paralyzed by sensory overload, Thorndike articulated the Law of Prepotency of Elements. This principle asserts that an animal does not respond uniformly to the holistic totality of a stimulus situation. Instead, it selectively isolates and reacts to a single, highly salient, diagnostic feature—a “prepotent element”—while ignoring the vast background of extraneous, irrelevant cues.

The prepotency of elements represents Thorndike’s mechanistic account of what cognitive psychologists would later designate as selective attention and perceptual abstraction. In the puzzle box, an experienced cat learns to ignore the grain of the wood, the ambient shadows in the laboratory room, and the particular scent of the experimental table, focusing its motor response exclusively upon the prepotent visual cue of the hanging cord or the floor treadle. In humans, this capacity reaches its zenith in analytical problem solving, where a mathematician or chess grandmaster abstracts the single essential variable from an intricate problem array, linking their next intellectual response directly to that prepotent core.

When an organism encounters a novel situation for which it possesses no specific, pre-existing bonds, it relies upon the Law of Response by Analogy (or the Principle of Assimilation). Thorndike maintained that an animal will respond to an unfamiliar environment by executing behaviors that are firmly connected to previously encountered situations that share common, identical physical elements with the new array. The organism “assimilates” the new problem into old behavioral circuits. This principle laid the conceptual groundwork for what modern learning theorists term stimulus generalization, but Thorndike stripped it of cognitive mystery: transfer occurs not because the mind grasps a profound structural metaphor, but because the identical sensory components of the two situations mechanically activate the same physical conduction pathways.

7.3 Associative Shifting and Belongingness

One of Thorndike’s most sophisticated auxiliary formulations was the Law of Associative Shifting. Thorndike demonstrated that by introducing subtle, progressive, and incremental modifications to a stimulus situation while continuously maintaining the execution of a specific motor response, the response could ultimately be transferred to an entirely novel stimulus that shared nothing in common with the original trigger. In a famous demonstration, Thorndike trained an animal to stand up at the presentation of an edible treat held high. He then paired the presentation of the treat with a verbal command (“Stand!”). Over successive trials, he gradually diminished the physical prominence of the food while maintaining the vocalization, until eventually the animal would stand up instantly in response to the vocal command alone.

Associative shifting is historically and conceptually significant because it represents an early, independent American discovery of what Ivan Pavlov formalized as classical respondent conditioning. However, while Pavlov approached the phenomenon as the cortical substitution of an unconditioned stimulus by a conditioned stimulus within autonomic reflex arcs (such as salivation), Thorndike viewed associative shifting as the gradual migration of voluntary somatic motor impulses across a shifting landscape of afferent sensory terminals. It provided a powerful technique for behavioral modification, anticipating modern shaping and stimulus fading protocols.

In his later revisions during the 1930s, Thorndike introduced an auxiliary concept that represented an intriguing concession to structural organization: the Principle of Belongingness. Through extensive verbal learning studies, Thorndike observed that two items presented in temporal contiguity are not bound together with equal facility; they form a strong S-R connection much more rapidly if they naturally “belong” together in semantic, structural, or functional terms. If a subject hears the sentence “John is a butcher, Henry is a carpenter,” the association between “John” and “butcher” is established with immense strength, whereas the association between “butcher” and “Henry”—despite their immediate temporal contiguity across the comma—is exceptionally weak. By acknowledging belongingness, Thorndike admitted that temporal contiguity and mechanical consequence alone were insufficient to explain human learning; the intrinsic relational structure of the stimulus elements profoundly influences the rate and permanence of bond formation.

8. Comparative Analysis: Thorndike’s Connectionism versus Classical and Operant Conditioning

8.1 Thorndike versus Ivan Pavlov: S-R Bonds versus Conditioned Reflexes

To fully contextualize Thorndike’s historical significance, his connectionist model must be systematically contrasted with the two other dominant conditioning paradigms of the twentieth century: Ivan Pavlov’s classical conditioning and B.F. Skinner’s operant conditioning. Working contemporaneously in St. Petersburg, Russia, Ivan Pavlov developed his paradigm of the conditioned reflex, focusing on the digestive secretions of canines. The divergence between the Thorndikian and Pavlovian frameworks is profound, spanning experimental methodology, the nature of the behavioral response, and the underlying neurophysiological models.

The primary point of divergence lies in the operational role of the organism’s behavior:

• Pavlovian Classical Conditioning: The organism is largely passive and structurally immobilized. The unconditioned stimulus (UCS, e.g., meat powder) is delivered regardless of the animal’s actions; the conditioned response (CR, e.g., salivation) is an involuntary, autonomic, vegetative reflex that is elicited from the animal through stimulus-substitution. The animal’s behavior has zero causal efficacy in altering the external physical environment.
• Thorndikian Connectionism: The animal is active, freely moving, and instrumental. The delivery of the reinforcing outcome (the fish) is strictly contingent upon the animal’s execution of an active, somatic motor response (pressing the treadle). The response is not elicited by an involuntary reflex arc; it is emitted by the animal as an instrumental manipulation of the physical environment.

Their underlying neurophysiological theories diverged with equal sharpness. Pavlov conceptualized learning through the macro-dynamics of the cerebral cortex, hypothesizing systemic waves of cortical excitation and inhibition that swept across the cerebral hemispheres like physical tides. Thorndike, conversely, rejected this cortical macro-field approach, positing a micro-architectural model of discrete, localized synaptic junctions whose physical resistance was permanently modified by metabolic consequences. While Pavlov established the mechanics of involuntary visceral conditioning, Thorndike founded the science of voluntary instrumental acquisition.

8.2 Thorndike versus B.F. Skinner: The Law of Effect versus Operant Reinforcement

The direct intellectual heir to Thorndike’s Law of Effect was B.F. Skinner, the architect of radical behaviorism and operant conditioning. Skinner frequently acknowledged that his concept of the “operant” was a direct, linear descendant of Thorndike’s instrumental learning. However, Skinner executed a profound epistemological purification of the Law of Effect, stripping away the mentalistic and subjective terminology that had left Thorndike vulnerable to decades of critical assault.

Skinner attacked Thorndike’s use of the terms “satisfaction” and “annoyance,” arguing that despite Thorndike’s operational definitions, these words inevitably imported anthropomorphic, emotional connotations into the scientific lexicon. Skinner substituted the strictly empirical term reinforcer: a stimulus event that, when presented contingently following an operant response, increases the subsequent frequency or probability of that response class. Skinner eliminated all appeals to internal affective states, feelings, or speculative neural conductance. Where Thorndike spoke of a neural bond being “stamped in” by a “satisfying state of affairs,” Skinner spoke solely of a mathematically verifiable shift in the rate of response within a dynamic operant chamber.

Methodologically, Skinner revolutionized Thorndike’s discrete-trial procedure. In Thorndike’s puzzle box, an experiment was organized around isolated, episodic trials: the cat was placed in the box, escaped once, was caught by the experimenter, recorded on a chronometer, and physically replaced for trial two. This procedure created massive experimental friction and conflated the animal’s escape latency with the human experimenter’s handling speed. Skinner introduced the free-operant method via the Skinner Box: the animal was left undisturbed in the chamber for hours, free to depress a lever continuously. The primary dependent variable shifted from Thorndike’s “escape latency” to Skinner’s “rate of response,” recorded automatically on a cumulative recorder. Skinner automated and purified what Thorndike had engineered by hand.

8.3 Thorndike versus Clark Hull: Mathematical Formalization

During the 1930s and 1940s, Clark L. Hull attempted to synthesize Thorndikian S-R mechanics, Pavlovian conditioning, and logical positivism into a monumental, hypothetico-deductive mathematical system of behavior. Hull took Thorndike’s qualitative concept of the S-R bond and converted it into a formal, mathematically defined intervening variable termed habit strength, designated symbolically as $_sH_R$.

Hull sought to formalize what Thorndike had described descriptively. In Hull’s system, habit strength grew as a continuous, logarithmic function of the number of reinforced trials ($N$), governed by the mathematical equation:
$$_sH_R = M(1 – 10^{-iN})$$
where $M$ represents the physiological maximum of the habit, and $i$ represents an empirical constant. Furthermore, Hull provided a physiological reinterpretation of Thorndike’s “satisfaction” via his famous drive-reduction hypothesis. For Hull, satisfaction was not an ambiguous affective state, but the biological reduction of a homeostatic tissue deficit: food consumption was reinforcing precisely because it lowered the biological drive ($D$) of hunger, reducing systemic metabolic strain.

While Thorndike was an empirical pragmatist who resisted hyper-formalized deductive systems, Hull represented the apex of behavioral axiomatic architecture. Hull introduced intervening organismic variables ($O$) between the environmental stimulus ($S$) and the motor output ($R$), transforming Thorndike’s direct S-R bond into a mediated $S\text{-}O\text{-}R$ system. Yet, despite Hull’s complex equations involving reaction potential ($_sE_R$), reactive inhibition ($I_R$), and conditioned inhibition ($_{s}I_R$), the fundamental core of his system remained unaltered Thorndikian connectionism: the progressive strengthening of an associative path through biological consequence.

9. Educational Connectionism: Instructional Design, Arithmetic, and Curriculum Reform

9.1 The Doctrine of Identical Elements in Transfer of Learning

Edward Thorndike’s transition to Teachers College at Columbia University marked the beginning of a profound revolution in American pedagogical theory. At the turn of the twentieth century, American education was completely dominated by the doctrine of formal discipline, an educational philosophy rooted in classical faculty psychology. This doctrine maintained that the human mind was comprised of general, distinct faculties—such as “Reasoning,” “Memory,” “Will,” and “Attention”—which functioned precisely like physical muscles. Proponents argued that by forcing students to study exceedingly difficult, abstract subjects that possessed zero direct vocational utility—most notably Latin, ancient Greek, and Euclidean geometry—the general mental faculties would be systematically strengthened, transferring seamlessly to any practical domain in later life.

In a historic series of empirical investigations conducted with Robert S. Woodworth in 1901, titled “The Influence of Improvement in One Mental Function upon the Efficiency of Other Functions,” Thorndike delivered a death blow to formal discipline. Thorndike and Woodworth subjected adult learners to rigorous training regimes designed to improve their capacity to estimate the area of small rectangles (measuring between 10 and 100 square centimeters). After the subjects demonstrated profound mastery in this specific task, the researchers tested their ability to estimate the areas of rectangles of different sizes and shapes, as well as triangles and trapezoids. The results were shocking: despite massive improvement on the trained rectangles, the subjects demonstrated virtually zero improvement—and in many cases, outright regression—when estimating the areas of novel geometric figures.

From these data, Thorndike formulated the Doctrine of Identical Elements. This doctrine asserted that transfer of learning between Task A and Task B occurs if, and only if, Task A and Task B share specific, identical constituent S-R bonds. There is no broad, general strengthening of an indivisible “mental muscle.” Improvement in one function alters another only to the extent that the two functions employ the exact same neuromuscular connections, the same objective physical information, and the same procedural habits. The Doctrine of Identical Elements shattered the classical curriculum, paving the way for vocational education, direct domain-specific instruction, and modern standardized testing systems calibrated to measure specific educational competencies rather than nebulous intellectual faculties.

9.2 Re-engineering Pedagogy: The Psychology of Arithmetic and Language

Armed with connectionist principles, Thorndike systematically dismantled and rebuilt the curricula of elementary mathematics and language arts. In his monumental 1922 treatise, The Psychology of Arithmetic, Thorndike argued that mathematics was not an abstract, intuitive art, but an intricate, hierarchical lattice of hundreds of thousands of discrete, specific S-R bonds. To teach arithmetic successfully, the educator could not rely on sweeping conceptual explanations; one had to deconstruct the discipline into its atomic associative components, establish each bond independently, and then sequence their historical integration with engineering precision.

Thorndike demonstrated that a seemingly simple operation, such as two-digit column addition with carrying, was not a single cognitive skill, but an interconnected network of distinct, microscopic bonds:

• Knowing that addition proceeds from right to left.
• Binding the visual symbol “7 + 8” to the automated motor response “write 5, carry 1”.
• Holding the carried “1” in short-term storage while executing the subsequent addition of the left column.
• Adding the carried digit to the intermediate sum.

Thorndike conducted comprehensive audits of existing arithmetic textbooks, exposing that they frequently introduced massive cognitive gaps—introducing advanced operations without establishing the requisite lower-order S-R bonds—or forced students to drill on archaic, socially useless skills, such as calculating complex fractional dividends for currencies that no longer existed. He designed new textbooks structured around deliberate practice, graduated difficulty, and immediate feedback.

Thorndike executed an equally sweeping transformation in reading and literacy instruction through the 1921 publication of The Teacher’s Word Book. Prior to this work, reading primers presented vocabulary arbitrarily, often exposing young children to exceedingly rare, literary words while ignoring common terms. Thorndike compiled the first massive, objective frequency count of the English language, reading through millions of pages of literature, legal documents, newspapers, and textbooks to identify the most frequently occurring words in written communication. He isolated the first 10,000 most common words, ranking them by empirical frequency. This allowed educators to construct basal readers that introduced vocabulary systematically, ensuring that students established rock-solid S-R bonds with high-frequency foundational words before encountering rare, complex terminology, a paradigm that remains the architectural core of early childhood literacy curricula to this day.

9.3 Classroom Management and the Demise of Punitive Discipline

The pedagogical consequences of Thorndike’s 1930s revision of the Law of Effect were nowhere more visible than in the transformation of classroom management and disciplinary practices. For centuries, traditional Western education had relied heavily upon aversive control: physical corporal punishment, verbal humiliation, public shaming, retention, and punitive homework assignments were standard operating procedures designed to coerce compliance and expunge errors. This pedagogical philosophy operated under the unexamined assumption of the symmetrical Law of Effect: punishment was believed to actively stamp out indolence, misbehavior, and cognitive mistakes.

Thorndike’s experimental demonstration that punishment fails to weaken S-R bonds provided a devastating empirical indictment of punitive classroom cultures. Thorndike demonstrated that when a child is punished for giving an incorrect answer, the punitive consequence does not magically insert the correct cognitive bond into the child’s brain. Rather, the aversive stimulus generates an overwhelming emotional disruption—fear, resentment, anxiety, and task avoidance—which completely paralyzes the child’s behavioral variability, preventing the Law of Readiness and the Law of Use from operating effectively. In many instances, the public delivery of punishment merely fixated the child’s attention upon the error, causing the incorrect bond to be exercised and inadvertently stamped in.

Thorndike advocated for a complete restructuring of the instructional environment, demanding the systemic replacement of aversive control with positive reinforcement. Classroom architecture had to be re-engineered so that learning tasks were broken down into small, highly manageable increments that guaranteed high rates of initial student success. Immediate, targeted praise, public acknowledgment, and tangible educational rewards had to follow every correct execution of a target bond. Educators were instructed to ignore errors whenever possible, redirecting the student’s behavior immediately toward the execution of the correct alternative response. Thorndike became an early champion of individualized, self-paced instructional sequences, insisting that educational mastery could only occur when every child was permitted to progress through the curriculum at an associative acquisition rate calibrated to their unique neurological constitution.

10. Methodological Critique, Experimental Controversies, and Gestalt Counterarguments

10.1 The Gestalt Challenge: Wolfgang Köhler and Insight Learning

Despite its widespread institutional adoption, Thorndike’s connectionism faced fierce theoretical and empirical challenges. The most profound and intellectually devastating counter-offensive was mounted by the Berlin school of Gestalt psychology, spearheaded by Wolfgang Köhler. Trapped on the island of Tenerife during the First World War, Köhler conducted a series of classic experiments on chimpanzees, culminating in his 1925 landmark publication, The Mentality of Apes. Köhler’s findings presented an empirical portrait of animal cognition that appeared completely irreconcilable with Thorndike’s blind, mechanical puzzle-box studies.

Köhler presented his subjects—most famously a brilliant ape named Sultan—with complex, multi-component problems that could not be solved by a single, immediate motor act. In the classic “two-stick problem,” Sultan was placed in a large cage with a banana positioned far beyond the wire perimeter, out of reach of any single tool. Inside the cage were two hollow bamboo poles of different diameters, neither of which was long enough on its own to reach the fruit. Sultan initially attempted to reach the banana with one stick; failing, he discarded it. Rather than engaging in the frantic, continuous, blind thrashing characteristic of Thorndike’s cats, Sultan sat quietly on a wooden box, visually surveying the spatial arrangement of the room.

Suddenly, Sultan leaped up, retrieved both sticks, inserted the thinner bamboo pole into the hollow opening of the thicker pole to create an elongated composite implement, and immediately dragged the banana into the cage. The solution was executed smoothly, without trial and error, without accidental motor success, and without gradual learning curves. Köhler designated this phenomenon insight learning (Einsicht), defining it as the sudden, holistic cognitive restructuring of the perceptual field, leading to an immediate grasp of the essential structural relationships within a problem.

Köhler launched a blistering critique against the ecological and methodological validity of Thorndike’s puzzle box. He argued that the puzzle box was a methodological trap that artificially compelled the animal to behave like an idiot. The vital mechanical connections of the box—the external cords, the hidden pulleys, the spring-loaded bolts—were deliberately concealed from the animal’s perceptual field. The cat could not possibly grasp the functional structure of the apparatus because it was physically impossible to see the entire mechanism simultaneously. By placing the animal in a state of high panic and starving it into extreme biological drive, while simultaneously blinding it to the structural relations of the problem, Thorndike had eliminated all possibilities for intelligent, perceptual restructuring. The jagged, gradual learning curve was not an intrinsic property of animal learning, Köhler argued, but an experimental artifact of an impoverished and claustrophobic apparatus. Thorndike defended his connectionism by arguing that Sultan’s sudden “insight” was merely the rapid covert execution of previously formed, lower-order S-R habits acquired during the ape’s early life outside the experimental chamber.

10.2 The Subjectivity of ‘Satisfaction’ and Circularity Charges

From the philosophical and behaviorist camps, Thorndike was subjected to a relentless barrage of criticism targeting the conceptual coherence of the Law of Effect itself. The most enduring of these challenges was the charge of vicious circularity (tautology), a critique famously analyzed by twentieth-century philosophers of science and learning theorists such as Leo Postman and Paul Meehl. The critique attacked the operational integrity of Thorndike’s primary definitions:

1. Premise 1: What is a satisfier? A satisfier is an event that an animal does nothing to avoid, often doing things to attain and preserve it.
2. Premise 2: How do we know an event has reinforced or strengthened an S-R bond? Because the animal did things to attain it, and the response recurred.
3. The Tautology: The Law of Effect asserts that responses followed by satisfiers are strengthened; but satisfiers are defined exclusively as those events that strengthen responses!

If the definition of the independent variable (satisfier) is completely conflated with the measurement of the dependent variable (bond strengthening), the Law of Effect ceases to be an empirically testable scientific hypothesis; it degenerates into a circular, analytic truism that is fundamentally unfalsifiable.

Radical behaviorists, led by Watson, were equally unsparing in their attacks on the terms “satisfaction” and “annoyance.” They argued that these constructs were mentalistic contaminants that smuggled subjective hedonism into what purported to be an objective materialist science. Watson pointed out that an outside observer could never directly measure a feline “satisfaction”; one could only observe physical movements. To attribute the physical alteration of a neural conduction path to an emotional state was, in Watson’s eyes, a dangerous retreat into prescientific animism. While Thorndike labored extensively to operationalize these terms through objective approach and avoidance criteria, the circularity critique persisted, ultimately prompting B.F. Skinner to discard the concept of satisfaction entirely in favor of the strictly functional, empirical definition of a reinforcer.

10.3 Ecological Validity and Animal Ethology

With the mid-twentieth-century emergence of modern ethology, pioneered by Nikolaas Tinbergen and Konrad Lorenz, Thorndike’s connectionism faced a profound ecological critique. Ethologists demonstrated that animal behavior could not be understood by treating non-human subjects as arbitrary, interchangeable biological blank slates dropped into unnatural, wooden torture chambers. Every animal species possesses an evolutionarily canalized behavioral repertoire—instinctive response tendencies, sensory biases, and species-specific defense reactions—that evolved to resolve specific ecological dilemmas in the wild.

When Thorndike dropped a domestic cat into a claustrophobic, wire-slatted box, he was not observing a neutral learning machine; he was observing an animal subjected to extreme existential terror and severe confinement stress. The frantic biting, clawing, and thrashing that Thorndike chronicled as “trial-and-error” were, in reality, hardwired species-specific defense reactions (SSDRs), an evolutionary panic script activated by predatory trapping or entrapment. These instinctual emergency programs actively interfered with the acquisition of subtle, novel motor habits.

Decades later, in 1961, Keller and Marian Breland, former students of B.F. Skinner, published their famous paper, “The Misbehavior of Organisms,” demonstrating that evolutionary adaptations systematically override and disrupt operant and instrumental conditioning, a phenomenon they termed “instinctive drift.” Furthermore, Edward Tolman’s groundbreaking work at the University of California, Berkeley, established that rodents navigate physical space not through linear, chain-like sequences of localized S-R muscular twitches, but through the construction of holistic, internal cognitive maps. Animals, it turned out, learned about spatial geography even in the total absence of food reinforcement (latent learning). Thorndike’s assertion that all animal cognition could be reduced to the blind stamping in of isolated S-R bonds was exposed as an oversimplified laboratory abstraction that largely ignored the immense cognitive and evolutionary richness of the natural world.

11. Neurobiological Parallels: Historical S-R Bonds and Modern Neural Network Connectionism

11.1 From Thorndike to Donald Hebb: The Synaptic Interface

One of the most extraordinary developments in the history of psychology is the extent to which Edward Thorndike’s intuitive, early twentieth-century speculative neurophysiology anticipated the physical breakthroughs of modern cellular neurobiology. In his 1949 foundational masterwork, The Organization of Behavior, Canadian neuropsychologist Donald O. Hebb sought to bridge the chasm between behavioral learning theories and cortical neuroanatomy. Hebb formulated what has become the foundational postulate of cellular neuroscience, universally known as the Hebb Synapse:

“When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A’s efficiency, as one of the cells firing B, is increased.”

Often popularized as the aphorism “neurons that fire together, wire together,” Hebb’s learning rule is the precise cellular and physiological instantiation of Thorndike’s Law of Exercise and Law of Effect. Thorndike’s historical assertion that learning represents the reduction of synaptic resistance within specific conduction units found its definitive physical verification in 1973 with the discovery of Long-Term Potentiation (LTP) by Terje Lømo and Timothy Bliss in the mammalian hippocampus. LTP demonstrated that high-frequency electrical stimulation of a presynaptic neuron induces a stable, long-lasting enhancement in signal transmission across the synapse to the postsynaptic neuron.

At the molecular level, LTP operates via the activation of NMDA (N-methyl-D-aspartate) and AMPA glutamate receptors. When a strong postsynaptic depolarization coincides with presynaptic glutamate release, magnesium ions ($Mg^{2+}$) are expelled from the NMDA receptor channel pore, permitting an influx of calcium ions ($Ca^{2+}$) into the postsynaptic dendritic spine. This intracellular calcium cascade triggers the insertion of additional AMPA receptors into the postsynaptic density, physically strengthening the synaptic connection. Furthermore, modern neuroscience has identified retrograde neurotransmission—via agents such as nitric oxide and endocannabinoids—whereby the postsynaptic neuron sends a biochemical signal backward across the synaptic cleft to enhance presynaptic neurotransmitter release. This retrograde transmission provides the biophysical solution to the retroaction paradox that plagued Thorndike: an outcome (depolarization and calcium influx) reaches backward across the junction to permanently modify the antecedent conduction unit.

11.2 Modern Computational Connectionism (PDP Models)

During the cognitive revolution of the 1950s and 1960s, Thorndike’s behavioral connectionism was largely marginalized by the dominant symbolic paradigm, which conceptualized the human mind as a serial, digital, von Neumann computer executing formal, logical rules over discrete propositional symbols. However, in the mid-1980s, an intellectual revolution swept cognitive science with the publication of the two-volume foundational treatise Parallel Distributed Processing: Explorations in the Microstructure of Cognition (1986), authored by David E. Rumelhart, James L. McClelland, and the PDP Research Group. This movement resurrected Thorndike’s historical terminology, christening itself Connectionism.

While sharing Thorndike’s commitment to network-based, associative mechanics, computational connectionism operated at an unprecedented level of mathematical and computational sophistication. The divergence and convergence between the two paradigms can be modeled across their structural components:

Structural Dimension	Thorndikian Behavioral Connectionism	Computational PDP Connectionism
Network Node	Macroscopic, observable physical entity (discrete Stimulus or discrete Motor Act).	Microscopic, sub-symbolic artificial neuron (activation value within a continuous vector).
Connection Architecture	Direct, single-layer neuromuscular paths ($S \rightarrow R$).	Deep, multi-layered architectures (Input Layer $\rightarrow$ Hidden Layers $\rightarrow$ Output Layer).
Information Storage	Localized, discrete S-R bonds stored at individual anatomical reflex points.	Fully distributed representation; concepts exist as global weight matrices across all connections.
Learning Mechanism	Biochemical stamping in governed by the biological Law of Effect.	Mathematical gradient descent algorithms, specifically the backpropagation algorithm.

The mathematical core of modern computational connectionism is the backpropagation algorithm, formulated by Rumelhart, Hinton, and Williams. Backpropagation is the rigorous mathematical realization of Thorndike’s Law of Effect operating across deep, hidden layers of representation. During a forward pass, an input vector generates an output activation pattern. An objective loss function computes the mathematical error (the discrepancy between the network’s generated output and the desired target output). This error signal is then propagated backward through the network layers using the chain rule of differential calculus, calculating the partial derivative of the error with respect to each individual connection weight:

$$\frac{\partial E}{\partial w_{ij}}$$

Each weight is then adjusted via gradient descent to minimize overall network error. Just as Thorndike’s puzzle box iteratively reduced the temporal escape latency through the stamping in and out of bonds, backpropagation iteratively drives down the global error surface of an artificial neural network, achieving speech recognition, computer vision, and linguistic translation through the continuous adjustment of millions of associative weights.

11.3 Reinforcement Learning in Modern Artificial Intelligence

The most direct, explicit, and monumental realization of Thorndike’s Law of Effect in contemporary computing resides within the discipline of Reinforcement Learning (RL), a subfield of artificial intelligence that powers autonomous robotics, algorithmic game engines, and modern large language models. In their definitive masterwork, Reinforcement Learning: An Introduction, computer scientists Richard S. Sutton and Andrew G. Barto explicitly identify Edward Thorndike’s 1898 dissertation as the foundational intellectual origin of the entire field, dedicating significant historical exposition to the Law of Effect.

In modern reinforcement learning, an autonomous agent interacts with an environment modeled as a Markov Decision Process (MDP). At each discrete time step $t$, the agent perceives the environmental state $S_t in \mathcal{S}$ and executes an action $A_t in \mathcal{A}$. The environment transitions to a new state $S_{t+1}$ and emits a scalar numerical feedback signal: the reward $R_{t+1}$. The fundamental goal of the agent is to discover an optimal behavioral policy $\pi(a|s)$ that maximizes the expected cumulative discounted future reward:

$$G_t = \sum_{k=0}^{\infty} \gamma^k R_{t+k+1}$$

This mathematical objective is the precise algorithmic formalization of Thorndike’s cat attempting to maximize “satisfaction” over successive iterations.

The computational engine driving this adaptation is the Temporal Difference (TD) learning algorithm, and its action-control offshoots such as Q-learning and SARSA. TD learning updates the value of executing a specific action within a given state by calculating the Reward Prediction Error ($\delta_t$):

$$\delta_t = R_{t+1} + \gamma V(S_{t+1}) – V(S_t)$$

The prediction error represents the mathematical difference between the satisfaction the agent actually experienced and the satisfaction it expected to receive. The state-action value $Q(S_t, A_t)$ is then adjusted proportionally to this error signal:

$$Q(S_t, A_t) \leftarrow Q(S_t, A_t) + \alpha \delta_t$$

Remarkably, modern computational neuroscience has demonstrated that this exact algorithm is physically instantiated within the mammalian brain. In landmark neurophysiological studies conducted by Wolfram Schultz, Peter Dayan, and P. Read Montague, recording electrodes inserted into the ventral tegmental area (VTA) and the substantia nigra of primates revealed that midbrain dopaminergic neurons do not fire in response to raw pleasure; they fire in exact correspondence to the mathematical reward prediction error ($\delta_t$). A burst of dopamine stamps in corticostriatal synaptic weights when an outcome is better than expected, while a pause in dopaminergic firing signals an annoying, worse-than-expected outcome. When modern deep reinforcement learning agents, such as DeepMind’s AlphaGo or DQN, master complex virtual environments, they are operating as hyper-computational incarnations of Thorndike’s felines, recursively traversing the Law of Effect through the automated mathematics of reward prediction errors.

12. Enduring Legacy and Pedagogical Implications in Contemporary Psychology

12.1 Thorndike’s Place in the Behavioral Canon

Edward Lee Thorndike stands as a monumental figure in the history of science, serving as the essential intellectual bridge that transported psychology across the threshold from nineteenth-century philosophical introspection to twentieth-century empirical natural science. His historical footprint is vast: he served as president of the American Psychological Association in 1912, was elected president of the American Association for the Advancement of Science in 1934, and published over five hundred monographs, standardized tests, and empirical articles. His quantitative dedication established psychometrics as a central pillar of educational psychology, institutionalizing mental measurement across global educational systems.

Yet, an honest historical evaluation of Thorndike’s legacy requires confronting his participation in the darker sociopolitical movements of his era. Thorndike was an unapologetic and fervent proponent of the American eugenics movement. Deeply committed to biological determinism, Thorndike believed that human intellectual capacity, moral integrity, and social efficiency were almost entirely hardwired by genetics, viewing environmental and educational interventions as capable of only marginally polishing the biological limits established at conception. In works such as Human Nature and the Social Order (1940), he advocated for selective breeding, intelligence-based social stratification, and the tracking of educational opportunities based on standardized IQ metrics. He argued that society should invest its resources primarily in the intellectual elite, while relegating those with lower measured psychometric capacity to narrow vocational tracks. This rigid determinism led to decades of educational tracking that disproportionately marginalized racial minorities and lower socioeconomic classes, a historical legacy that has prompted modern educational institutions to critically re-evaluate his institutional honors.

Nevertheless, Thorndike’s core scientific contribution—the connectionist bond and the Law of Effect—remains an invariant principle of cross-species behavioral adaptation. While his radical eugenicist claims have been thoroughly discredited by modern behavioral genetics, epigenetics, and developmental systems theory, his formulation of consequence-driven learning remains untouched. He stripped psychology of prescientific animism, replacing mystical conceptions of the will with verifiable, objective mechanics that established a permanent, cross-phylogenetic behavioral baseline.

12.2 Contemporary Educational Architecture and EdTech

In the twenty-first century, the architectural core of the global educational technology (EdTech) industry does not derive from the radical constructivist theories of Jean Piaget or the sociocultural models of Lev Vygotsky; it is built upon the connectionist engineering of Edward Thorndike. The contemporary proliferation of algorithmic micro-learning software, computerized adaptive testing, and gamified educational applications represents the industrial-scale automation of discrete S-R bond formation.

Consider the design architecture of modern learning platforms such as Duolingo, Khan Academy, or continuous medical education apps:

• Deconstruction into Atomic Units: Curricula are systematically decomposed into minute, modular, highly localized S-R associations (a single vocabulary translation, an isolated geometric formula, a targeted grammar rule), executing Thorndike’s prescription in The Psychology of Arithmetic.
• Immediate Consequence Delivery: The student inputs a response and is greeted within milliseconds by an unmistakable algorithmic consequence: an affirmative musical chime accompanied by a green interface (“Right!”), or a dull vibrational buzz accompanied by a red flag (“Wrong!”).
• Gamified Dopaminergic Feedback: Experience points (XP), daily completion streaks, badge acquisitions, and progressive leveling mechanics function as computerized proxies for Thorndike’s “satisfying states of affairs,” deliberately designed to stamp in target associations through recursive exposure.
• Algorithmic Pacing: Modern spaced-repetition algorithms (such as the SuperMemo SM-2 algorithm or Anki) mathematically operationalize the revised Law of Exercise and the Law of Disuse, presenting a stimulus bond for review at the precise mathematical instant before memory decay drops below retrieval thresholds.

Furthermore, the modern institutional transition toward Competency-Based Education (CBE) is the direct administrative realization of the Doctrine of Identical Elements. CBE rejects the traditional model of passive “seat time,” requiring students to demonstrate verifiable, quantitative mastery over granular, specific learning objectives before advancing to the next pedagogical tier. While modern educational theorists frequently critique this architecture as overly reductionist, mechanical, and hostile to holistic critical inquiry, the algorithmic efficiency of connectionist instructional design remains unmatched in the rapid, scalable acquisition of foundational procedural competencies.

12.3 Epistemological Synthesis: The Permanence of Connectionist Principles

More than a century after Edward Thorndike closed the latch on his wooden puzzle boxes in the basement of William James’s Cambridge home, the fundamental dialectic between connectionist mechanics and cognitive constructivism has reached a mature, integrated epistemological synthesis. The historical conflicts between Thorndike and Köhler, or between Skinner and Chomsky, are no longer viewed as irreconcilable, mutually exclusive explanations of human mental life. Instead, contemporary cognitive science resolves this tension through the framework of dual-process theories of cognition, famously synthesized by Daniel Kahneman in Thinking, Fast and Slow.

Under this dual-process taxonomy:

• System 1 (Fast, Automatic, Heuristic): This system operates almost exclusively under the computational architecture of Thorndikian connectionism. It is sub-symbolic, associative, driven by continuous pattern matching, heavily calibrated by reinforcement histories, and computationally parallel. It governs rapid motor execution, habitual choice, athletic reflexes, emotional threat detection, and automated procedural routines. The neurobiological substrate of System 1 is anchored precisely within the basal ganglia, the cerebellum, and the dopaminergic striatal pathways that execute the Law of Effect.
• System 2 (Slow, Deliberative, Symbolic): This system embodies the insight, planning, and conscious deliberation championed by Köhler and the cognitive constructivists. It is serial, rule-governed, linguistically mediated, mentally taxing, and localized primarily within the human prefrontal cortex. System 2 allows an individual to override automated System 1 habits, mentally simulate hypothetical futures, and restructure complex problem fields through abstract symbolic manipulation.

Far from being rendered obsolete by the rise of cognitive science, Thorndike’s connectionism provides the indispensable, sub-symbolic foundation upon which higher-order human cognition is erected. In modern behavioral pharmacology, clinical addiction treatment protocols operate by systematically identifying and severing the potent S-R-Reinforcement loops that bind environmental triggers to compulsive drug-seeking motor acts. In modern public health, “habit architecture” and behavioral “nudges” successfully reshape societal wellness by modifying the immediate micro-consequences of daily human behavior. Edward Thorndike’s immortal vision—a purely empirical science of behavior constructed brick by associative brick through the deterministic power of consequence—has stood the test of deep historical time. In the cellular mechanics of our synapses, in the digital architecture of our educational software, and in the deep reinforcement learning algorithms that navigate our computational future, the Law of Effect continues to operate as an invariant law of intelligent life.

References

• Bain, A. (1855). The Senses and the Intellect. John W. Parker and Son. https://archive.org/details/sensesandintelle00bainuoft
• Darwin, C. (1871). The Descent of Man, and Selection in Relation to Sex. John Murray. https://www.darwin-online.org.uk/
• Hartley, D. (1749). Observations on Man, His Frame, His Duty, and His Expectations. S. Richardson. https://plato.stanford.edu/entries/hartley/
• Hebb, D. O. (1949). The Organization of Behavior: A Neuropsychological Theory. John Wiley & Sons. https://archive.org/details/organizationofbe00hebb
• Hull, C. L. (1943). Principles of Behavior: An Introduction to Behavior Theory. Appleton-Century-Crofts. https://archive.org/details/principlesofbeha00hull
• Hume, D. (1739). A Treatise of Human Nature. John Noon. https://plato.stanford.edu/entries/hume/
• James, W. (1890). The Principles of Psychology. Henry Holt and Company. https://psychclassics.yorku.ca/James/Principles/
• Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
• Köhler, W. (1925). The Mentality of Apes (E. Winter, Trans.). Harcourt, Brace & Company. https://archive.org/details/mentalityofapes00kh
• Locke, J. (1690). An Essay Concerning Human Understanding. Thomas Basset. https://plato.stanford.edu/entries/locke/
• Morgan, C. L. (1894). An Introduction to Comparative Psychology. Walter Scott. https://archive.org/details/anintroductiont01morggoog
• Pavlov, I. P. (1927). Conditioned Reflexes: An Investigation of the Physiological Activity of the Cerebral Cortex (G. V. Anrep, Trans.). Oxford University Press. https://psychclassics.yorku.ca/Pavlov/
• Postman, L. (1947). The history and present status of the Law of Effect. Psychological Bulletin, 44(6), 489–563. https://doi.org/10.1037/h0054707
• Romanes, G. J. (1882). Animal Intelligence. Kegan Paul, Trench, & Co. https://archive.org/details/animalintell00roma
• Rumelhart, D. E., McClelland, J. L., & the PDP Research Group. (1986). Parallel Distributed Processing: Explorations in the Microstructure of Cognition (Vols. 1–2). MIT Press. https://mitpress.mit.edu/9780262680530/parallel-distributed-processing/
• Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. Science, 275(5306), 1593–1599. https://doi.org/10.1126/science.275.5306.1593
• Skinner, B. F. (1938). The Behavior of Organisms: An Experimental Analysis. Appleton-Century-Crofts.
• Skinner, B. F. (1953). Science and Human Behavior. Macmillan. https://www.bfskinner.org/
• Sutton, R. S., & Barto, A. G. (2018). Reinforcement Learning: An Introduction (2nd ed.). MIT Press. http://incompleteideas.net/book/the-book-2nd.html
• Thorndike, E. L. (1898). Animal Intelligence: An Experimental Study of the Associative Processes in Animals (Psychological Review Monograph Supplements, No. 8). Macmillan. https://psychclassics.yorku.ca/Thorndike/Animal/
• Thorndike, E. L. (1905). The Elements of Psychology. A. G. Seiler. https://archive.org/details/elementsofpsycho00thoruoft
• Thorndike, E. L. (1911). Animal Intelligence: Experimental Studies. Macmillan. https://archive.org/details/animalintell00thor
• Thorndike, E. L. (1913). Educational Psychology: The Psychology of Learning (Vol. 2). Teachers College, Columbia University. https://archive.org/details/educationalpsych02thor
• Thorndike, E. L. (1921). The Teacher’s Word Book. Teachers College, Columbia University. https://archive.org/details/teacherswordbook00thor
• Thorndike, E. L. (1922). The Psychology of Arithmetic. Macmillan. https://archive.org/details/psychologyofarit00thorrich
• Thorndike, E. L. (1931). Human Learning. Century Company.
• Thorndike, E. L. (1932). The Fundamentals of Learning. Teachers College, Columbia University. https://archive.org/details/fundamentalsofle00thor
• Thorndike, E. L. (1935). The Psychology of Wants, Interests, and Attitudes. D. Appleton-Century Company.
• Thorndike, E. L. (1940). Human Nature and the Social Order. Macmillan. https://archive.org/details/humannaturesocia00thor
• Thorndike, E. L., & Woodworth, R. S. (1901). The influence of improvement in one mental function upon the efficiency of other functions. Psychological Review, 8(3), 247–261. https://doi.org/10.1037/h0074898
• Watson, J. B. (1913). Psychology as the behaviorist views it. Psychological Review, 20(2), 158–177. https://doi.org/10.1037/h0074428