The Transposition Experiment (Relational vs. Absolute Learning) – Kenneth Spence

In the history of experimental psychology and comparative cognition, few empirical puzzles have generated as fierce an epistemological battle as the phenomenon of stimulus transposition. Arising at the intersection of European phenomenological inquiry and American quantitative behaviorism, the transposition paradigm interrogated a deceptively simple question: when an organism learns to choose between two physical stimuli based on a distinguishing physical continuum—such as selecting a brighter square over a duller one, or a larger disc over a smaller one—what has it actually internalized? Has the subject formed an associative bond with the specific, absolute physical energy values impinging upon its sensory receptors, or has it apprehended an abstract relational rule governing the structural configuration of the stimuli?

For early twentieth-century Gestalt psychology, spearheaded by Wolfgang Köhler, transposition was hailed as incontrovertible proof that animal perception is inherently holistic, dynamic, and relational. According to Köhler, organisms do not perceive isolated sensory elements; rather, they experience organized perceptual totalities, or Gestalten. If a chimpanzee or a chicken trained to choose a stimulus of size 100 over size 60 subsequently chooses a novel stimulus of size 160 over the familiar 100, the organism is demonstrating that it learned the relation “choose the larger,” completely bypassing the absolute physical traces of reinforcement. This interpretation posed a radical threat to the mechanistic, atomistic architectures of behaviorism, which sought to reduce all psychological phenomena to associative bonds forged through reinforcement across physical stimulus-response dimensions.

Enter Kenneth Wartenbee Spence, an intellectual titan of American neobehaviorism and the premier quantitative theorist of the Hullian tradition. In a brilliant pair of theoretical papers published in 1936 and 1937, Spence achieved what many considered an intellectual impossibility: he formulated a purely mechanistic, elementistic, and absolute mathematical model of stimulus-response learning that accounted for the emergence of relational choice behavior without conceding a shred of cognitive abstraction to the organism. By postulating overlapping algebraic gradients of excitatory and inhibitory potential across a continuous physical metric, Spence demonstrated that relational transposition was not only compatible with absolute associationism, but was its inevitable deductive consequence under specific boundary conditions. This treatise examines Spence’s historic solution, its empirical confirmations and anomalies, and its enduring relevance to contemporary cognitive neuroscience, developmental psychology, and computational modeling.

1. Introduction to the Transposition Paradigm and the Core Debate

1.1 Conceptual Definition and the Fundamental Phenomenon

The transposition paradigm is an experimental procedure designed to determine whether discrimination learning is anchored to absolute physical stimulus values or to the invariant relational properties existing between those stimuli. In a standard two-choice simultaneous discrimination protocol, an animal is presented with two stimulus objects that differ along a single physical dimension, such as surface area, spectral luminance, auditory frequency, or reflectance. One stimulus, designated as the positive stimulus (S+), is consistently paired with an appetitive reinforcer, such as a food reward. The alternative stimulus, designated as the negative stimulus (S-), is paired with non-reinforcement or, in certain experimental designs, mild punishment.

Once the experimental subject achieves a stable criterion of differential responding—consistently approaching S+ and avoiding S-—the critical transposition test phase commences. In this test phase, the experimenter introduces a novel configuration of stimuli. In the classic test design, the previously reinforced stimulus (S+) is paired with a completely novel stimulus (S++) that lies further along the physical continuum in the direction established during training. For example, if S- had a surface area of 60 square centimeters and S+ possessed an area of 100 square centimeters, the test pairing might consist of S+ (100 square centimeters) paired with S++ (160 square centimeters).

The operational distinction between absolute and relational learning becomes strikingly manifest in the subject’s behavior during this critical choice trial. If the animal operates strictly under absolute learning, it should select S+, because S+ possesses an established, direct history of positive reinforcement, whereas S++ is entirely novel and carries no associative reinforcement history. Conversely, if the animal has learned the relational principle—specifically, the rule to “choose the larger stimulus”—it should transpose this relational rule to the novel stimulus pair and systematically choose S++ over S+, despite the fact that S++ has never been reinforced in the subject’s lifetime.

Historically, the transposition paradigm emerged from early comparative investigations into avian and primate sensory capacities in the late nineteenth and early twentieth centuries. However, it quickly escalated into one of the most critical theoretical battlegrounds in the history of psychology. The phenomenon was not merely an empirical curiosity about how animals discriminate shapes or shades of gray; it was a conceptual stress test for the viability of associationist psychology itself. The empirical observation that animals frequently choose the novel, never-reinforced stimulus over the reliably reinforced familiar stimulus appeared to strike at the foundational premises of Thorndikian and Pavlovian conditioning.

1.2 The Epistemological Conflict: Gestalt Holism versus Neobehaviorist Atomism

The theoretical collision catalyzed by the transposition paradigm exposed a profound epistemological divide regarding the nature of perception, learning, and mental architecture. On one side stood Gestalt holism, an intellectual movement originating in Germany that rejected the atomistic reductionism of Anglo-American associationism and Wundtian structuralism. The Gestalt theorists contended that mental experience and behavioral orientation cannot be decomposed into a mosaic of elementary sensory inputs connected by mechanical associative glue. Instead, they posited that perception is fundamentally organized into emergent structures, where the whole is qualitatively different from, and psychologically prior to, the sum of its isolated parts.

Within this Gestalt framework, relational learning was viewed as prima facie evidence for the perceptual grasp of structural invariants. When an organism confronts two visual stimuli, it does not record two distinct photometric sensations and link one to a motor execution; rather, it apprehends a visual field organized by a perceptual gradient or ratio. The cognitive implications of this view were profound: even humble organisms like domestic fowl were presumed to possess an inherent capacity for perceptual abstraction, responding to dynamic relational properties rather than static physical values. Learning, according to Gestalt theory, involves the reorganization of the perceptual field—an intuitive restructuring often referred to as insight (Einsicht)—rather than the blind accumulation of conditioned reflex arcs.

Opposing this holistic paradigm was American neobehaviorism, an intellectual tradition committed to mechanistic atomism, logical positivism, and strict operationalism. For neobehaviorists, invoking unobservable cognitive phenomena such as “relational apprehension,” “structural insight,” or “perceptual Gestalten” represented an unacceptable retreat into mentalistic mysticism. The primary mandate of behaviorism, crystallized by John B. Watson and extended by Clark L. Hull, was to construct an objective, deterministic science of behavior that explained complex actions strictly through peripheral, observable events: physical stimuli acting upon sensory receptors, muscle contractions, and quantifiable reinforcement histories.

The central question driving the transposition controversy was therefore existential for behaviorist psychology: could the apparent perception of relationships be systematically reduced to elementistic stimulus-response (S-R) associative chains? If behaviorists failed to explain transposition using physical, absolute stimulus properties, they would be forced to concede that even non-human animals possess cognitive mechanisms capable of abstracting relational concepts independently of localized physical reinforcement. The transposition paradigm was thus transformed into a philosophical testing ground where the future direction of cognitive versus behavioral psychology would be contested.

1.3 Kenneth W. Spence and the Neobehaviorist Research Program

Kenneth Wartenbee Spence (1907–1967) emerged as the definitive champion of neobehaviorist quantitative rigor in this debate. A Canadian-born psychologist who completed his doctoral training at Yale University under the guidance of Robert Yerkes and Clark Hull, Spence possessed a rare combination of exceptional experimental finesse and mathematical sophistication. Serving as a central architect of the Hull-Spence learning theory during his distinguished tenure at the University of Iowa, Spence dedicated his career to refining Hull’s grand theoretical architecture into an operationalized, predictive, and mathematically closed system of behavioral laws.

Spence was acutely aware of the philosophical and empirical vulnerabilities of early behaviorist models. While Hull was prone to expansive, highly complex theoretical formulations that often outpaced their empirical verification, Spence maintained an unyielding commitment to formal deductivism. He believed that a psychological theory must consist of a tightly linked chain of mathematical postulates from which specific, quantifiable empirical outcomes could be deduced with absolute logical necessity. His overarching ambition was to demonstrate that quantitative associative learning theory could withstand the fiercest critiques launched by European Gestalt theorists and cognitive psychologists such as Edward C. Tolman.

To defend the associative paradigm, Spence undertook a rigorous analysis of the transposition phenomenon. Rather than dismissing the empirical findings of Köhler and his followers, Spence accepted their data as genuine behavioral realities that demanded a rigorous, mechanistic explanation. His revolutionary contribution was published in two seminal papers in the Psychological Review: “The Nature of Discrimination Learning in Animals” (1936) and “The Analysis of the Interaction of Two Basic Factors in Discrimination Learning” (1937). In these landmark works, Spence constructed a mathematical model based on the spatial interaction of continuous generalization gradients that fundamentally redefined the debate over relational versus absolute learning.

Spence’s 1936 and 1937 papers did not merely offer a post-hoc rationalization of transposition; they provided an axiomatic deductive system. By applying the principles of Pavlovian stimulus generalization and Thorndikian reinforcement to discrimination learning, Spence proved that preference for a novel, more extreme stimulus was the mathematically mandated outcome of overlapping absolute excitatory and inhibitory potentials. Furthermore, his model yielded bold, counterintuitive empirical predictions that Gestalt theory could neither anticipate nor easily accommodate. Through these papers, Spence positioned Hullian behaviorism at the zenith of its predictive power, creating an intellectual framework that dominated comparative learning theory for half a century.

2. The Gestalt Foundation: Wolfgang Köhler and Relational Learning

2.1 Köhler’s Classical Transposition Experiments

The empirical foundation of the transposition debate was laid during the First World War by the German psychologist Wolfgang Köhler while conducting research at the Prussian Academy of Sciences’ anthropoid research station on the island of Tenerife. Although Köhler is most famous for his classic work on problem-solving and tool use in chimpanzees, his rigorous investigations into discrimination learning in domestic fowl (hens) and young chimpanzees provided the empirical impetus for the relational challenge to behaviorism. Köhler designed a series of deceptively simple discrimination protocols that exposed the inadequacies of classical associationist assumptions.

In his standard experimental arrangement with domestic hens, Köhler affixed two sheets of paper of differing shades of gray onto a testing platform. One sheet was a medium gray (designated as stimulus b), while the second sheet was a darker gray (designated as stimulus a). Grains of corn were glued firmly to the darker sheet a, rendering them inaccessible, while edible, unglued grains were placed upon the lighter sheet b. Whenever the hen attempted to peck at the dark gray paper (a), it was shooed away or experienced the frustration of unyielding grains; when it pecked at the lighter gray paper (b), it was permitted to consume the food reward. This training continued across hundreds of trials until the birds exhibited near-perfect discrimination, pecking almost exclusively at the positive stimulus, b.

Once this baseline discrimination was mastered, Köhler introduced the critical test phase. Retaining the medium gray positive stimulus (b), he removed the dark gray negative stimulus (a) and replaced it with a novel sheet of paper, stimulus c, which was a substantially lighter shade of gray than b. The hen was therefore confronted with a choice between the familiar, consistently reinforced stimulus (b) and an entirely novel stimulus (c) that had never been associated with food. Classical associationism predicted that the bird, driven by habits established through hundreds of successful pecks, would overwhelmingly choose b. Instead, Köhler observed a striking and counterintuitive outcome: the hens bypassed the familiar reinforced stimulus b and directed their pecks systematically toward the novel lighter gray sheet c, achieving transposition rates frequently exceeding 70% to 80%.

Köhler replicated this phenomenon across multiple sensory dimensions and diverse animal species. When chimpanzees were trained to select the larger of two wooden boxes to retrieve fruit, they exhibited an identical behavioral pattern when presented with the previously reinforced box alongside an even larger, novel box: they systematically selected the novel, larger container. For Köhler, these consistent empirical demonstrations proved that the subjects were not conditioning their motor responses to localized physical wavelengths, specific surface reflectances, or exact physical areas. Instead, they were responding to the dynamic structural gradient inherent in the stimulus array.

2.2 The Gestalt Interpretation: Perception of Relationships (Gestaltqualität)

Köhler interpreted these empirical findings through the theoretical lens of Gestalt psychology, asserting that the fundamental unit of perceptual experience is not the isolated sensation, but the organized configuration or relation—termed a Gestaltqualität. Köhler argued that the nervous system does not function as an assemblage of independent, passive telephone wires transmitting localized sensory inputs to peripheral motor switches. Rather, the sensory cortex acts as a continuous physical field within which incoming neural excitations interact dynamically according to holistic physical laws, establishing structural equilibria.

In Köhler’s view, when an animal views two shades of gray side by side, its perceptual system does not process gray a and gray b as discrete, disconnected events. The sensory receptors register the step-like transition, the contrast, or the ratio between the two surfaces. The phenomenal experience is that of a “brighter-than” or “darker-than” relational vector. Therefore, what the animal learns during the discrimination training phase is not a habit attached to the absolute physical properties of stimulus b, but a behavioral orientation toward a structural relation: “choose the lighter of the two.” When presented with stimuli b and c during the test phase, the perceptual relation “lighter-than” remains structurally identical, even though the absolute physical components have shifted. The animal chooses stimulus c because c embodies the identical relational role that b occupied during training.

This formulation represented a frontal assault on the “constancy hypothesis”—the long-standing associationist assumption that a constant one-to-one correspondence exists between a localized physical stimulus and a specific sensory experience or conditioned response. Köhler rejected elementistic sensory psychology, asserting that an individual sensory element has no fixed psychological reality outside of the broader context or configuration in which it is embedded. What an organism experiences, remembers, and acts upon is the contextualized whole, governed by field forces and internal self-distribution of energy within the brain’s sensory systems.

Furthermore, Köhler linked transposition directly to his broader concept of insight (Einsicht). He posited that intelligent behavior across the animal kingdom does not proceed through the blind, mechanical stamping-in of random motor habits via gradual reinforcement, as proposed by Edward L. Thorndike. Instead, learning involves a dynamic restructuring of the organism’s cognitive and perceptual field. The transposition of a learned discrimination to a novel set of stimuli was held up as clear empirical verification that animals comprehend functional configurations, demonstrating an elementary form of conceptual abstraction that eluded the crude mechanics of peripheral behaviorism.

2.3 Theoretical Challenges Posed by the Relational View

Despite its intuitive appeal and phenomenological elegance, the Gestalt relational interpretation suffered from profound scientific and theoretical vulnerabilities. The most debilitating weakness was its chronic lack of quantitative precision and operational definitions. Gestalt theorists frequently invoked terms such as “insight,” “perceptual reorganization,” “dynamic field forces,” and “structural closure,” yet they consistently failed to translate these concepts into rigorous mathematical models that could generate precise, falsifiable predictions regarding behavioral choices.

For an experimental science grounded in measurement, Gestalt theory provided no formal algorithm to determine how relational perception would behave across varying degrees of stimulus change. If an animal transposes the rule “choose the larger” from a pair of 60 and 100 square-centimeter stimuli to a pair of 100 and 160 square-centimeter stimuli, what happens if the novel pair consists of stimuli measuring 500 and 800 square centimeters? Gestalt theory implied that as long as the invariant ratio or structural relationship (“larger-than”) was preserved, transposition should persist indefinitely across the entire continuum. Because Gestalt theorists viewed the relational rule as a unitary perceptual property, they possessed no theoretical mechanism to predict the degradation or collapse of transposition as stimuli moved further away from the original training coordinates.

Moreover, the relational hypothesis mounted a severe challenge to the prevailing mechanistic models of reinforcement without offering a viable mechanistic alternative. In Thorndikian and early behaviorist paradigms, reinforcement was conceptualized as a biological state of affairs that strengthened the physical associative bond between an antecedent stimulus and an execution pattern. If an animal could systematically reject a stimulus that possessed a direct, prolonged history of reinforcement in favor of an unknown stimulus that had never been accompanied by reinforcement, the absolute connection between reinforcement and habit formation was thrown into theoretical chaos.

Gestalt psychologists were largely content to demonstrate that mechanistic behaviorism had failed to explain the phenomenon, treating transposition as an empirical proof of behaviorism’s bankruptcy. However, by relying on descriptive phenomenological language rather than quantifiable predictive curves, the Gestalt school left a critical vacuum. They failed to anticipate that an astute behaviorist theorist might construct an absolute, non-relational mechanism capable of producing the exact relational outcomes observed in Köhler’s laboratory—a feat that would decisively alter the course of comparative psychology.

3. The Behavioral Paradigm: Absolute Stimulus-Response Learning

3.1 Core Premises of Absolute Associationism

The foundation of the behaviorist approach to discrimination learning rested upon the core premises of absolute associationism. Originating in the British empiricist traditions of John Locke and David Hume and subsequently formalized into an objective experimental methodology by Ivan Pavlov and Edward Thorndike, associationism asserted that all learning consists of the formation and modification of connections between discrete, physical elements. When transposed into twentieth-century behaviorism, this principle dictated that learning must attach directly and exclusively to measurable physical stimulus dimensions impinging upon specific receptor organs.

In absolute associationism, the physical continuum—whether calibrated in units of surface area, photometric foot-candles, auditory cycles per second, or electromagnetic wavelengths—serves as an absolute objective metric. When an organism is exposed to a visual stimulus, that stimulus is not processed as an abstract cognitive signifier; it is an aggregation of absolute physical energies activating distinct sensory pathways. The conditioned response strength that accrues to that stimulus is a direct mathematical function of the cumulative physical reinforcement history experienced at that precise physical locus. Learning, therefore, is elementistic, localized, and atomistic.

A non-negotiable axiom of absolute behaviorism was the operational refusal to invoke internal, unobservable cognitive mechanisms to bridge the gap between stimulus and response. To introduce concepts such as “relational apprehension,” “mental representations,” or “conscious insight” was viewed as an epistemological surrender to Cartesian dualism. For the neobehaviorist, the organism was a biological machine operating within a physical field governed by determinism. Behavior was the observable dependent variable, completely determined by the interaction of experimental history, biological drives, and current environmental inputs. If relational behavior occurred, it had to be derived from the mechanical operations of physical stimulus traces interacting within the nervous system.

Consequently, the neobehaviorist research program demanded that every discrimination task be conceptualized as an absolute competition between discrete stimulus points. If stimulus S+ is presented, an associative vector linking S+ to an approach response is strengthened whenever reinforcement is delivered. If stimulus S- is presented, the associative vector linking S- to an approach response is weakened or actively suppressed when reinforcement is withheld. In this elemental view, there is no conceptual room for an overarching “relation” linking the two stimuli; there are only two independent physical events acquiring distinct associative weights through discrete temporal encounters with reinforcement and non-reinforcement.

3.2 Hullian Hull-Spence Drive-Reduction and Habit Strength

The preeminent theoretical framework operationalizing this absolute associationist philosophy was the system developed by Clark L. Hull and expanded by Kenneth Spence, widely known as the Hull-Spence learning theory. At the heart of this comprehensive, deductive neobehaviorist architecture was the postulate that learning is fundamentally driven by physiological drive-reduction. When an organism experiences a biological deficit—such as nutritional or fluid deprivation—a physiological drive state (D) is established, energizing the organism into behavioral output.

Within this framework, the central associative construct was Habit Strength, formally symbolized as sHr. Habit strength represented the enduring, structural connection forged between a specific conditioned stimulus (S) and a conditioned motor response (R). Hull postulated that whenever a response occurs in temporal contiguity with a stimulus and is immediately followed by the reduction of an organic drive, a quantum of habit strength is permanently deposited into that specific S-R circuit. Mathematically, the growth of habit strength was defined as a negatively accelerating monotonic function of the number of reinforced training trials (N):

sHr = M * (1 – e^(-i * N))

where M represents the physiological asymptote of habit formation and i represents the empirical rate parameter of acquisition. Habit strength, however, was an unobservable intervening variable that did not directly translate into behavioral action. To generate overt behavior, habit strength had to interact multiplicatively with the current level of physiological drive (D) and, in later iterations of the theory, the incentive value of the reinforcer (K), generating an active behavioral tendency designated as Excitatory Potential or Reaction Potential (sEr):

sEr = sHr * D * K

This reaction potential was balanced by counteracting forces of unconditioned and conditioned inhibition, designated as sIr and Ir, which accrued whenever responses were executed without drive-reduction. The ultimate probability, latency, and amplitude of the overt behavioral response were dictated by the effective reaction potential (sĒr), derived from the subtraction of total inhibitory potential from total excitatory potential.

Crucially, throughout this intricate mathematical system, the associative bonding remained strictly localized between physical receptor stimulation and physical motor effectors. The Hull-Spence model required absolute continuity across the sensory-motor interface. To suggest that an organism could form a habit strength attached not to a physical stimulus point, but to an abstract ratio or relational vector between two stimuli, undermined the core foundation of Hull’s drive-reduction mechanics. The associative matrix could only store values calibrated to physical sensory-neural coordinates.

3.3 The Problem Transposition Presented to Classic Associationism

It was precisely this unyielding commitment to absolute physical stimulus values that rendered the phenomenon of stimulus transposition an existential crisis for early associationist theory. When analyzed through the classical associationist lens, the empirical outcome of Köhler’s transposition experiments appeared to be a logical impossibility and an outright refutation of the law of effect. The paradox was both acute and undeniable.

Consider the empirical coordinates of the standard transposition paradigm. During the acquisition phase, an animal experiences dozens or hundreds of discrete trials. Every time it approaches the positive stimulus (S+), its sensory receptors are stimulated by that precise physical energy, its motor behavior is executed, and it receives immediate drive-reduction. According to the foundational postulates of Thorndike and Hull, maximal habit strength (sHr) and maximal reaction potential (sEr) must inevitably coalesce directly around S+. Conversely, the novel stimulus introduced during the test phase (S++) has never been encountered within the experimental apparatus; it has never stimulated the subject’s receptors in temporal contiguity with drive-reduction, and it therefore possesses zero directly conditioned habit strength (sHr = 0).

How, then, could an animal presented with a choice between S+ and S++ systematically bypass S+ and choose S++? Classical associationist models had no theoretical mechanism to explain this preference. Under standard associative logic, the choice probability should have heavily favored S+, the stimulus bearing the maximal history of reinforced trials. The organism’s systematic preference for the novel stimulus appeared to demonstrate that behavior was not governed by the accumulation of reinforcement at absolute physical coordinates.

For the opponents of behaviorism, this failure was fatal. Gestalt theorists and cognitive psychologists argued that the associationist model was dead, wrecked upon the reality of transposition. They insisted that the animal’s rejection of the reinforced stimulus in favor of the novel stimulus proved that the creature had formed an internal cognitive representation of the relationship—a dynamic rule that rendered absolute associative reinforcement histories irrelevant. To save the neobehaviorist research program from theoretical collapse, Kenneth Spence needed to construct an entirely new mathematical formulation: an absolute mechanism that could account for the relational outcome while remaining completely faithful to the physicalist axioms of Hullian conditioning.

4. The Mechanics of Spence’s Model: Excitatory and Inhibitory Gradients

4.1 The Excitatory Gradient (S+)

Spence’s revolutionary solution to the transposition dilemma was anchored in an integration of Hullian reinforcement theory and Pavlovian principles of stimulus generalization. In his 1936 and 1937 formulations, Spence postulated that reinforcement does not confine its associative effects exclusively to the precise physical stimulus coordinate present during conditioning. Instead, drawing upon Pavlov’s physiological observations of sensory irradiation, Spence asserted that when an approach response to a positive stimulus (S+) is reinforced, an enduring Excitatory Potential (symbolized as E) is deposited not only at S+, but also spreads or “generalizes” across adjacent stimulus values along the continuous physical dimension.

This generalized associative field forms a mathematical distribution termed the Excitatory Gradient. At the exact physical locus of S+, the excitatory potential achieves its maximum asymptotic magnitude, reflecting the direct history of drive-reduction. However, as one moves away from S+ in either direction along the physical continuum—whether toward higher or lower values of luminance, area, or frequency—the magnitude of excitatory potential progressively decays. Spence conceptualized this gradient as a continuous, symmetrical, bell-shaped or Gaussian-like distribution centered symmetrically upon S+.

The mathematical properties of this excitatory generalization gradient can be formally modeled through an exponential decay function or a normal density distribution. If we designate the physical dimension as a continuous metric x, and the precise physical locus of the positive stimulus as x₊, the excitatory potential E(x) at any arbitrary point x along the continuum can be mathematically expressed as:

E(x) = E_max * e^(-((x – x₊)^2 / (2 * σ_E^2)))

where E_max represents the maximal excitatory potential accumulated at S+ as an asymptotic function of reinforcement trials, and σ_E represents the generalization parameter governing the dispersion or width of the excitatory spread. Points on the physical dimension immediately adjacent to S+ inherit substantial quantities of excitatory potential through generalization, whereas points situated at great physical distances from S+ retain negligible traces of excitation, approaching an asymptotic baseline of zero.

Spence emphasized that this excitatory gradient is a strictly absolute physical phenomenon. The curve does not reflect an organism’s cognitive inference about potential rewards; it represents the neurobiological reality that sensory stimulation activates overlapping populations of afferent neural pathways. An adjacent physical stimulus excites a subset of the identical neural units conditioned during S+ training, mechanically evoking a proportion of the original approach habit. The physical continuum serves as the spatial substrate upon which absolute excitation is distributed.

4.2 The Inhibitory Gradient (S-)

The second foundational component of Spence’s model was the concept of active, conditioned inhibition. In contrast to simple single-process models that viewed non-reinforcement merely as a passive absence of learning, Spence embraced the Pavlovian construct of internal inhibition. He postulated that when an organism approaches the negative stimulus (S-) and experiences the non-reinforcement or frustration of an anticipated reward, an active Inhibitory Potential (symbolized as I) is acquired at that physical coordinate. This inhibitory potential acts as a direct, active counter-force to behavioral execution, functioning as a behavioral brake that suppresses the tendency to approach S-.

Critically, Spence asserted that just as excitation generalizes across adjacent physical values, so too does inhibition. The non-reinforcement of responses to S- generates an Inhibitory Gradient that spreads outward along the continuous physical spectrum, centered symmetrically upon the locus of S-. At the exact coordinate of S-, the inhibitory potential reaches its maximal value, reflecting the cumulative history of non-reinforced approaches. As the physical distance from S- increases in either direction, this inhibitory potential systematically diminishes toward zero.

Formally, the inhibitory potential I(x) across any point x on the continuum can be defined using an analogous spatial decay function:

I(x) = I_max * e^(-((x – x_–)^2 / (2 * σ_I^2)))

where I_max is the peak inhibitory potential accumulated at the negative stimulus coordinate x_–, and σ_I represents the spatial spread parameter of the inhibitory gradient.

A crucial theoretical postulate introduced by Spence, which proved decisive for the internal mechanics of his model, concerned the relative geometric properties of the excitatory and inhibitory curves. Spence postulated that the inhibitory gradient is generally narrower and steeper than the excitatory gradient (i.e., σ_I < σ_E). While excitatory tendencies spread broadly across wide expanses of the sensory continuum, active conditioned inhibition remains more localized, clustering tightly around the unreinforced stimulus coordinate. As will become evident, this structural asymmetry between the broad, sweeping excitatory curve and the steep, compact inhibitory curve provided the mathematical engine that drove the displacement of behavioral choice in transposition tasks.

4.3 Algebraic Summation: Deriving Net Associative Strength

The definitive intellectual breakthrough of Spence’s 1936 and 1937 papers resided in the principle of algebraic summation. Spence recognized that during simultaneous discrimination training, the animal is not subjected to an isolated excitatory experience or an isolated inhibitory experience; rather, the organism is exposed to the concurrent interaction of both associative forces operating simultaneously within the identical sensory continuum.

Spence formulated the fundamental postulate that the overt behavioral tendency to approach any given physical stimulus point x is dictated by its Net Effective Excitatory Potential, formally designated as E_net (or in Hullian notation, sĒr). This net value is derived strictly through the linear algebraic subtraction of the generalized inhibitory potential from the generalized excitatory potential at that precise spatial coordinate:

E_net(x) = E(x) – I(x)

To visualize the operational mechanics of this formulation, one must picture a single, continuous physical metric—such as visual area—plotted along the horizontal axis. Centered at the physical value corresponding to S+ is a broad, bell-shaped excitatory curve rising into positive values. Centered at the physical value corresponding to S- is a narrower, steep inhibitory curve, extending downward. Because S+ and S- are separated by a finite physical distance along the continuum, their generalization tails overlap extensively.

When the subtraction E_net(x) = E(x) – I(x) is executed across every infinitesimal point along the horizontal continuum, the overlapping curves cancel each other out in complex, asymmetric ways. In the immediate physical vicinity of S-, the powerful, steep inhibitory gradient heavily depresses the net value, frequently driving effective reaction potential below the absolute behavioral threshold of response, ensuring that the animal actively avoids S-. Conversely, in the region surrounding S+, the broad excitatory potential dominates, yielding substantial positive net values that drive approach behavior.

However, the most profound and unexpected consequence of this algebraic summation occurs not at S- or S+, but in the physical territory immediately adjacent to S+ on the side opposite to S-. Because the inhibitory gradient decays rapidly as physical distance increases from S-, its suppressive influence reaches zero much faster than does the broad, sweeping tail of the excitatory gradient. Consequently, on the side of S+ that faces away from S-, the excitatory potential is left almost entirely unencumbered by subtraction. This differential interaction establishes an entirely new, transformed landscape of effective associative strength across the physical continuum, containing within its geometry a revolutionary deductive surprise.

5. The Peak Shift Phenomenon and Spence’s Deductive Solution

5.1 The Mathematical Shift of Maximal Response Strength

The crowning achievement of Kenneth Spence’s theoretical work was the mathematical discovery of what would later be termed the peak shift phenomenon. Through pure deductive logic, Spence realized that when a steep inhibitory gradient is subtracted from a broad excitatory gradient that sits adjacent to it, the point of maximal net associative strength (E_net) does not remain at the physical coordinate of the reinforced stimulus (S+). Instead, the peak of maximal net excitation is geometrically displaced outward, shifting systematically away from S+ in the direction opposite to the inhibitory stimulus S-.

The geometric proof of this displacement is both elegant and incontrovertible. Consider the physical coordinate of S+ itself. Although the excitatory potential E is at its absolute maximum at this point, S+ is located uncomfortably close to S-. Because the generalization tail of the inhibitory gradient extends across the intervening physical distance, a measurable quantity of inhibition, I(S+), spills over onto S+. Therefore, the net associative value at S+ is reduced: E_net(S+) = E_max – I(S+).

Now, examine a novel stimulus coordinate, designated as S++, which lies further away from S+ in the direction opposite to S-. At this novel coordinate, the excitatory potential E(S++) is slightly lower than E_max, because it resides on the descending generalized flank of the excitatory curve. However, because S++ is significantly farther away from S- than S+ was, the inhibitory potential reaching S++ from the distant S- has decayed to virtually zero: I(S++) ≈ 0. When the algebraic subtraction is performed:

E_net(S++) = E(S++) – 0 = E(S++)

If the quantity of inhibition subtracted at S+ is greater than the slight loss of excitation incurred by moving outward to S++—that is, if I(S+) > [E_max – E(S++)]—then the net associative strength at the novel stimulus will be mathematically larger than the net associative strength at the actual reinforced stimulus:

E_net(S++) > E_net(S+)

This mathematical inequality provided the deductive solution to the transposition paradox that had eluded behavioral psychology for decades. When the experimental subject is placed in the transposition test phase and presented with a choice between the familiar reinforced stimulus (S+) and the novel stimulus (S++), the animal does not engage in perceptual abstraction, nor does it deduce the relational concept of “larger-than.” Instead, the animal functions as a mechanical Hullian selector, automatically choosing the physical stimulus that possesses the highest effective net reaction potential. Because algebraic summation has displaced the peak of net excitation outward, the animal selects S++ precisely because S++ commands a greater absolute net associative charge than S+. Spence had succeeded in generating a relational choice through purely absolute, non-cognitive associative mechanics.

5.2 Step-by-Step Numerical Reconstruction of Spence’s 1937 Model

To appreciate the absolute mechanical clarity of Spence’s 1937 model, it is instructive to reconstruct his exact numerical demonstration using standardized, hypothetical units of associative potential. Let us calibrate a stimulus continuum based on the visual surface area of circular geometric discs, utilizing the specific logarithmic step values common to historical discrimination protocols: 39 sq cm, 62 sq cm, 100 sq cm, 160 sq cm, 256 sq cm, and 409 sq cm.

In this classic scenario, let the negative stimulus (S-) be established at 62 sq cm, and the positive stimulus (S+) be established at 100 sq cm. Following exhaustive discrimination training, we assign standardized theoretical values for the acquired excitatory and inhibitory gradients across the continuum, consistent with Spence’s empirical assumptions that the maximal excitation at S+ equals 100 units (with a broad spread) and maximal inhibition at S- equals 70 units (with a steeper spread):

• Stimulus 39 sq cm (Distant lower coordinate): Excitatory potential = 20.0; Inhibitory potential = 45.0. Net associative strength: E_net = 20.0 – 45.0 = -25.0 (Absolute inhibition/avoidance).
• Stimulus 62 sq cm (Training negative stimulus, S-): Excitatory potential = 50.0; Inhibitory potential = 70.0 (Maximal I). Net associative strength: E_net = 50.0 – 70.0 = -20.0 (Robust avoidance).
• Stimulus 100 sq cm (Training positive stimulus, S+): Excitatory potential = 100.0 (Maximal E); Inhibitory potential = 35.0 (Substantial generalized inhibition spilling over from S-). Net associative strength: E_net = 100.0 – 35.0 = 65.0.
• Stimulus 160 sq cm (Novel transposition test stimulus, S++): Excitatory potential = 82.0 (Mild generalization decay from S+); Inhibitory potential = 8.0 (Inhibitory gradient has nearly fully decayed). Net associative strength: E_net = 82.0 – 8.0 = 74.0.
• Stimulus 256 sq cm (Far transposition test stimulus, S+++): Excitatory potential = 52.0; Inhibitory potential = 1.0. Net associative strength: E_net = 52.0 – 1.0 = 51.0.
• Stimulus 409 sq cm (Extreme distant coordinate): Excitatory potential = 15.0; Inhibitory potential = 0.0. Net associative strength: E_net = 15.0 – 0.0 = 15.0.

Now, let us examine the mathematical reality confronting the experimental subject during the critical near transposition test trial, where S+ (100 sq cm) is paired simultaneously with S++ (160 sq cm). The associative strength pulling the animal toward the familiar, reinforced 100 sq cm disc is E_net = 65.0. The associative strength pulling the animal toward the novel, never-reinforced 160 sq cm disc is E_net = 74.0. Because behavioral choice is determined by the differential magnitude of effective reaction potential, the animal must inevitably approach the 160 sq cm stimulus. The animal chooses the larger disc not because it possesses a concept of “larger,” but because algebraic subtraction has rendered 160 sq cm the absolute point of maximal behavioral excitation within that pair.

5.3 Empirical Verification: Norman Guttman and Harry Kalish (1956)

While Spence’s 1937 paper provided a flawless deductive proof, the physical displacement of the behavioral peak remained a theoretical deduction for nearly two decades. The empirical vindication of Spence’s mathematical peak shift arrived in 1956 in an experimental study conducted by Norman Guttman and Harry Kalish at Duke University, which became an instant classic in the psychology of learning.

Guttman and Kalish adapted B. F. Skinner’s automated operant conditioning technology to evaluate discrimination generalization along a continuous physical spectrum with unprecedented experimental control. Utilizing pigeons as subjects, they projected precise monochromatic light wavelengths onto a translucent response key inside an operant chamber. In their baseline discrimination protocols, pigeons were reinforced on a variable-interval schedule for pecking the key when illuminated by a specific wavelength—for instance, an amber-yellow light of 550 millimicrons (designated as S+). Simultaneously, an adjacent wavelength—such as an orange light of 570 millimicrons—served as the negative stimulus (S-), during which key pecks were never reinforced.

Following training, Guttman and Kalish administered a generalization test under extinction conditions, systematically exposing the pigeons to a wide, pseudo-randomized spectrum of monochromatic wavelengths spanning from 480 to 620 millimicrons, while recording the exact rate of key-pecking at each discrete spectral band. When the generalization curves were plotted, the resulting data vindicated Spence’s mathematical model. The highest rate of pecking did not occur at the training wavelength of 550 millimicrons. Instead, the maximum response frequency was systematically displaced to 540 or 535 millimicrons—shifting directly outward along the spectral continuum away from the inhibitory 570-millimicron stimulus.

The Guttman-Kalish experiment provided undeniable empirical verification that the post-discrimination generalization gradient undergoes an actual physical displacement. The peak shift was not an artifact of theoretical imagination; it was an objective biological phenomenon that emerged whenever an excitatory gradient interacted with an overlapping inhibitory gradient along a continuous sensory dimension. By confirming Spence’s deduction through high-resolution operant testing, Guttman and Kalish established Spence’s algebraic model as one of the most successful predictive frameworks in the history of behavior analysis.

6. The Critical Falsification Test: The Transposition Distance Effect

6.1 Near Transposition versus Far Transposition

The true mark of a scientific theory lies not merely in its capacity to explain existing empirical data, but in its ability to generate risky, counterintuitive predictions that expose the theory to potential falsification. Kenneth Spence understood that if his model merely explained the near transposition observed by Köhler (e.g., testing the adjacent pair 100 vs. 160 sq cm), it would remain locked in a theoretical stalemate with Gestalt psychology. To decisively defeat the relational hypothesis, Spence needed to identify an empirical domain where the predictions of Gestalt holism and absolute associationism diverged completely.

This critical empirical battleground was the Transposition Distance Effect. Spence differentiated between two operational categories of testing: near transposition, where the test pair consists of stimuli one incremental step removed from the training stimuli along the continuum; and far transposition, where the test pair consists of novel stimuli positioned several steps, or octaves, removed from the original training pair.

Gestalt theory made an unequivocal, categorical prediction regarding distance: because the organism learns an invariant structural relation or perceptual rule (“choose the larger”), the physical distance between the training stimuli and the test stimuli should have no meaningful impact on behavioral choice. As long as the novel test stimuli maintained the identical ratio and spatial relationship (e.g., 256 vs. 409 sq cm, or even 1000 vs. 1600 sq cm), the animal should perceive the identical Gestalt configuration and systematically select the larger stimulus. For Gestalt psychology, relational transposition should remain robust across the entire physical continuum.

Spence’s algebraic model generated the exact opposite prediction. According to Spence, transposition is not driven by an enduring perceptual rule, but by the local algebraic interaction of overlapping generalization gradients. Because both excitatory and inhibitory generalization gradients are spatially bounded phenomena that decay as physical distance increases, their differential interaction must inevitably attenuate at remote stimulus coordinates. Therefore, Spence predicted that while transposition would reliably succeed at near distances, it must progressively weaken, break down, and ultimately fail as the novel test stimuli were moved farther and farther along the physical continuum away from the original training coordinates.

6.2 Mathematical Deduction of Transposition Failure

The mathematical deduction of transposition failure at extreme physical distances emerges naturally from the numerical properties of Spence’s decaying gradients. Returning to the numerical values established in Section 5.2, let us examine the net associative values when the organism is tested with the far transposition pair: Stimulus 256 sq cm (S+++) versus Stimulus 409 sq cm (S++++):

• At 256 sq cm, the generalized excitatory potential has decayed to 52.0, while the generalized inhibitory potential has decayed to 1.0. Net associative strength: E_net(256) = 52.0 – 1.0 = 51.0.
• At 409 sq cm, the generalized excitatory potential has decayed to 15.0, while the generalized inhibitory potential has reached 0.0. Net associative strength: E_net(409) = 15.0 – 0.0 = 15.0.

When the animal is forced to choose between 256 sq cm and 409 sq cm in a far transposition test, what must happen according to Spence’s model? The net reaction potential driving the approach to the 256 sq cm stimulus is 51.0, whereas the net reaction potential driving the approach to the 409 sq cm stimulus is only 15.0. Therefore:

E_net(256) > E_net(409)

The mathematical outcome is astonishing: Spence’s model deduces that when presented with this distant novel pair, the animal will systematically choose the smaller of the two stimuli (256 sq cm over 409 sq cm). Far from transposing the relational rule “choose the larger,” the animal will exhibit a complete choice reversal! At extreme physical distances, because both gradients have decayed toward an absolute asymptote of zero, the difference in net associative strength between test stimuli becomes negligible (E_net(A) ≈ E_net(B) ≈ 0), causing behavior to collapse into chance-level responding (50% random choice).

This mathematical deduction represented a monumental theoretical fork in the road. Gestalt psychology predicted stable, continuous relational choice across all physical distances. Spence predicted a predictable trajectory of behavioral collapse: robust relational choice at near distances, transitioning into systematic choice reversal at intermediate-far distances, and finally terminating in absolute random performance at extreme distances. If experimental testing confirmed this distance-dependent decay, the foundational claim of unconstrained Gestalt relational learning would be completely falsified.

6.3 Experimental Evidence for the Distance Effect

Experimental psychology immediately mobilized to test Spence’s bold deduction, launching an era of rigorous empirical testing across diverse vertebrate species. The empirical results overwhelmingly confirmed the existence of the transposition distance effect, delivering a severe blow to the classical Gestalt interpretation.

In a seminal series of experiments conducted with chimpanzees at the Yale Laboratories of Primate Biology, Spence (1937) empirically validated his own predictions. When tested on stimulus pairs immediately adjacent to the training stimuli, the chimpanzees transposed the “larger” choice with near 100% accuracy. However, when the test pairs were shifted several steps along the dimensional scale, the frequency of relational choices systematically declined. At far distances, transposition completely collapsed, with chimpanzees selecting the smaller stimulus or exhibiting behavioral hesitation and chance-level performance, precisely as deduced by the gradient summation model.

Subsequent investigations across diverse animal models confirmed the universality of the distance effect in non-human subjects. Studies conducted with domestic fowl, rats, and rhesus macaques repeatedly revealed the identical behavioral profile: near transposition was robust, but far transposition failed. Avian studies, in particular, demonstrated that when pigeons and chickens were presented with luminance or size stimuli separated by multiple steps from the training values, the birds consistently exhibited preference reversals or random choice distributions.

The discovery of the distance effect represented a triumph for the Hull-Spence research program. Gestalt psychologists could provide no coherent explanation for why an organism that supposedly possessed the abstract conceptual rule “choose the larger” would suddenly abandon that rule when the stimuli were physically larger, yet structurally identical in ratio. Spence’s absolute associative model not only explained why transposition occurred at near distances, but uniquely explained why it had to fail at far distances. Through the verification of the distance effect, neobehaviorism successfully reclaimed the empirical high ground.

7. Methodological Paradigms in Transposition Research

7.1 Apparatus and Experimental Controls

The fierce empirical debates surrounding transposition necessitated unprecedented advancements in apparatus design, measurement precision, and experimental methodology. To evaluate whether learning was governed by absolute physical values or relational properties, researchers had to ensure that the physical stimuli were standardized with extreme scientific rigor, eliminating all extraneous sensory artifacts that could confound the subjects’ choices.

The preeminent apparatus developed to study discrimination and transposition in primates was the Wisconsin General Test Apparatus (WGTA), conceptualized by Harry Harlow. The WGTA consisted of a standardized testing enclosure where a primate subject faced a presentation tray containing stimulus wells covered by experimental objects (such as wooden blocks of varying geometric areas or painted luminance values). The experimenter was concealed behind a one-way observation screen, preventing subtle experimenter-expectancy effects or unintended social cues—a critical control that eliminated the “Clever Hans” phenomenon that had compromised earlier animal cognition studies.

In avian and rodent research, the introduction of automated operant chambers (Skinner boxes) revolutionized transposition testing. Rather than relying on manually presented stimuli, computer- or relay-controlled projectors presented visual stimuli onto translucent response keys. Operant chambers allowed for the microsecond-level recording of response latencies, response forces, and discrete pecking frequencies, eliminating subjective human scoring. Stimulus dimensions were calibrated using rigorous physical instrumentation: photometers to measure exact candelas per square meter for luminance, spectroradiometers to calibrate monochromatic light wavelengths, and precision micrometers to verify physical surface areas.

A vital methodological refinement was the mathematical standardization of stimulus dimensions along logarithmic scales rather than linear physical scales. Psychophysical research pioneered by Ernst Heinrich Weber and Gustav Fechner had demonstrated that sensory perception operates along proportional, logarithmic ratios rather than arithmetic increments. By spacing training and test stimuli along logarithmic progressions (e.g., area increments of 1.6:1), researchers ensured that the sensory intervals separating the stimuli remained psychophysically constant, preventing artifacts arising from non-linear sensory compression in peripheral receptor organs.

7.2 Simultaneous versus Successive Discrimination Protocols

A critical procedural variable that significantly modulated transposition behavior was the operational distinction between simultaneous and successive discrimination training protocols. The method of stimulus presentation fundamentally altered the sensory-motor dynamics of the learning task, leading to distinct patterns of associative gradient interaction.

In a simultaneous discrimination protocol, both S+ and S- are presented to the animal concurrently on every training trial, positioned side-by-side in space. The subject must visually scan both objects, actively orienting toward one while turning away from the other. This arrangement facilitates immediate, direct visual comparison. Gestalt theorists argued that simultaneous presentation was the only ecologically valid protocol for studying relational perception, as it presented an integrated perceptual field wherein the structural Gestalt could emerge naturally.

Conversely, in a successive discrimination protocol (often implemented as a “go/no-go” procedure), the stimuli are presented individually in time. On trial n, the animal is exposed exclusively to S+; if it responds, it receives reinforcement. On trial n+1, the animal is exposed exclusively to S-; if it responds, it receives no reinforcement and incurs a time-out penalty. In the successive design, the animal never views S+ and S- side-by-side; it must compare the current sensory input against an internal memorial trace of the alternative stimulus.

Kenneth Spence conducted a meticulous theoretical analysis comparing these two presentation modes. Spence demonstrated that simultaneous protocols generate significantly stronger peak shifts and more pronounced near-transposition effects than successive protocols. In simultaneous training, the animal frequently executes visual comparisons, scanning rapidly back and forth between S+ and S- before executing a final approach. Spence pointed out that this comparative scanning behavior exposes the organism’s receptors to immediate, rapid successions of excitation and inhibition, sharpening the boundaries of both generalization gradients. In successive training, the temporal separation between stimuli leads to broader, less differentiated associative fields, often resulting in flatter generalization curves and reduced transposition.

7.3 Reinforcement Schedules and Extinction Testing

A persistent methodological challenge in transposition research was the preservation of pristine gradient profiles during the critical transposition test phase. In classical testing, an animal is presented with novel stimulus pairs that have no established reinforcement history. If the experimenter reinforces responses to these novel stimuli, new associative learning instantly occurs, contaminating the very phenomenon under evaluation. Conversely, if the experimenter withholds reinforcement entirely during testing (extinction testing), the animal experiences extinction, rapidly ceasing to respond altogether after a small handful of trials.

To overcome this methodological conundrum, researchers developed sophisticated reinforcement architectures. One prevalent strategy was the implementation of intermittent reinforcement schedules (such as variable-interval or variable-ratio schedules) during the terminal phases of acquisition training. By habituating the subject to enduring long sequences of non-reinforced responses prior to the test phase, researchers generated high behavioral resistance to extinction, allowing for extensive testing with novel stimulus configurations under pure extinction conditions without degrading response rates.

Another widely adopted methodological compromise was the use of nondifferential reinforcement testing. In this protocol, every choice made during the test phase—whether the animal chooses S+ or S++, or chooses between far transposition pairs—is reinforced with equal probability. Because reinforcement is delivered nondifferentially across all options, no new systematic associative bias is introduced favoring one physical dimension over another. This technique preserved high response vigor while allowing researchers to map stable choice preferences across hundreds of trials.

Furthermore, researchers utilized randomized, Latin-square presentation sequences during the test phase, interspersing baseline training trials (S+ vs. S-) among non-reinforced test trials. This ensured that the original associative anchor points (the underlying excitatory and inhibitory peaks) remained biologically active and dynamically stable, preventing the total decay of the associative landscape while novel coordinates were being systematically probed.

8. Developmental and Comparative Evidence: Human Children and Higher Primates

8.1 The Kuenne Experiments: Language and Transposition Distance

As the transposition controversy matured, experimental psychologists recognized that the debate could not be resolved solely through non-human animal research. The critical question emerged: do human beings process discrimination tasks through absolute associative gradient summation, or through relational cognitive abstraction? The answer to this question was uncovered in a landmark developmental study published in 1946 by Margaret Kuenne, a brilliant student of Kenneth Spence at the University of Iowa.

Kuenne hypothesized that whether a human subject demonstrates absolute gradient-driven transposition or abstract relational transposition is directly contingent upon the developmental acquisition of language and verbal mediation. To test this hypothesis, Kuenne assembled a diverse cohort of young children ranging in age from two and a half to eight years old, stratifying them based on their chronological and mental age, with a specific focus on their expressive linguistic capabilities.

Kuenne trained the children on a standard size-discrimination task using a modified Wisconsin General Test Apparatus, requiring them to choose between two white square stimulus blocks (for instance, a small square versus a medium square) to find a hidden toy reward. Once discrimination was mastered, the children were evaluated on both near transposition test pairs and far transposition test pairs. The empirical results yielded a striking developmental dissociation that perfectly bridged the gap between Spence’s behaviorism and cognitive theory:

• Pre-verbal and low-linguistic children (Ages 2.5 to 4 years): These younger children exhibited a behavioral profile identical to that of Spence’s chimpanzees and pigeons. On near transposition trials, they systematically selected the larger object, exhibiting high rates of transposition. However, on far transposition trials, their relational choice completely broke down; they exhibited the classic transposition distance effect, collapsing into random responding or preference reversal. In the absence of verbal mechanisms, their behavior was completely governed by the algebraic summation of physical generalization gradients.
• Linguistically proficient, verbal children (Ages 5 to 8 years): These older children demonstrated a radically different behavioral architecture. They exhibited near 100% transposition on both near test pairs and far test pairs. The physical distance of the test stimuli along the continuum had zero effect on their choice behavior. When questioned about their choices, these older children articulated an explicit verbal rule: “I always picked the bigger one.”

The Kuenne experiments proved that Spence’s absolute gradient model accurately described sensory-motor learning in organisms lacking symbolic language systems (including non-human animals and pre-verbal infants). However, once language developed, human cognitive architecture transitioned into a higher-order representational system capable of symbolic relational coding, rendering the physical decay of sensory gradients irrelevant.

8.2 Verbal Mediation Theory: The Kendler and Kendler Reversal Shift Continuum

Building upon Kuenne’s developmental discoveries, Howard Kendler and Tracy Kendler formulated the comprehensive framework of Verbal Mediation Theory. The Kendlers sought to map the precise psychological and developmental trajectory through which human cognition transitions from the single-stage, absolute S-R associative learning formalized by Hull and Spence to the multi-stage, mediational learning characteristic of mature human thought.

To evaluate this cognitive transition, the Kendlers designed the reversal-shift and extradimensional-shift paradigm. In this protocol, subjects were trained to discriminate stimuli possessing two independent dimensions—such as size (large vs. small) and brightness (black vs. white). After initial acquisition, the reinforcement contingencies were unexpectedly shifted. In a reversal shift, the dimensional category remained relevant, but the reinforcement values inverted (e.g., if large had been S+ and small had been S-, small now became S+). In an extradimensional shift, the relevant dimension changed entirely (e.g., size became irrelevant, and white became S+).

The Kendlers demonstrated that for non-human mammals and young, pre-verbal children, an extradimensional shift was acquired significantly faster than a reversal shift. This occurred because, within Spence’s absolute associative model, an animal in a reversal shift must painstakingly extinguish a massive, fully established excitatory peak at the old S+ while overcoming a massive inhibitory barrier at the old S-, requiring a prolonged period of algebraic restructuring. In contrast, mature, verbal human beings executed reversal shifts with astonishing speed—often within a single trial. The Kendlers proved that older children and adult humans utilize an internal, covert verbal response that acts as an autonomous mediational link:

Physical Stimulus → Covert Verbal Mediator (“Size is relevant”) → Overt Choice Response

This verbal mediation theory effectively synthesized the Spencean and Gestalt positions within a developmental continuum. In lower organisms and pre-linguistic humans, learning is predominantly single-stage and absolute, governed strictly by the automatic algebraic summation of physical generalization gradients. As language develops, covert symbolic labels create abstract relational rules that override physical gradient decay. Transposition across great physical distances was thus revealed to be not an inherent primitive property of sensory perception as Köhler had assumed, but a higher-order cognitive achievement unlocked through the evolutionary and developmental emergence of symbolic language.

8.3 Primate Cognition: Rhesus Monkeys, Chimpanzees, and Great Apes

The comparative evaluation of transposition across non-human primates revealed nuanced complexities that pushed Spence’s purely absolute model to its theoretical boundaries. While lower mammals and avians adhered strictly to Spencean gradient dynamics, higher primates—specifically rhesus macaques, chimpanzees, and other great apes—demonstrated a behavioral plasticity that reflected an intermediate evolutionary state between pure associative conditioning and abstract symbolic cognition.

A critical determinant of whether a non-human primate exhibited absolute gradient decay or stable relational transposition was the animal’s prior learning history, operationalized by Harry Harlow through the concept of Learning Sets (“learning-to-learn”). Harlow demonstrated that when a naive rhesus monkey is trained on a single discrimination problem, its learning proceeds slowly, trial-by-trial, adhering closely to the gradual, absolute habit accumulation curves modeled by Spence. If tested on far transposition at this naive stage, the monkey exhibits the classic distance effect, with relational performance breaking down rapidly.

However, when primates were exposed to hundreds of successive, distinct discrimination problems across months of testing, a profound cognitive transformation occurred. The animals formed a generalized learning set. When presented with a completely novel stimulus pair, an experienced primate could master the discrimination within a single trial, extracting the relevant dimensional rule. Crucially, primates equipped with well-developed learning sets began to exhibit robust transposition across substantially wider physical distances, maintaining relational choices even when the stimuli were shifted significantly along the continuum.

This boundary condition became even more pronounced in modern comparative research involving enculturated, symbol-trained great apes—such as chimpanzees and bonobos trained in lexigrams or sign language. When these symbol-trained primates were evaluated on transposition and relational-matching-to-sample tasks, their behavioral performance mirrored that of older, verbal children. Armed with symbolic cognitive tools, these higher primates completely bypassed the physical decay predicted by Spence’s model, demonstrating true relational transfer across vast physical intervals. These findings indicated that while Spence’s gradient summation represented the universal default architecture of basic sensory-motor conditioning, higher primates possessed latent neural systems capable of abstract relational computation under specific ecological and training conditions.

9. The Intermediate-Size Problem and Other Theoretical Anomalies

9.1 The Structure of the Intermediate-Size Problem

As the Hull-Spence theoretical architecture solidified its dominance, critics sought new experimental paradigms capable of probing the fundamental mathematical limits of gradient summation. The most formidable challenge to emerge from this effort was the intermediate-size problem, a three-stimulus discrimination paradigm first introduced by Gestalt researchers and later systematically analyzed by experimental psychologists throughout the 1940s and 1950s.

In the standard intermediate-size protocol, an organism is confronted with three stimuli simultaneously, arranged along a continuous physical dimension such as surface area. The stimuli are designated as Small (S1), Medium (S2), and Large (S3). Crucially, the intermediate-sized stimulus (S2) is designated as the positive stimulus (S+), rewarded with food. Both the smaller stimulus (S1) and the larger stimulus (S3) are designated as negative stimuli (S-), paired with non-reinforcement. The animal must learn to approach exclusively the intermediate stimulus while rejecting both flanking alternatives.

This three-stimulus arrangement presented an exceptionally demanding test for associative learning models. Instead of a single excitatory peak interacting with a single inhibitory flank, the intermediate-size paradigm required an excitatory potential centered at S2 to be completely bounded on both sides by two flanking inhibitory potentials centered at S1 and S3. Gestalt psychologists heralded this paradigm as the definitive experimentum crucis. They asserted that an animal would readily perceive the emergent relational concept of “intermediateness”—a structural property that cannot be derived from a simple linear directional vector like “larger-than” or “brighter-than.”

Gestalt theory predicted that if the animal masters the concept of “intermediateness,” it should immediately transpose this relational rule when presented with a novel triplet of stimuli—for example, Stimuli S3, S4, and S5 (where S3 was formerly the unreinforced Large stimulus, but now occupies the intermediate position within the novel triplet). Gestalt theory claimed the animal would immediately select S4, the novel intermediate stimulus.

9.2 Empirical Outcomes and Associative Modeling

Kenneth Spence recognized the grave theoretical threat posed by the intermediate-size problem and directly addressed it in a major 1942 theoretical paper titled “The Basis of Solution by Chimpanzees of the Intermediate Size Problem.” Spence applied his algebraic summation mechanics to the three-stimulus configuration. He demonstrated that because S2 is reinforced, a central excitatory gradient develops, centered at S2. Concurrently, non-reinforcement at S1 and S3 generates two separate inhibitory gradients that spread inward from both sides.

When algebraic subtraction is executed (E_net = E – I_S1 – I_S3), the two flanking inhibitory gradients symmetrically compress the central excitatory gradient. Spence proved mathematically that this double subtraction creates an exceptionally narrow, highly peaked, symmetrical island of net positive excitation tightly restricted to the physical locus of S2. Because this net excitatory island is flanked by deep valleys of net negative associative potential on both sides, the animal can master the intermediate-size discrimination through purely absolute associative mechanics.

However, Spence’s mathematical model hit a profound empirical wall when applied to far transposition with novel triplets. Spence’s algebraic calculations deduced that if an animal is tested with novel stimuli that are shifted completely outside the narrow, localized excitatory island of S2, transposition must fail completely. If tested on novel triplets (such as S3, S4, S5, or S7, S8, S9), the net associative values for all three stimuli should be essentially zero, or the animal should exhibit a preference for the stimulus closest to the original training values, resulting in severe choice anomalies.

When empirical experiments were conducted with chimpanzees and monkeys, the results were mixed and contentious. Naive primates and lower mammals frequently failed the intermediate-size transposition test, behaving precisely as Spence’s localized algebraic model predicted. However, highly trained primates—particularly chimpanzees exposed to extensive intermediate-size problems across multiple stimulus sets—demonstrated a remarkable capacity to select the novel intermediate stimulus across novel triplets, completely defying the predictions of Spence’s mathematical model. The intermediate-size problem established an enduring empirical boundary condition, demonstrating that while algebraic summation could account for initial acquisition, it could not fully explain the emergence of an abstract relational concept of “middle” in cognitively advanced species.

9.3 Contextual, Configural, and Background Effects

Beyond the intermediate-size problem, a broader cluster of theoretical anomalies began to accumulate regarding the pervasive influence of contextual and configural cues in discrimination learning. Spence’s classical 1937 model operated upon an idealized assumption: that stimuli reside upon an isolated, abstract physical continuum, devoid of environmental context.

However, experimental psychologists demonstrated that changing the perceptual background or the contextual framing of a stimulus fundamentally altered transposition performance. For example, if two gray stimulus patches were presented against a pure white background during training, but against a pitch-black background during testing, transposition frequently collapsed, even though the absolute physical reflectances of the test patches were identical to those utilized in successful transposition trials. The perceived brightness of a visual stimulus was profoundly modulated by simultaneous luminance contrast with its surrounding background frame—a phenomenon deeply embedded in the physiology of lateral inhibition within the retina.

To address these complex contextual phenomena, post-Hullian theorists were forced to move beyond elementary unidimensional models. John M. Pearce formulated the Configural Theory of Learning, which posited that organisms do not process individual stimulus elements independently, nor do they rely on simple linear algebraic summation. Instead, Pearce proposed that an organism processes the entire sensory array present on any given trial—including the primary stimuli, the background frame, and testing apparatus cues—as a single, integrated configural representation.

Pearce’s configural model demonstrated that generalization occurs between integrated multidimensional configurations based on their overall perceptual similarity, rather than along isolated physical axes. Similarly, Robert Rescorla introduced models of configural conditioning demonstrating that compound stimuli acquire unique associative properties that cannot be predicted from the mere linear addition of their elemental parts. These theoretical developments demonstrated that while Spence’s unidimensional gradient model was historically revolutionary, real-world biological perception relies upon complex, non-linear configural interactions that challenge simple absolute associationism.

10. Synthesizing the Debate: Absolute, Relational, or Dual-Process?

10.1 Evaluating the Explanatory Power of Spence’s Gradient Summation

When Kenneth Spence’s gradient summation theory is evaluated through the philosophy of science, its intellectual brilliance remains undisputed. Spence achieved one of the greatest feats of parsimony in the history of psychology: he took a phenomenon that had been widely accepted as definitive proof of cognitive insight, structural holism, and perceptual abstraction, and demonstrated that it could be fully derived from the interaction of two elementary, physically grounded associative variables: excitation and inhibition.

The explanatory power of Spence’s model rested upon its predictive fruitfulness. A purely post-hoc model merely accommodates known data; a genuine scientific theory predicts novel, unsuspected empirical phenomena. Spence’s mathematical model did not merely account for Köhler’s near transposition; it deductively anticipated the peak shift phenomenon two decades before Guttman and Kalish confirmed it in the laboratory. Furthermore, it uniquely predicted the transposition distance effect and intermediate choice reversals—empirical realities that Gestalt theory had completely failed to anticipate and could not coherently explain.

However, viewing the model with contemporary historical hindsight exposes its clear limitations. Spence’s model was extraordinarily successful within simple, continuous, unidimensional sensory modalities tested in non-linguistic organisms. It performed magnificently when predicting how a pigeon or a rat would choose between two shades of gray, two auditory tones, or two circular discs. But its explanatory power deteriorated rapidly when confronted with categorical rule-learning, multi-dimensional relational matching, abstract concept formation, and the symbolic flexibility of human and higher-primate cognition. Spence had not disproved cognitive relational learning; he had discovered the powerful, low-level associative substrate upon which higher cognitive architectures were ultimately constructed.

10.2 Re-evaluating the Relational Viewpoint

If Spence successfully demonstrated that transposition could emerge from absolute gradients, does this mean Wolfgang Köhler and the Gestalt theorists were entirely wrong? A nuanced examination of modern sensory neurobiology reveals that Köhler’s intuitive phenomenological observations contained profound biological truth, even if his theoretical formulations lacked quantitative rigor.

Gestalt psychologists were fundamentally correct in their rejection of the “constancy hypothesis.” Biological sensory systems do not function as passive, absolute physical meters. From the earliest stages of sensory processing in the retina, cochlea, and primary sensory cortices, the nervous system is physiologically engineered to encode ratios, contrast, and relative differences rather than absolute physical magnitudes. Neural circuits utilize automatic gain control, sensory adaptation, and lateral inhibition precisely to discard absolute illumination levels—which vary wildly between sunlight and shade—in order to preserve invariant surface reflectance ratios (color and lightness constancy).

Therefore, when an animal views two stimuli side-by-side, the afferent signals transmitted to the brain are already heavily filtered through relational neural mechanisms. Early sensory processing computes contrast ratios directly. Köhler’s claim that organisms experience relational vectors was neurobiologically prescient; his error lay in assuming that this perceptual relational coding was synonymous with conscious cognitive insight, and in failing to recognize that even a contrast-sensitive sensory system must lay down associative traces that generalize across physical continua, generating the very gradient dynamics described by Spence.

10.3 Modern Dual-Process Architectures in Animal and Human Learning

The century-long resolution of the absolute versus relational debate has culminated in the widespread adoption of dual-process architectures within contemporary cognitive psychology and comparative neuroscience. Rather than viewing absolute associationism and relational cognitive theory as mutually exclusive, modern science recognizes that both systems operate concurrently within the biological brain, forming a complementary, hierarchical cognitive architecture.

Under this dual-process framework, learning systems can be categorized into two primary functional layers:

• System 1: The Associative-Gradient Substrate (Absolute/Elemental): An evolutionarily ancient, phylogenetically conserved associative system shared across all vertebrates. This system operates automatically, incrementally, and mechanistically, laying down localized excitatory and inhibitory potentials across sensory tuning dimensions. It is modeled with exceptional fidelity by Spencean gradient summation, Rescorla-Wagner associative equations, and modern reinforcement learning algorithms. This system dominates in lower animals, pre-verbal infants, and in adult humans under conditions of high cognitive load or rapid sensory-motor execution.
• System 2: The Cognitive-Relational Engine (Rule-Based/Mediational): An evolutionarily advanced, phylogenetically specialized cognitive system that reaches its zenith in primates, cetaceans, and humans. This system operates through dimensional attention, relational rule induction, working memory, and symbolic/verbal mediation. It extracts abstract structural relations (“larger-than,” “intermediate,” “different”) and applies them invariantly across arbitrary physical coordinates, completely overriding the decay of underlying associative gradients.

The dominance of one system over the other on any given experimental trial is dynamically determined by an array of interacting variables: the phylogenetic sophistication of the organism, its developmental stage, the availability of symbolic language, the extent of prior learning-set training, and the presence of simultaneous versus successive perceptual cues. The historical controversy between Kenneth Spence and Wolfgang Köhler was not a battle between truth and falsehood, but an extended scientific dialectic that illuminated the two primary computational engines governing animal and human intelligence.

11. Computational Models, Neural Networks, and Neurobiological Substrates

11.1 Connectionist and Neural Network Replications of Transposition

The advent of modern computational neuroscience and connectionist modeling provided a rigorous formal platform to test whether Spence’s gradient summation and Gestalt relational learning could be reconciled within artificial neural networks. Computational models have demonstrated that the emergence of transposition, peak shift, and the distance effect are natural emergent properties of multilayer feedforward neural networks trained with reinforcement learning or backpropagation algorithms.

In standard connectionist simulations of discrimination learning, input layers are constructed as an array of continuous, overlapping sensory units possessing localized receptive fields, mimicking the sensory topography of the mammalian cortex. When the network is trained on a simulated simultaneous discrimination task—pairing input unit x₊ with weight increments and input unit x_– with weight decrements—the resulting distribution of connection weights across the hidden and output layers mirrors the exact mathematical geometry of Spence’s excitatory and inhibitory gradients.

When these trained neural networks are subsequently evaluated on novel test inputs, they naturally reproduce the peak shift phenomenon and near transposition. Because the hidden layer units compute weighted sums of their inputs, units adjacent to the positive training coordinate on the side opposite to the negative coordinate inherit high activation potentials unencumbered by negative weights. Furthermore, when connectionist networks are tested on distant input coordinates, their activations decay toward baseline, flawlessly replicating the transposition distance effect.

Crucially, advanced deep neural network architectures—such as recurrent neural networks equipped with self-attention mechanisms (transformers)—have demonstrated how abstract relational rules can emerge naturally from deep associative processing. When exposed to broad, multi-task training environments similar to Harlow’s learning-set protocols, deep networks transition from storing localized, elementistic weight configurations to developing higher-order dimensional vectors that encode abstract relational operations. Modern artificial intelligence has demonstrated that the dichotomy between elementistic associationism and relational abstraction is computationally false; higher-order relational abstractions are the natural, emergent structural products of deep associative architectures.

11.2 Neurobiological Mechanisms of Generalization Gradients

What are the physical, biological substrates within the living brain that correspond to Kenneth Spence’s theoretical constructs of excitatory and inhibitory gradients? Modern neurophysiology has revealed that the physical continuum postulated by Hull and Spence is physically instantiated in the topographic sensory maps of the mammalian brain.

In the primary sensory cortices—the primary visual cortex (V1), primary auditory cortex (A1), and primary somatosensory cortex (S1)—neurons are physically arranged in precise spatial arrays that mirror the physical dimensions of the external world. In the auditory cortex, neurons are organized in a tonotopic map along an axis of acoustic frequency; in the visual cortex, neurons are organized retinotopically along axes of spatial location, orientation, and spatial frequency. Individual sensory neurons within these maps display classic tuning curves: an individual neuron fires maximally to its preferred physical stimulus value (e.g., a specific light wavelength or sound frequency), with its firing rate decaying symmetrically in a bell-shaped, Gaussian distribution as the stimulus deviates from that preferred coordinate. These physiological tuning curves represent the literal neurobiological implementation of Spence’s excitatory generalization gradient.

Furthermore, the physical engine of Spence’s inhibitory gradient has been definitively localized to the micro-circuitry of GABAergic interneuron networks mediating lateral inhibition. When a sensory stimulus is paired with non-reinforcement or punishment, local inhibitory interneurons (such as parvalbumin-positive basket cells) within cortical and subcortical structures release gamma-aminobutyric acid (GABA), actively suppressing pyramidal projection neurons tuned to the negative stimulus coordinate. This active GABAergic suppression forms a spatial field of neurochemical inhibition that spreads laterally across adjacent cortical columns, displaying the exact steep, localized spatial geometry postulated by Spence.

Finally, the reinforcement driving the accumulation of associative habit strength is biochemically mediated by phasic dopaminergic signaling originating in the ventral tegmental area (VTA) and substantia nigra pars compacta. As formalized in modern computational neuroscience by Wolfram Schultz and colleagues, dopamine neurons encode reward prediction errors (RPEs), matching the mathematical functions of the Rescorla-Wagner model. When unexpected reinforcement is delivered at S+, dopamine bursts trigger long-term potentiation (LTP) at active cortical synapses, permanently stamping in habit strength. When reinforcement is withheld at S-, dips in dopamine firing below baseline trigger long-term depression (LTD) and facilitate conditioned inhibition. Spence’s mathematical mechanics of effective associative strength (E_net = E – I) are thus directly realized in the biological balance of glutamatergic synaptic excitation and GABAergic synaptic inhibition across topographically organized cortical maps.

11.3 Hippocampal and Prefrontal Contributions to Relational Abstraction

While primary sensory cortices and striatal dopaminergic circuits provide the neurobiological substrate for Spence’s absolute gradient summation, higher-order brain structures are required to execute abstract relational learning and verbal mediation. Extensive neuroimaging, electrophysiological, and lesion studies have isolated the biological machinery of relational cognition to the coordinated interaction of the hippocampus and the prefrontal cortex (PFC).

The hippocampus and the broader medial temporal lobe system play an indispensable role in relational binding and configural processing. While elementary S-R conditioning can proceed unimpaired following hippocampal destruction, neurobiological research pioneered by Howard Eichenbaum has proven that the hippocampus is uniquely specialized for encoding relational representations. Hippocampal networks create flexible, associative cognitive maps that represent the relationships between stimuli, spatial contexts, and temporal events. When an animal must solve complex transposition tasks—such as the intermediate-size problem or transverse patterning—hippocampal lesioning completely destroys relational transfer, forcing the animal to rely exclusively on rudimentary, absolute cortical generalization gradients.

Simultaneously, the prefrontal cortex—specifically the dorsolateral prefrontal cortex (dlPFC) and the ventrolateral prefrontal cortex (vlPFC)—provides the executive machinery necessary for dimensional attention, rule extraction, and cognitive control. Electrophysiological recordings in primates executing discrimination and transposition tasks demonstrate that prefrontal neurons do not tune their firing to specific physical energy values; instead, PFC neurons exhibit rule-selective firing. An individual prefrontal neuron will fire robustly whenever the active behavioral rule is “choose the larger object,” regardless of whether the stimuli are circles, squares, red objects, or green objects, and regardless of their absolute physical sizes.

Modern functional magnetic resonance imaging (fMRI) studies in humans confirm this neurobiological division of labor. When human subjects solve transposition tasks under conditions where absolute gradient summation dominates, neural activity is localized to primary visual cortices, the dorsal striatum, and the basal ganglia. However, when subjects transition into relational transposition, a massive shift in neural recruitment occurs: the prefrontal cortex, anterior cingulate cortex, and hippocampus become highly engaged, while language areas (Broca’s and Wernicke’s areas) activate to generate covert verbal mediators. The historical transition from Spencean associative conditioning to Gestalt relational insight is thus mapped directly onto the functional neuroanatomy of the mammalian brain.

12. Historical Legacy and Contemporary Relevance of Spence’s Work

12.1 The Methodological Triumph of Formal Quantitative Modeling

The enduring historical significance of Kenneth Spence’s transposition model extends far beyond its immediate resolution of an animal learning puzzle. Spence’s 1936 and 1937 papers represented a watershed moment in the evolution of psychology from a descriptive, qualitative discipline into a rigorous, quantitative natural science. Spence demonstrated how theoretical psychology could be conducted with the mathematical precision, deductive rigor, and falsifiability characteristic of theoretical physics.

Prior to Spence, psychological debates were frequently mired in vague, semantic arguments. Gestalt theorists asserted that animals possessed “insight” and perceived “wholes”; early behaviorists asserted that animals formed “habits.” Neither camp had provided mathematical formulations that could systematically predict the exact probability of an overt behavioral response across a continuous physical spectrum. Spence fundamentally transformed this dynamic. By defining generalization gradients as continuous mathematical functions and establishing linear algebraic summation as an operational rule, Spence replaced qualitative rhetoric with quantitative predictive curves.

Spence’s methodology established the operational template for the grand mathematical models of learning that emerged in the latter half of the twentieth century. The theoretical lineage running from Clark Hull and Kenneth Spence leads directly to the seminal work of Robert Bush and Frederick Mosteller (stochastic models of learning), William K. Estes (stimulus sampling theory), and most decisively, to the legendary Rescorla-Wagner Model of Conditioning (1972). The mathematical formalization of associative competition, delta-rule learning, and modern algorithmic reinforcement learning—which powers current artificial intelligence architectures such as DeepMind’s AlphaZero—traces its intellectual lineage back to the formal deductivism and gradient mechanics pioneered by Kenneth Spence.

12.2 Impact on Educational Psychology, Clinical Conditioning, and Psychometrics

The practical applications derived from Spence’s discrimination and generalization mechanics have reverberated across multiple applied domains, including educational technology, clinical psychology, behavioral therapy, and human factors engineering.

In educational psychology and applied behavior analysis (ABA), Spence’s work on discrimination gradients provided the scientific foundation for stimulus fading procedures and instructional scaffolding. When teaching complex discriminations to individuals with developmental delays or autism spectrum disorders, educators utilize Spencean principles to eliminate trial-and-error frustration. By pairing an easily discriminable stimulus along an established generalization axis and systematically fading physical dimensions in minute, calibrated steps, practitioners can transfer behavioral control across stimulus gradients without inducing the catastrophic behavioral breakdowns that occur when jumping across wide transposition intervals.

In clinical psychology, Spence’s excitatory and inhibitory gradient mechanics provide the theoretical foundation for understanding the etiology and treatment of phobias, post-traumatic stress disorder (PTSD), and clinical extinction. A traumatic event acts as an ultra-potent conditioning experience, establishing a massive, widespread excitatory fear gradient that generalizes broadly across any sensory stimulus physically or contextually resembling the original trauma. Modern exposure therapy relies on systematic desensitization protocols designed to systematically construct overlapping inhibitory gradients through non-reinforced exposure, gradually flattening the pathological reaction potential.

Furthermore, Spence’s peak shift analysis holds profound relevance for contemporary psychometrics, human-computer interaction, and advertising. Marketers and industrial designers exploit peak shift dynamics to create “supernormal stimuli”—designing consumer products, visual displays, and aesthetic logos whose physical features are displaced systematically away from baseline categories in the direction of maximal positive net attraction. In modern psycholinguistics and clinical conditioning, Relational Frame Theory (RFT) has emerged as a direct, sophisticated descendant of the relational-associative dialogue, exploring how verbally competent humans learn to frame relations arbitrarily across stimuli independently of their physical dimensional properties.

12.3 Concluding Synthesis: The Enduring Epistemological Lesson

The epic historical debate over the transposition experiment—commencing with Wolfgang Köhler’s domestic hens on the island of Tenerife and culminating in Kenneth Spence’s elegant mathematical curves at the University of Iowa—stands as one of the most intellectually exhilarating chapters in the history of cognitive science. The enduring epistemological lesson of Spence’s work is both profound and humbling: apparent higher-order cognitive insights can emerge naturally from the mechanical interaction of lower-level associative processes.

When an animal systematically chooses an unknown, novel stimulus over a familiar, reliably reinforced one, the human mind is intuitively compelled to attribute this behavior to conscious insight, structural reasoning, and abstract concept formation. It seems impossible that such an “intelligent,” relational choice could be born of blind, mechanical conditioning. Yet, Kenneth Spence demonstrated that when physical excitation and physical inhibition interact across continuous spatial gradients, the resulting algebraic geometry inevitably forces the organism to choose the novel stimulus. Spence revealed that complexity in behavioral output does not inherently mandate complexity in underlying cognitive architecture.

This insight remains a vital cautionary principle for contemporary artificial intelligence, cognitive neuroscience, and comparative cognition. In an era where deep neural networks and large language models exhibit behavior that mimics human reasoning, creativity, and comprehension, the lesson of Spence warns us against the seductive trap of uncritical anthropomorphism. We must continuously interrogate whether the sophisticated “insights” we observe in biological organisms or artificial agents represent genuine higher-order abstract representations, or whether they are the brilliant, emergent consequences of localized gradient interactions operating across high-dimensional associative spaces.

Ultimately, Kenneth Wartenbee Spence did not diminish the marvel of animal intelligence; he illuminated its mechanistic foundations. By constructing a quantitative bridge between elementistic sensory conditioning and relational behavior, Spence proved that the physicalist principles of natural science could conquer even the most daunting paradoxes of the mind, leaving behind an intellectual legacy that continues to instruct, challenge, and inspire the science of learning.

References

• Guttman, N., & Kalish, H. I. (1956). Discriminability and stimulus generalization. Journal of Experimental Psychology, 51(1), 79–88. https://doi.org/10.1037/h0046219
• Harlow, H. F. (1949). The formation of learning sets. Psychological Review, 56(1), 51–65. https://doi.org/10.1037/h0062474
• Hull, C. L. (1943). Principles of Behavior: An Introduction to Behavior Theory. Appleton-Century-Crofts.
• Kendler, H. H., & Kendler, T. S. (1962). Vertical and horizontal processes in problem solving. Psychological Review, 69(1), 1–16. https://doi.org/10.1037/h0044678
• Köhler, W. (1918). Nachweis einfacher Strukturfunktionen beim Schimpansen und beim Haushuhn: Über eine neue Methode zur Untersuchung des bunten Farbensystems. Abhandlungen der Königlich Preußischen Akademie der Wissenschaften, 2, 1–101.
• Köhler, W. (1925). The Mentality of Apes (E. Winter, Trans.). Harcourt, Brace & World.
• Kuenne, M. R. (1946). Experimental investigation of the relation of language to transposition behavior in young children. Journal of Experimental Psychology, 36(6), 471–490. https://doi.org/10.1037/h0055743
• Pearce, J. M. (1987). A model for stimulus generalization in Pavlovian conditioning. Psychological Review, 94(1), 61–73. https://doi.org/10.1037/0033-295X.94.1.61
• Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical Conditioning II: Current Research and Theory (pp. 64–99). Appleton-Century-Crofts.
• Spence, K. W. (1936). The nature of discrimination learning in animals. Psychological Review, 43(5), 427–449. https://doi.org/10.1037/h0056975
• Spence, K. W. (1937). The differential response in animals to stimuli varying within a single dimension. Psychological Review, 44(5), 430–444. https://doi.org/10.1037/h0062885
• Spence, K. W. (1942). The basis of solution by chimpanzees of the intermediate size problem. Journal of Experimental Psychology, 31(4), 257–271. https://doi.org/10.1037/h0063239