Core Knowledge Theory – Elizabeth Spelke

How does the human mind construct coherent, structured representations of reality out of the blooming, buzzing confusion of sensory experience? For centuries, classical philosophy oscillated between the radical empiricist assumption that the mind begins as a blank slate and the extreme rationalist conviction that mature metaphysical truths exist fully formed within the soul. In contemporary cognitive science, this foundational inquiry has been fundamentally transformed through the empirical paradigm known as Core Knowledge Theory. Formulated and exhaustively developed by cognitive psychologist Elizabeth Spelke and her collaborators, this theoretical framework posits that human infants do not enter the world intellectually destitute, nor do they rely solely on open-ended, domain-general statistical associations. Instead, human cognition is underwritten by a set of phylogenetically ancient, ontogenetically early-emerging, and domain-specific representational engines designated as core systems.

Spelke’s empirical program has dismantled long-standing dogmas inherited from both classical behaviorism and Piagetian constructivism. Where Jean Piaget posited that infants spend their initial years slowly piecing together the concepts of objecthood, causality, and permanence through arduous sensorimotor exploration, Spelke demonstrated that infants possess rich, deductively structured conceptual expectations within the first hours, days, and months of life. Through ingenious experimental protocols such as the Violation-of-Expectation (VOE) paradigm, Spelke revealed that pre-verbal infants reliably expect physical objects to be cohesive, solid, and continuous, perceive animate agents as goal-directed entities bound by the principle of rational efficiency, and navigate spatial and numerical environments via dedicated computational routines. These core domains are not mere perceptual reflexes; they are conceptual frameworks that parse the sensory continuum into distinct ontological categories.

Crucially, Core Knowledge Theory provides a coherent solution to the paradox of human epistemic uniqueness: how does a species equipped with biologically constrained, phylogenetically conserved cognitive primitives construct abstract geometry, non-Euclidean mathematics, formal physical science, and complex institutional societies? The answer, Spelke argues, lies in the combinatorial superpower of natural language. By serving as an open-ended, compositional medium of integration, language bridges the informational boundaries isolating separate core modules, permitting mature human minds to weave ancient, specialized computational systems into flexible, novel, and culturally transmissible conceptual architectures. What follows is an exhaustive, systematic examination of Core Knowledge Theory—its epistemological roots, empirical architectures, comparative neurobiology, critical critiques, and radical implications for modern education and artificial general intelligence.

1. Introduction to Core Knowledge Theory and Epistemological Foundations

1.1 Historical Context and Epistemological Lineage

The philosophical genesis of Core Knowledge Theory traces its lineage directly through the long-standing debate between classical rationalism and Lockean empiricism. In the seventeenth and eighteenth centuries, John Locke advanced the radical thesis that the human mind originates as a tabula rasa—a blank slate devoid of innate characters, marks, or ideas—upon which sensory impressions inscribe all subsequent knowledge through perceptual exposure, associative contiguity, and inductive habit. Conversely, René Descartes and later Immanuel Kant argued that raw sensory input is fundamentally chaotic, underdetermined, and uninterpretable without antecedent conceptual structures. In the Critique of Pure Reason, Kant established that the human experience of sensible reality is structurally conditioned by a priori forms of sensible intuition—namely, space and time—alongside pure concepts of the understanding, such as causality, substance, and unity. Spelke’s work can be conceptualized as an empirical translation and phylogenetic naturalization of Kantian synthetic a priori cognition, grounded firmly within contemporary evolutionary biology and cognitive science.

This rationalist resurgence gained modern empirical momentum during the cognitive revolution of the mid-twentieth century, spearheaded by Noam Chomsky’s transformative critique of behaviorism. Chomsky’s poverty-of-the-stimulus argument demonstrated that the linguistic environment of the developing child is far too fragmented, degenerate, and finite to account for the rapid, uniform, and error-resistant acquisition of complex generative grammar through general inductive learning. Chomsky demonstrated that children must possess an innate Universal Grammar—a biologically endowed Language Acquisition Device (LAD)—that severely restricts the hypotheses the infant mind entertains regarding linguistic syntax. Elizabeth Spelke systematically recognized that Chomsky’s poverty-of-the-stimulus logic applies with equal force to physical, mechanical, spatial, and numerical domains. Raw retinal projections do not explicitly code for the boundary lines of persistent three-dimensional objects, nor do sensory receptors directly record numerical cardinality or physical contact causality; the mind must contribute internal operational constraints to make sense of these sensory signals.

Consequently, Spelke parted company with Jean Piaget’s prevailing sensorimotor constructivism. Piaget had proposed that the human neonate begins life in a condition of radical sensorimotor solipsism, gradually assembling conceptual notions of object permanence, spatial coordinates, and physical causality across sequential, domain-general developmental stages driven by physical grasping, visual tracking, and manual manipulation. Piaget contended that object permanence does not solidify until approximately eight to twelve months of age, with complete mental representation culminating only around eighteen to twenty-four months. Spelke, armed with precision experimental methodologies, exposed the motoric confound underlying Piaget’s manual search tasks: infants fail manual search not because they lack conceptual representations of permanent objects, but because of immature executive function and motor-planning limitations. By liberating cognitive measurement from motoric performance, Spelke demonstrated that conceptual competence predates sensorimotor mastery.

This fundamental conceptual divergence shifted developmental psychology away from domain-general models of cognitive development toward domain-specific modular architectures. While classical structuralists envisioned a uniform intellectual engine operating symmetrically across all epistemic challenges, Spelke formulated an architectural framework wherein the mind comprises several independent, content-rich, domain-specific modules. Rather than a singular, generalized learning algorithm attempting to parse physical collisions, spatial geometry, and conspecific interactions through identical statistical calculations, the brain relies on specialized computational systems that evolved to solve specific, recurring ecological challenges encountered by ancestral organisms.

1.2 Defining the Core Knowledge Paradigm

The core knowledge paradigm rests upon several structural postulates that distinguish it from both radical nativism and connectionist empiricism. Foremost among these is the principle of domain specificity: each core system operates over a strictly delimited class of environmental entities and informational inputs. The computational algorithms governing core physical mechanics process information exclusively regarding bounded, physical objects, their spatial trajectories, and mechanical interactions. These algorithms are blind to social identity, communicative intent, or emotional valence. Conversely, the core system of agency calculates teleological efficiency and goal-directed trajectories for animate beings while remaining largely indifferent to geometric shape or inert physical boundaries. This domain-specific parsing ensures that the developing mind executes rapid, targeted inferences without becoming paralyzed by computational combinatorial explosions.

A second signature property of core knowledge systems is their innate evolutionary grounding and early ontogenetic emergence. Far from being acquired through extended schedules of trial-and-error conditioning or cultural transmission, core systems are present in fully functional, primitive states at the very threshold of postnatal life. Extensive experimental paradigms demonstrate that human neonates—within minutes, hours, or days of birth—exhibit signature sensitivities to physical boundaries, numerical disparities, facial geometries, and biological motion trajectories. These capacities reflect phylogenetically deep, evolutionarily selected adaptations designed to maximize survival by providing instant, reliable interpretations of essential ecological invariants.

Third, and perhaps most radically within developmental psychology, Spelke posits the invariance and permanent persistence of core systems. In traditional developmental models, early cognitive stages are superseded, overwritten, or structurally dismantled as children mature into concrete and formal operational thought. Spelke asserts that core systems are not temporary developmental scaffolding to be cast aside; they persist entirely unaltered and functional throughout human adulthood. The exact same computational constraints governing an infant’s perception of object cohesion, solidity, and numerical magnitude remain active in the adult neurocognitive architecture, functioning underneath culturally acquired beliefs and formal symbolic systems.

Fourth, core systems exhibit profound informational encapsulation. Borrowing from Jerry Fodor’s modularity thesis, Spelke underscores that core representations are cognitively impenetrable to conscious, high-level theoretical beliefs. Even after an adult attains a doctorate in quantum mechanics and conceptually accepts that a solid table is mostly empty space populated by subatomic particles governed by probabilistic wave functions, their core physical system continues to perceive and anticipate the table as an impenetrable, cohesive, continuously existing physical entity. Finally, core systems operate as foundational building blocks. They do not constitute the totality of human cognition; rather, they serve as dedicated conceptual anchors upon which formal education, abstract symbolic thinking, and advanced cultural transmission securely anchor.

1.3 Nativism in the Cognitive Sciences

To accurately situate Elizabeth Spelke’s contribution within contemporary cognitive science, one must rigorously distinguish her constrained developmental nativism from the radical, unconstrained nativism historically associated with Jerry Fodor. Fodor notoriously argued that virtually all conceptual primitives—including abstract notions such as “carburetor” or “bureaucrat”—must be innate, since genuine conceptual novelty cannot be logically generated via inductive hypothesis testing. Spelke, by contrast, rejects such pan-nativism. She restricts the scope of innate endowments to a small, privileged set of foundational domains: inanimate objects, intentional agents, approximate magnitudes, environmental geometry, and social partners. Spelke does not argue that humans are born with innate concepts of airplanes, microchips, or constitutional law; rather, she posits that an initial kit of primitive core systems provides the requisite representational infrastructure from which cultural and conceptual innovations emerge.

Furthermore, modern developmental nativism thoroughly repudiates the simplistic, deterministic framing of the historical nature-versus-nurture debate. Spelke does not present core knowledge as an immutable, genetic blueprint executing in isolation from environmental interaction. Instead, core knowledge represents a set of biologically endowed computational constraints that selectively guide attention, prioritize sensory processing, and constrain learning from the environment. Development is an epigenetic, transactional process: the innate cognitive architecture requires species-typical sensory and environmental affordances to stabilize, calibrate, and express its computational algorithms.

By conceptualizing innateness as a system of generative constraints rather than pre-fabricated knowledge databases, Spelke resolves the evolutionary puzzle of developmental plasticity. Human infants are capable of thriving across radically diverse ecological, geographic, and cultural environments precisely because their core cognitive systems provide universal, stable ontological anchors. The physical properties of terrestrial objects, the geometric logic of navigation, and the goal-directed nature of animate creatures are ecological constants across our planet. By biologically hardwiring computational sensitivity to these evolutionary invariants, natural selection endowed the human lineage with a resilient foundation upon which cultural learning can build.

2. Foundational Architecture and Key Postulates of Core Knowledge

2.1 Domain Specificity and Functional Autonomy

The architecture of core knowledge is fundamentally modular, defined by distinct neural and computational boundaries that divide knowledge into functional domains. Unlike a domain-general computational architecture, which applies an identical set of statistical pattern-recognition algorithms across visual scenes, phonetic transitions, mechanical interactions, and linguistic utterances, a domain-specific system possesses algorithms fine-tuned exclusively for a particular ontological class. In Spelke’s model, the computational mechanisms deployed to track an inanimate physical object through space and time are functionally segregated from the mechanisms designed to predict the teleological actions of a social agent or estimate the non-symbolic cardinality of a visual array.

This functional autonomy reflects distinct evolutionary trajectories shaped by disparate ecological pressures. The adaptive challenge of predicting where a dislodged rock will tumble down a steep incline demands computational algorithms prioritizing mass, trajectory, continuous spatiotemporal paths, and mechanical impact forces. Conversely, calculating whether an approaching predator will turn left or right requires a specialized computational framework prioritizing internal goals, perceptual access, biological kinematics, and metabolic energy efficiency. Attempting to process these disparate ecological phenomena through an undifferentiated cognitive algorithm would generate severe computational bottlenecks and lethal delays in behavioral response.

Crucially, this domain-specific segregation manifests as informational encapsulation. The core physical module operates automatically, involuntarily, and independently of the intentional states or emotional expressions of observed entities. For instance, if an animated human face is shown behaving like an inanimate physical projectile—colliding with a barrier and ricocheting according to the laws of Newtonian mechanics—the core physical system evaluates the spatial kinematics entirely separate from the facial identity or communicative expressions displayed. This informational modularity prevents cognitive cross-talk, ensuring that baseline predictions regarding physical survivability and mechanical stability remain robust and impervious to transient social or emotional dynamics.

2.2 Invariance and Ontogenetic Continuity

One of the central, empirically verifiable claims of Core Knowledge Theory is the Spelke-Carey continuity hypothesis, formulated in close dialogue with developmental psychologist Susan Carey. This hypothesis posits that the core representations identified in pre-verbal infants do not vanish, metamorphose, or undergo structural replacement over the course of human ontogeny. Instead, they remain structurally intact, functionally active, and computationally invariant across the entire lifespan. While adult humans acquire extensive theoretical knowledge, formal education, and abstract symbolic notations, the initial core systems persist beneath these cultural superstructures, operating as permanent sub-modules within the adult cognitive architecture.

Empirical support for ontogenetic continuity emerges directly from adult psychophysics and cognitive neuroscience. When adult human participants are subjected to rapid-fire visual tasks under demanding cognitive load—wherein conscious executive working memory, linguistic verbalization, and deliberate analytical reasoning are computationally suppressed—their intuitive judgments revert entirely to the primitive principles of core knowledge. For example, in split-second physical reasoning tasks, highly educated adults consistently exhibit systematic biases that mirror infant core physical constraints (such as expecting moving objects to follow continuous straight paths and experiencing difficulty tracking objects that violate cohesion or solidity), despite their formal academic mastery of advanced physics.

The continuity hypothesis decisively refutes developmental stage theories that propose sweeping, qualitative restructuring of mental faculties. While children undeniably acquire new, transformative concepts—such as rational numbers, negative integers, evolutionary natural selection, and electromagnetic radiation—these conceptual expansions do not replace core intuitions. Rather, they form distinct, reflective cultural theories that must constantly navigate, accommodate, or suppress the unyielding, foundational intuitions continuously projected by the underlying core systems.

2.3 Phylogenetic Primacy and Evolutionary Homology

If core knowledge mechanisms are deeply entrenched products of natural selection, they should exhibit phylogenetic primacy: their computational foundations should be observable across non-human animal taxa that share evolutionary history and ecological niches with our lineage. Research in comparative cognition has systematically confirmed this evolutionary prediction. Homologous manifestations of core physical mechanics, numerical magnitude discrimination, agent goal-attribution, and environmental geometry are readily identifiable across non-human primates, domestic chicks, rodents, and even non-mammalian marine species.

The presence of these shared cognitive architectures in species separated by hundreds of millions of years of divergent evolution demonstrates that core systems are ancient adaptations. An organism capable of instinctively parsing solid, bounded physical objects from background noise enjoys an immense selective advantage when foraging, fleeing predators, or navigating treacherous terrain. Similarly, an animal possessing an innate Approximate Number System can instantly assess whether a competing conspecific coalition outnumbers its own, or calculate which foraging patch offers a higher caloric density, without wasting precious seconds manually counting individuals.

Consequently, the cognitive capacities observed in human infants are not newly invented evolutionary novelties unique to Homo sapiens. Instead, humanity inherited a rich, highly conserved neurocomputational toolset shared with diverse branches of the phylogenetic tree. Spelke’s empirical breakthrough was recognizing that the unique intellectual achievements of our species do not stem from a total reconfiguration of these foundational modules, but from the novel evolutionary emergence of an integrative interface that connects these ancient, isolated animal modules into an interconnected, generative cognitive whole.

3. Methodological Breakthroughs: Operationalizing Infant Cognition

3.1 The Violation-of-Expectation (VOE) Paradigm

Investigating the internal mental landscape of pre-verbal, pre-motoric infants required an epistemological and methodological revolution. For decades, classical experimental psychology erroneously conflated motor capability with conceptual competence. Because neonates and young infants could not speak, reach, grasp, point, or execute coordinated manual searches, they were categorized as cognitively vacant or restricted strictly to momentary perceptual reflexes. Elizabeth Spelke, alongside pioneering collaborators such as Renée Baillargeon, bypassed this motoric bottleneck by operationalizing looking time via the Violation-of-Expectation (VOE) paradigm, paired with sophisticated habituation-dishabituation techniques.

The VOE paradigm exploits a fundamental, universal neurobiological drive: the orienting reflex toward novelty and conceptual discrepancy. When presented with visual displays that conform to natural, physically lawful regularities, infants quickly habituate, exhibiting progressively declining gaze durations as the stimulus becomes predictable and cognitively redundant. However, if the infant is subsequently exposed to two distinct test events—one that is visually novel but physically possible, and another that is visually familiar but physically impossible—their looking behavior provides an unmediated window into their cognitive expectations. If the infant looks significantly longer at the physically impossible event, researchers can decisively infer that the infant detected a violation of a conceptual expectation, rather than reacting to superficial perceptual changes.

Spelke designed rigorous experimental controls to disentangle low-level sensory novelty from high-level conceptual surprise. In classic occlusion displays, an infant is habituated to an object moving smoothly behind a screen and reappearing on the opposite side. During test events, the visual trajectories are meticulously balanced: one display features an unexpected physical impossibility (e.g., an object appearing to pass magically through an occluded barrier or vanishing and reappearing across a gap without occupying intermediate space), while the alternative display introduces striking perceptual variations (such as alterations in surface texture, color, or speed) that fully preserve core physical mechanics. Infants systematically demonstrate prolonged, elevated gaze durations specifically targeted at the impossible physical violations, confirming that their mental representations operate at the level of underlying, abstract principles rather than surface-level sensory features.

3.2 Eye-Tracking Technologies and Pupillometry

While early core knowledge studies relied on manual video recording paired with blind, frame-by-frame gaze coding conducted by independent observers, the twenty-first century introduced advanced micro-metric, high-speed infrared eye-tracking technologies. These instruments sample corneal reflections and pupil positions hundreds of times per second, enabling developmental cognitive scientists to generate granular, high-density scanpaths that chart the micro-structure of infant visual exploration. Scanpath analysis reveals not merely how long an infant looks at an unexpected display, but the precise spatial and temporal sequence of their predictive visual fixations.

Predictive gaze tracking provides definitive proof of forward-looking conceptual models. When an infant observes an object sliding rapidly toward an occluder, an eye-tracker records whether the infant’s gaze remains passively anchored to the object’s point of disappearance or predictive saccades land forward to the far edge of the occluder in anticipation of the object’s emergence. If the object fails to emerge within the lawful temporal window determined by its velocity and distance, the infant’s eye-movement patterns exhibit rapid corrective saccades, visual scanning back and forth across the occluding barrier, and physiological indices of disorientation. These predictive saccades demonstrate that the infant mind continuously computes forward simulations of physical events governed by internal mechanistic rules.

Furthermore, developmental laboratories integrate pupillometry as an autonomous physiological biomarker of cognitive dissonance, surprise, and mental effort. The pupillary aperture is innervated by the autonomic nervous system, mediated through locus coeruleus norepinephrine projections that respond instantly to unexpected occurrences. When pre-verbal infants witness an event that violates core physical principles—such as an object miraculously hovering in mid-air without support or two solid masses interpenetrating—their pupils undergo measurable dilation that correlates directly with the magnitude of the structural violation. This physiological metric provides objective verification of internal cognitive surprise, entirely free from human experimenter bias.

3.3 Neural Measures: fNIRS and Infant Electroencephalography

To directly connect behavioral looking-time data to neurobiological substrates, cognitive developmental neuroscience employs cutting-edge functional neuroimaging tailored specifically for awake, active infants. Chief among these methods is Functional Near-Infrared Spectroscopy (fNIRS). Traditional functional Magnetic Resonance Imaging (fMRI) demands that participants remain entirely motionless inside a confined, loud, and claustrophobic magnetic bore—an environment fundamentally incompatible with young, awake infants. In contrast, fNIRS utilizes lightweight, flexible optical caps embedded with near-infrared light emitters and detectors to non-invasively measure cortical hemodynamic changes by tracking the differential absorption spectra of oxygenated and deoxygenated hemoglobin.

Using fNIRS, researchers have successfully localized domain-specific cortical activity in infants as young as two to three months of age. When infants view displays violating core physical continuity or solidity, fNIRS identifies localized activations within the bilateral parietal and frontal cortices—regions known in human adults to mediate visuospatial manipulation and intuitive physics. Conversely, when infants process goal-directed social actions, distinct activations illuminate the temporal-parietal junction (TPJ) and superior temporal sulcus (STS), demonstrating that the functional partitioning predicted by Core Knowledge Theory is biologically reflected in localized cortical specialization during early infancy.

Complementing hemodynamic measures, high-density infant electroencephalography (EEG) and Event-Related Potentials (ERPs) capture the millisecond-by-millisecond temporal dynamics of infant cognitive operations. Neuroscientists monitor specific electrophysiological signatures, such as the Negative central (Nc) wave—an ERP component maximal over frontal and central scalp locations typically elicited between 400 and 800 milliseconds post-stimulus onset, which indexes attentional allocation, orienting, and conceptual novelty. When an infant witnesses a physical impossibility, such as an occluded object vanishing into nothingness, high-density EEG arrays record a sharp, statistically robust modulation of the Nc component. These electrophysiological markers prove that conceptual surprise occurs with lightning-fast neural timing, debunking the claim that looking-time paradigms merely capture sluggish, post-hoc perceptual fatigue.

4. Core System 1: Inanimate Objects and Physical Mechanics

4.1 The Cohesion Principle

The first core system delineated by Elizabeth Spelke is the system of inanimate physical objects, which governs our intuitive comprehension of the mechanics of the material world. Central to this system is the Cohesion Principle, which posits that the infant mind conceptualizes physical objects as bounded, internally unified wholes that maintain their physical integrity and move as connected entities. According to this computational constraint, a physical object cannot spontaneously rupture into disparate, uncoordinated fragments, nor can two distinct objects spontaneously merge into an undifferentiated mass, unless subjected to overt, externally applied mechanical forces.

Spelke established this principle through foundational habituation experiments involving multi-component displays. In these experiments, infants were exposed to an apparatus featuring two adjacent, visually distinct segments. When the segments moved synchronously together across a visual field, infants uniformly parsed the array as a single, cohesive, unified object, expecting the two halves to remain structurally linked if lifted or manipulated. Conversely, if the two halves exhibited independent, asynchronous motion paths, infants immediately partitioned the visual display into two distinct, autonomous objects. Infants demonstrate profound surprise—indexed by sustained VOE looking times—if an entity that previously displayed rigid, unified motion spontaneously dissociates into independent pieces when pulled or translated.

Crucially, this cohesion constraint establishes a clear ontological boundary between solid, cohesive physical bodies and non-cohesive substances, such as sand, fluids, or powders. Studies conducted by Spelke, Susan Carey, and their associates demonstrate that infants do not apply the cohesion principle to non-rigid materials. If an infant observes a pourable liquid or a pile of sand being partitioned, they do not register conceptual surprise when it separates, pours around obstacles, or splits into discontinuous pools. The core object system is finely tuned specifically to solid, cohesive bodies, confirming that infant physics is organized around discrete, tangible entities rather than generic, undifferentiated visual textures.

4.2 The Continuity Principle

The second inviolable law governing infant intuitive physics is the Continuity Principle. This principle specifies that physical objects exist permanently in time and travel strictly along spatiotemporally continuous, unbroken trajectories through space. An object cannot simply cease to exist at one point in space and instantaneously materialize at another spatial coordinate without traveling through all intermediate points. Nor can an object pass directly through an opaque spatial barrier without displacing or disrupting that barrier.

In her classic occlusion experiments, Spelke decisively overturned Piaget’s dogma regarding object permanence. Piaget maintained that an object hidden behind an opaque screen ceases to exist for an infant younger than eight months. Spelke refuted this by presenting young infants with an experimental setup featuring two wide, spatially separated occlusion screens placed side by side. An object rolled smoothly behind the first screen, remained unseen while traversing the empty gap between the two screens, and emerged from behind the second screen. Infants exposed to this possible, continuous trajectory readily habituated. However, in the impossible test condition, the object disappeared behind the first screen, was completely absent within the open gap between the two screens, and yet emerged normally from behind the second screen.

Despite the visual simplicity of the trick, infants as young as three to four months of age stared with sustained disbelief at the impossible event. Their looking-time profile revealed that their internal cognitive architecture tracked the continuous spatiotemporal trajectory of the hidden object even when it was obscured from retinal view. The infants expected the object to maintain its continuous identity across space and time; its failure to appear in the visible gap constituted a direct violation of the continuity constraint. This established that pre-verbal infants possess robust, continuous representational tracking of hidden entities, entirely dismantling the claim that human object representations depend upon manual reaching capabilities.

4.3 The Contact Principle and Solidity

The third cornerstone of core physical mechanics comprises the Contact Principle and the Solidity Postulate. The Contact Principle establishes that inanimate physical bodies cannot act upon one another at a distance without physical mediation; mechanical movement or change of state requires direct, unmediated surface-to-surface contact. If a stationary billiard ball abruptly rolls forward the exact instant an approaching ball stops three inches short of it, infants register immediate conceptual dissonance. They expect an inanimate object to remain stationary unless mechanically impacted by an external physical force, reflecting an intuitive grasp of Newtonian action-at-a-contact long before any formal academic instruction.

Interlinked with contact is the absolute constraint of Solidity, which mandates that two physical bodies cannot occupy the same spatial coordinates simultaneously. Renée Baillargeon and Elizabeth Spelke operationalized this principle in famous drawbridge paradigms. Infants were habituated to a solid wooden panel that rotated backward and forward across a 180-degree arc like a drawbridge. Subsequently, a solid, three-dimensional wooden block was placed directly in the path of the rotating panel. In the physically possible test event, the panel rotated upward, struck the wooden block, and halted abruptly at the expected mechanical angle (e.g., 112 degrees). In the impossible test event, an optical trick permitted the panel to continue rotating smoothly through the space fully occupied by the solid block, completing its full 180-degree flat trajectory.

Infants looked significantly longer at the impossible event where the rotating screen appeared to pass directly through the solid block. This selective attention demonstrated that infants mentally calculate the spatial volume and unyielding solidity of hidden objects, projecting that one solid mass cannot interpenetrate or occupy the space of another. However, developmental research also illuminates the strict limitations of this core system. While infants possess innate, unshakeable commitments to cohesion, continuity, contact, and solidity, they do not possess an innate, mature grasp of continuous parameters such as gravity, exact center of mass, or fractional support relations. Infants must gradually learn—through structured observation—the exact proportions of base support an object requires before toppling, demonstrating where innate core constraints end and empirical learning begins.

5. Core System 2: Agents, Teleology, and Goal-Directed Action

5.1 Differentiating Agents from Inanimate Matter

While the first core system governs the unyielding, mechanical behavior of inanimate objects, the second core system is dedicated to understanding agents and intentional actions. Human survival depends upon the ability to rapidly distinguish passive physical matter from active, animate living beings. The infant visual system detects agents through primitive, highly sensitive perceptual signatures, the most prominent being self-generated motion and non-rigid, biological kinematics. If an entity initiates motion spontaneously without prior physical contact from an external object, infants immediately categorize it as an agent governed by an internal locus of energy and intentionality.

This fundamental distinction between objects and agents radically alters the computational rules the infant mind deploys. While infants strictly demand physical contact causality for inanimate entities, they effortlessly accept and expect action-at-a-distance when interpreting agents. An agent can pause, change direction, or react to another distant agent through vocalizations, visual gestures, or posturing without requiring mechanical physical contact. The infant mind readily assigns separate ontological status to entities possessing biological features—such as eyes, faces, and reciprocal social vocalizations—exempting them from the mechanical constraints of inert physics.

Studies utilizing point-light displays demonstrate that infants within days of birth preferentially attend to biological motion patterns over inverted or scrambled kinematic displays. The infant brain is evolutionarily primed to detect the structural trajectories of animate conspecifics and predators. When an entity possesses eyes, infants track the direction of its gaze to infer its internal attentional focus, demonstrating that the agent system instantly activates an interpretive framework directed toward internal psychological orientations rather than external physical collisions.

5.2 The Principle of Rational Efficiency

Once the infant mind categorizes an entity as an animate agent, its actions are systematically interpreted through what developmental psychologists Gergely Csibra and György Gergely define as the teleological stance, governed by the Principle of Rational Efficiency. This principle specifies that agents are expected to pursue their internal goals through the most efficient, cost-effective spatial pathways permissible within the physical constraints of their environment. Agents do not execute arbitrary, random, or energetically wasteful trajectories; they navigate the world rationally.

In Csibra and Gergely’s landmark detouring experiments, infants were habituated to an animated small circle that hopped over a tall rectangular barrier to reach a larger circle on the other side. Because a physical obstruction stood between the two agents, the curved, arching trajectory of the small circle was entirely rational and efficient. Following habituation, the central barrier was completely removed from the environment, leaving an open, unobstructed horizontal path between the two agents. The infants were then presented with two alternative test events: (1) the small circle traversed a straight, direct, flat horizontal path to the other circle (a physically novel trajectory, but teleologically efficient), or (2) the small circle executed the exact same upward hopping, curved trajectory it had performed during habituation, despite the complete absence of any barrier.

Strikingly, infants looked significantly longer at the arching, hopping trajectory. Even though this jumping action was visually identical to the habituated display, it was teleologically irrational and mechanically inefficient in an open environment lacking physical obstacles. The infants expected the agent to immediately modify its trajectory toward the most direct, efficient path. This elegant finding proved that pre-verbal infants do not simply encode surface-level kinematic shapes; they construct high-level teleological representations regarding unobserved goals, assessing whether an observed trajectory represents a rational expenditure of kinetic energy toward a targeted outcome.

5.3 Attribution of Mental States and Intentionality

The core agent system serves as the foundational phylogenetic precursor to mature Theory of Mind (ToM). Long before children master explicit, linguistically mediated false-belief tasks (typically achieved around four years of age), pre-verbal infants demonstrate sophisticated sensitivities to the implicit intentional states, attentional foci, and moral valence of interactive agents. Infants instinctively interpret pointing gestures and directional gaze saccades not merely as geometric lines in space, but as communicative signals referring to specific external targets of interest.

Furthermore, the agent system evaluates the qualitative, social dimensions of interpersonal interactions. In groundbreaking research conducted by Kiley Hamlin, Karen Wynn, and Paul Bloom, infants were exposed to puppet-based social scenarios featuring a character attempting to climb a steep hill. In the cooperative condition, a “helper” character gently pushed the climber up the incline from behind; in the antagonistic condition, a “hinderer” character deliberately collided with the climber from above, knocking it back down the slope. Following the presentation, infants were offered a direct choice between the two characters.

Infants as young as three to six months of age systematically exhibited robust preferences for the prosocial helper over the antisocial hinderer, as indexed by both manual reaching choices and preferential looking durations. Furthermore, when infants witnessed subsequent interactions where the climber chose to approach either the helper or the hinderer, they registered elevated looking times if the climber approached the hinderer, indicating an expectation that rational agents affiliate with cooperative rather than hostile entities. These remarkable discoveries demonstrate that the core agent system does not merely compute physical navigation paths; it encodes the social and normative dimensions of interaction, providing the primitive bedrock upon which human moral psychology and cooperative social institutions are constructed.

6. Core System 3: Number Representation and Pre-Verbal Arithmetic

6.1 The Approximate Number System (ANS)

The third core cognitive system identified by Spelke and extensively explored alongside cognitive neuroscientist Stanislas Dehaene is the Approximate Number System (ANS). The ANS is an evolutionary ancient, non-verbal, non-symbolic computational cognitive system dedicated to representing numerical magnitude across space and time. Long before children acquire verbal counting sequences or formal mathematical symbols (such as Arabic numerals “1, 2, 3”), they possess an internal analog magnitude system that computes relative quantities across visual, auditory, and tactile modalities.

A defining computational hallmark of the Approximate Number System is its strict adherence to Weber-Fechner’s Law: the discriminability of two numerical quantities is governed not by their absolute numerical difference, but by the proportional ratio between them. For example, discriminating between 10 and 20 items requires the exact same proportional computation as discriminating between 50 and 100 items (a 1:2 ratio), despite the massive difference in absolute quantity. The mental representation of number within the ANS takes the form of noisy, overlapping Gaussian curves aligned along an internal, logarithmically compressed mental number line.

Developmental psychophysics reveals a predictable, universally conserved ontogenetic maturation of the ANS ratio threshold throughout human life. Human neonates within hours of birth successfully discriminate between arrays of visual items or auditory tones that contrast at a coarse 1:3 ratio (e.g., 4 versus 12 sounds). By six months of age, this perceptual resolution sharpens to an exact 1:2 ratio (e.g., 8 versus 16 items), but infants fail at a tighter 2:3 ratio (e.g., 8 versus 12). By nine to twelve months, the threshold refines to a 2:3 ratio; by childhood, it achieves a 5:6 ratio; and in typical human adults, it reaches an optimal discriminative precision of approximately 9:10 to 10:11. This continuous, quantifiable developmental trajectory occurs prior to and largely independent of formal mathematical instruction, proving the existence of an endogenous, continuous magnitude metric.

6.2 The Object Tracking System (OTS)

Alongside the approximate magnitude system, the human mind possesses a radically different, parallel numerical architecture known as the Object Tracking System (OTS), sometimes referred to as the subitizing or parallel individuation mechanism. Rather than generating noisy, statistical estimations of aggregate magnitude, the OTS relies upon visual-spatial working memory indexes—often conceptualized as Zenon Pylyshyn’s “visual fingers” or FINSTs (Fingers of Instantiation)—to simultaneously lock onto, track, and precisely enumerate individual, discrete physical objects.

The operational signature of the Object Tracking System is its exceptional precision combined with a rigid, non-negotiable capacity limit: it can track exactly three to four items, and absolutely no more. In pre-verbal infants, this capacity threshold is strictly bounded at three. In classic behavioral paradigms, infants watch an experimenter place two or three crackers into an opaque container, and one or two crackers into a second container. Infants reliably choose the container holding the larger quantity, demonstrating exact numerical evaluation. However, the moment the quantity placed into one container exceeds the capacity limit of the system—such as comparing 2 crackers against 4, or 3 against 6—infants perform completely at chance.

This dramatic performance collapse highlights the qualitative dissociation between the OTS and the ANS. If infants were utilizing the Approximate Number System in these cracker choices, comparing 2 versus 4 items would be an easy 1:2 ratio discrimination. Yet, because the objects are presented as discrete, small-scale physical units, the visual system automatically recruits the Object Tracking System. When the array exceeds three elements, the OTS index buffer completely overflows and fails, rendering the infant unable to construct a stable representation. This signature discontinuity provides empirical proof that the human brain operates two functionally separate, competing non-symbolic systems for processing quantity.

6.3 Primitive Operations: Addition, Subtraction, and Cross-Modal Number

Pre-verbal infants do not merely register static numerical quantities; they execute dynamic computational operations, including basic arithmetic transformations. In a groundbreaking series of experiments that reshaped developmental psychology, Karen Wynn demonstrated that five-month-old infants possess an intuitive grasp of basic addition and subtraction over small sets of occluded objects.

Wynn’s experimental design utilized an occlusion stage where infants watched a small Mickey Mouse figure placed onto an empty surface. A screen rose to conceal the figure, after which the infant observed the experimenter’s hand place a second, identical figure behind the screen (representing the mathematical operation 1 + 1). The screen dropped to reveal either two figures (the possible, correct outcome) or a single figure (the impossible outcome, achieved via a hidden trapdoor). Wynn found that infants looked significantly longer at the impossible 1 + 1 = 1 outcome. Conversely, in a subtraction condition (2 – 1), infants were habituated to two figures, watched one figure visibly removed from behind the screen, and exhibited elevated looking times when the screen dropped to reveal two figures remaining instead of one (2 – 1 = 2). These findings demonstrated that infants maintain exact internal tallies of occluded objects and calculate the arithmetic transformations resulting from physical additions and subtractions.

Furthermore, core numerical representations are deeply abstract and cross-modal. Pre-verbal infants do not merely count visually presented shapes; they extract numerical invariants across entirely different sensory systems. In classic experiments, infants were habituated to a continuous sequence of auditory drumbeats consisting of either two or three distinct beats. When subsequently presented with visual arrays displaying either two or three visual items, infants looked longer at the visual array whose numerical cardinality matched the auditory tone sequence they had just heard. The infant mind effortlessly extracted the abstract numerical value “twoness” or “threeness,” translating freely between acoustic vibrations and retinal spatial arrays. This cross-modal fluency underscores that core numerical knowledge is conceptual, operational, and invariant across sensory modalities.

7. Core System 4: Environmental Geometry and Spatial Navigation

7.1 The Geometric Module in Spatial Reorientation

The fourth core cognitive system identified by Elizabeth Spelke and Linda Hermer is the environmental geometry module, an autonomous spatial computational system dedicated to reorienting the organism when lost within an enclosed physical space. When an animal or human becomes disoriented, it must determine its position and heading relative to its surroundings. Through systematic experimental inquiry, Spelke demonstrated that this fundamental navigational capacity relies upon an encapsulated computational engine that prioritizes the macroscopic, macro-geometric properties of the terrain—specifically distance metrics and directional angles—while actively ignoring local surface features.

In the classic Hermer and Spelke reorientation paradigm, young children (between eighteen and twenty-four months of age) were placed within a rectangular chamber characterized by distinct, asymmetrical geometric properties: two long walls and two short walls. A brightly colored, attractive toy was visibly hidden in one of the four corners in full view of the child. The child was blindfolded and gently spun around in place until completely disoriented, nullifying their vestibular, proprioceptive, and dead-reckoning egocentric tracking. The blindfold was removed, and the child was prompted to locate the hidden toy.

Because a rectangle exhibits two-fold rotational symmetry, navigating purely by geometry yields two computationally identical solutions: the true correct corner (e.g., with a long wall on the left and a short wall on the right) and the diagonally opposite, geometrically congruent rotational equivalent corner. The disoriented children systematically concentrated their searches exclusively across these two geometrically equivalent corners, splitting their reaches equally (roughly 50% each) while never searching the geometrically incorrect corners. Most astonishingly, even when the researchers introduced a salient non-geometric landmark—painting one of the short walls a vibrant, contrasting blue color (which uniquely disambiguated the exact corner where the toy was hidden)—toddlers continued to search the two geometrically symmetrical corners equally. They completely ignored the salient blue wall, demonstrating that the geometric navigation module is initially encapsulated, operating strictly over macroscopic spatial layout geometry while remaining blind to surface features.

7.2 Comparative Evidence Across Phylogeny

The geometric reorientation module represents one of the clearest examples of phylogenetic homology in cognitive science. When the exact same rectangular disorientation paradigm was administered to non-human animal species—including laboratory rats, mice, domestic chicks, rhesus macaques, and even teleost fish—the results were identical to those observed in human toddlers. Disoriented animals consistently navigate by the macroscopic geometric ratios of the enclosure, committing the identical rotational error by splitting their searches between the two geometrically congruent corners.

This widespread conservation across vertebrate phylogeny highlights the immense evolutionary survival value of a dedicated geometric module. In natural ecosystems, surface features such as foliage color, soil texture, lighting conditions, and snow cover undergo rapid, ephemeral transformations across seasonal cycles, weather events, or day-night transitions. A navigational system that relied heavily on transient surface features would be fragile and prone to failure. Conversely, the macroscopic, structural geometry of the landscape—the relative distances, elevations, and structural boundaries of ridges, valleys, and cavernous enclosures—remains invariant across decades. By anchoring reorientation to geometry, natural selection endowed vertebrates with an unyielding, dependable navigational anchor.

Cognitive neuroscience has localized the neural mechanisms supporting this geometric module to evolutionarily conserved structures within the hippocampal formation and the parahippocampal cortex. Specifically, the parahippocampal place area (PPA) in primates responds selectively to the geometric layout of physical boundaries, spatial layouts, and three-dimensional room structures, exhibiting insensitivity to transient surface decorations. At the single-unit level, this module is powered by place cells in the hippocampus and grid cells and border cells within the entorhinal cortex, which collectively construct an internal, metric coordinate map of physical space that functions identically across mammalian and avian species.

7.3 Integration of Geometric Boundaries with Feature Markers

The critical developmental and evolutionary puzzle posed by the geometric module is understanding how and when the human mind overcomes its initial modular encapsulation. While human toddlers and non-human animals remain trapped within pure geometry, ignoring salient colored walls and feature markers when disoriented in small enclosures, mature human adults easily combine geometric coordinates with landmark features (e.g., “the toy is in the corner with the long wall to the left of the blue wall”), finding the exact hidden location without committing rotational errors.

Through extensive developmental studies, Spelke and Hermer mapped the precise timeline of this cognitive transformation. The breakthrough occurs systematically between four and six years of age. What explains this transition? Spelke discovered that the ability to integrate geometric boundaries with feature markers correlates directly not with general chronological age, IQ, or raw spatial experience, but with the child’s linguistic acquisition of spatial language—specifically, the production of flexible, relational prepositional phrases that conjoin spatial terms with non-geometric feature adjectives (e.g., “to the left of the red door”).

To demonstrate that language is the causal engine driving this conceptual integration, Spelke and her team subjected mature human adults to verbal shadow tasks. When adult participants were placed in the rectangular chamber with a colored wall and required to continuously repeat a stream of recorded speech playing through headphones (a task that ties up the phonetic loop and grammatical combinatorial circuits of natural language), their spatial navigational performance instantly reverted to that of an eighteen-month-old toddler or a rat. They committed the classic rotational error, navigating strictly by the geometric layout while failing to integrate the salient colored wall. When the same adults engaged in an equally demanding non-verbal shadowing task (such as continuously tapping a complex rhythm), their ability to integrate geometry with landmarks remained entirely intact. These findings provide compelling empirical evidence that natural language acts as an active, online combinatorial bridge that breaks the informational encapsulation of core modules.

8. Core System 5: Social Partners, Group Identity, and Social Evaluation

8.1 Conspecific Identification and Language-Based Grouping

The fifth core knowledge system identified by Spelke and Katherine Kinzler governs the representation of social partners, group identity, and intergroup affiliation. While the core agent system processes the general, teleological mechanics of any goal-directed entity, the social partner system is tailored specifically to the unique complexities of conspecific human sociality. Infants must rapidly identify members of their social in-group to optimize survival, cultural learning, and collaborative affiliation.

A striking discovery emerging from Spelke and Kinzler’s research is that human infants prioritize linguistic cadence and spoken accent far above phenotypic racial characteristics when categorizing social partners. In a series of preference studies, young infants were exposed to video displays of two individuals: one who spoke the infant’s native language with a local, native accent, and another who spoke either a foreign language or their native language with a foreign accent. During silent test phases where both individuals smiled passively, infants exhibited pronounced preferential looking and reaching toward the native speaker.

When this paradigm was pitted against racial phenotypic variations, the primacy of language was unmistakable. White and Black infants who were presented with a choice between an individual of their own racial phenotype who spoke with a foreign accent versus an individual of a different racial phenotype who spoke with their native accent overwhelmingly preferred the native speaker. They reached for toys offered by the native-accented speaker and preferentially accepted food items that the native-accented speaker endorsed. From an evolutionary perspective, this priority makes complete sense. Throughout ancestral hominin history, human populations rarely encountered individuals of radically different geographical phenotypes, as migrations were limited by foot travel. However, neighboring human tribes that competed for resources, territory, and survival spoke distinct dialects and languages. Spoken accent served as a hyper-reliable, unforgeable ecological index of group affiliation and coalitionary loyalty.

8.2 Social Evaluation and Prosociality Primitives

The social partner system operates in intimate coordination with early-emerging moral and normative evaluation systems. Human infants do not view the social landscape as an undifferentiated collection of individuals; they continuously evaluate conspecifics on the basis of their fairness, prosocial inclinations, and distributive justice. The infant mind exhibits an endogenous aversion to antisocial behavior and an early expectation of resource equity.

In distributive justice experiments, infants are exposed to scenes where an adult distributor allocates resources (such as cookies, crackers, or toys) between two identical recipients. In the equal condition, the distributor awards two items to each recipient; in the unequal condition, the distributor gives three items to one recipient and one item to the other. Pre-verbal infants look significantly longer at the unequal distribution outcome, indexing cognitive surprise and normative violation at the unequal distribution. This early inequality aversion operates prior to any direct, personal experience with socio-economic transactions or explicit parental lectures regarding fairness.

Furthermore, this social evaluative system quickly interfaces with in-group favoritism and out-group wariness. While infants expect fairness as an unconstrained default, their expectations become modulated when group markers are introduced. Studies demonstrate that infants tolerate unequal distributions when resources are distributed in favor of an in-group member versus an out-group member. These findings reveal that the seeds of in-group solidarity, coalitionary psychology, and tribal categorization are deeply anchored within core social primitives, providing both the adhesive that cements human community cooperation and the evolutionary biases that modern societies must actively navigate.

8.3 Shared Intentionality and Cultural Transmission

The ultimate functional utility of the core social partner system is to facilitate cultural transmission via mechanisms of shared intentionality. While non-human primates possess complex social dominance hierarchies and acute awareness of visual sightlines, they exhibit virtually no propensity to engage in cooperative, pedagogical information sharing. As developmental psychologist Michael Tomasello and Spelke emphasize, human infants are unique in their profound drive to initiate and sustain joint attention.

By nine to twelve months of age, infants participate in the “nine-month revolution,” pointing to objects not merely to request physical items (imperative pointing), but simply to direct an adult’s attention to a shared point of interest (declarative pointing). This capacity interfaces directly with what György Gergely and Gergely Csibra designate as Natural Pedagogy. Human infants are innately receptive to ostensive-referential communicative signals: direct eye contact, high-pitched infant-directed speech (“parentese”), and communicative nodding.

When an adult emits an ostensive signal before demonstrating an action on an arbitrary physical artifact, the infant instantly assumes that the demonstrated action is culturally meaningful, generalizable, and universally normative. The infant does not merely view the interaction as a transient event, but encodes it as an enduring piece of cultural technology. This automatic, pedagogical mindset allows human cultural transmission to proceed with lightning speed and high fidelity, acting as the socio-cognitive engine that powers the cumulative ratcheting of human culture across successive generations.

9. Language as the Combinatorial Medium of Conceptual Integration

9.1 Overcoming Modular Encapsulation

The existence of domain-specific, informationally encapsulated core knowledge systems presents a profound epistemological conundrum: If the infant mind consists of segregated modules that process physical mechanics, intentional agents, approximate numbers, and environmental geometry in computational isolation, how do adult humans achieve fluid, flexible, and open-ended conceptual thought? How do we conceive of physical objects that act as agents (such as autonomous robotics), or map exact numbers to spatial trajectories (such as Cartesian coordinate geometry)?

Elizabeth Spelke’s central theoretical breakthrough is her proposal that natural language is the uniquely human combinatorial medium that breaks the encapsulation of core modules. Unlike non-human animal communication systems, which are closed, referentially narrow, and structurally restricted, human natural language is a generative, compositional system driven by hierarchical recursive syntax. Language provides a universal compositional template that allows concepts computed within one modular domain to be conjoined, modified, and integrated with concepts computed in an entirely separate module.

Spelke demonstrated this combinatorial phenomenon empirically in her spatial reorientation studies. The core geometric system computes purely metric relations (such as “short wall on the right”), while the core object and visual systems process surface feature attributes (such as “blue wall”). Prior to the acquisition of spatial prepositions, these two computational systems operate in absolute parallel, unable to share data to guide action. Once the child’s brain develops the grammatical syntax necessary to formulate sentences featuring conjoined spatial and feature modifiers (e.g., “The toy is to the left of the long blue wall”), the linguistic string acts as a shared computational bus. Language functions as a mental workspace, allowing previously incompatible representations to be welded into entirely novel, integrated conceptual structures.

9.2 Symbolic Mathematics as a Fusion of Number Systems

Perhaps the most powerful and transformative illustration of Spelke’s linguistic integration thesis lies in the evolutionary and developmental origin of symbolic mathematics. As established, humans possess two evolutionarily ancient, pre-verbal numerical systems: the Approximate Number System (which estimates large numerical sets with noisy ratio precision) and the Object Tracking System (which tracks up to three individual objects with absolute precision). Crucially, neither of these core systems contains the concept of an exact, infinite integer. The ANS lacks exactness, while the OTS lacks an open-ended numerical scale.

How does the child’s mind arrive at the revolutionary concept of natural numbers (1, 2, 3, 4, … 100, … infinity)? Spelke and Susan Carey demonstrated that the conceptual bridge is constructed via the acquisition of verbal counting lists and linguistic syntax. When children learn the rhythmic, memorized sequence of verbal count words (“one, two, three”), they initially possess no genuine mathematical understanding of their meaning. They treat the sequence as an arbitrary verbal chant.

Through an intense, years-long developmental process occurring between two and four years of age—often progressing through discrete stages of being a “one-knower,” “two-knower,” and “three-knower”—the child maps the first three count words directly onto the precise visual indexes provided by the Object Tracking System. Then, a profound conceptual synthesis occurs: the child recognizes that moving one step forward in the verbal counting sequence corresponds precisely to adding an exact physical individual to an existing set. At this moment, the recursive syntax of language—which permits indefinite concatenation (e.g., N + 1)—fuses the absolute precision of the OTS with the unbounded scope of the ANS.

This linguistic synthesis creates the foundational concept of the exact integer, an intellectual breakthrough that is completely absent in non-linguistic animals. The validity of this linguistic engine is corroborated by cross-cultural research. In indigenous Amazonian cultures such as the Pirahã and Munduruku, whose native languages lack a continuous sequence of counting words or exact numerical terms, adult members exhibit normal Approximate Number System abilities, yet they are unable to perform exact, large-scale arithmetic calculations or reliably identify precise cardinalities beyond three or four items. Without the combinatorial scaffolding of linguistic number terms, human numerical thought remains permanently anchored to its biological core foundations.

9.3 Representational Flexibility and Scientific Thinking

The transformative capacity of natural language extends far beyond elementary mathematics into the realm of abstract scientific thinking and philosophical reflection. Science routinely requires the human mind to grasp concepts that directly contradict our baseline core physical intuitions. In core physics, continuous contact and absolute solidity are inviolable axioms. Yet, modern physics demands that we conceptualize empty atoms, action-at-a-distance gravitational fields, curved space-time, and probabilistic quantum superpositions.

How can the human brain—anchored to the cognitive hardware of a Pleistocene hunter-gatherer—possibly compute quantum mechanics or evolutionary biology? Spelke argues that natural language permits the creation of complex conceptual metaphors, analogies, and formal external symbol systems that bootstrap core intuitions into novel theoretical spaces. We understand electrical currents by borrowing and metaphorically mapping the core intuitions of water flowing through physical pipes; we comprehend the subatomic nucleus by projecting the core mechanical model of a miniature planetary system.

However, this bootstrapping process reveals the enduring resilience and stubbornness of our core knowledge systems. Because our core physical and biological modules remain active across the lifespan, counter-intuitive scientific theories are exceptionally difficult to learn and psychologically sustain. When a physics student studies Newtonian inertia, their core physical system continues to project an intuitive impetus model (believing that a moving object must continually consume an internal force to stay in motion). When an adult ponders evolutionary biology, their core agent system instinctively projects teleological, intentional design onto random natural selection mutations. Advanced scientific education is not simply the passive acquisition of information; it is a continuous, effortful cognitive struggle to utilize the compositional power of language to construct reflective formal theories that consciously override our persistent core intuitions.

10. Phylogenetic Continuities: Evolutionary Origins and Comparative Cognition

10.1 Object Representation Across Animal Clades

The evolutionary roots of core knowledge are vividly illustrated by examining object representation across diverse non-human animal clades. If Spelke’s model is correct, non-human animals living in complex, three-dimensional physical environments should exhibit identical mechanical constraints regarding cohesion, continuity, and solidity. Decades of comparative cognitive research have systematically validated this hypothesis across primates, canines, corvids, and domestic avians.

In extensive studies conducted with chimpanzees, bonobos, and rhesus macaques, non-human primates consistently demonstrate looking-time patterns and predictive reaching behaviors that mirror human infants. When presented with occlusion displays wherein an object appears to magically teleport across a barrier, or two solid objects appear to occupy the same spatial coordinates, primates display elevated looking times, increased pupillary dilation, and avoidance of physically unstable trajectories. In the wild, primate foraging and locomotion demand an immediate, intuitive comprehension of physical mass, tensile branch strength, and spatial trajectories, making the core physical module indispensable to mammalian survival.

Even more dramatic evidence emerges from avian species, particularly corvids and parrots, whose evolutionary lineage diverged from mammals over three hundred million years ago. Corvids—such as New Caledonian crows and ravens—routinely solve complex multi-step physical puzzles involving water displacement (the classic Aesop’s fable paradigm), choosing solid sinking stones over buoyant hollow objects to raise the water level and retrieve floating food. Furthermore, stunning research with newly hatched domestic chicks (Gallus gallus domesticus)—reared in darkness with zero prior visual or motoric experience—demonstrates that upon their very first visual exposure to illuminated screens, chicks instinctively track the continuous, occluded spatiotemporal trajectories of objects, expecting them to maintain cohesion and solidity. The presence of these fully operational physical mechanics in neonate birds utterly refutes the constructivist assumption that core physics requires lengthy sensorimotor manual learning.

10.2 Comparative Numerical and Geometric Capacities

The comparative architecture of the Approximate Number System and environmental geometry reveals a similarly staggering evolutionary conservation. The capacity to discriminate between varying numerical magnitudes according to Weber’s Law has been documented across an immense array of species, including lions calculating whether to attack an intruding pride based on acoustic roar counts, honeybees counting landmarks to calibrate distance flight paths, and mosquitofish selecting the larger shoal for protective schooling.

Across these vastly disparate species, the quantitative discriminability ratio precisely matches the Weber-fraction curves established in human infants. A domestic chick, a guppy, and a four-month-old human infant all exhibit an identical baseline capacity to reliably discriminate between arrays exhibiting a 1:2 ratio, while simultaneously failing at a 2:3 ratio. This cross-species mathematical convergence proves that the Approximate Number System is an ancient, highly optimized evolutionary algorithm that was locked into vertebrate neural circuitry deep in the evolutionary past, conserved across hundreds of millions of years of natural selection.

Similarly, the geometric module demonstrates near-universal phylogenetic consistency. Whether testing a desert ant calculating navigational vectors, a subterranean rodent reorienting within an underground tunnel system, or a mountain goat scaling treacherous vertical cliffs, the macroscopic, metric geometry of the physical landscape forms the universal computational frame of reference. Animals continuously extract the metric distances, angles, and aspect ratios of their structural environment. When disoriented in identical rectangular experimental chambers, fish, chicks, rodents, and primates commit the identical rotational errors observed in human toddlers, searching the geometrically symmetrical corners while initially ignoring salient non-geometric feature patches. These navigational engines are evolutionary homologous modules wired directly into primitive vertebrate subcortical and hippocampal systems.

10.3 What Remains Uniquely Human?

Given the staggering evolutionary conservation of core physical, agentive, numerical, and geometric modules across non-human species, cognitive science is confronted with an imperative comparative question: What constitutes the uniquely human difference? If animal minds share the exact same core foundations with human infants, why did non-human animals never invent written mathematics, legal codes, architecture, or modern science? What is the unique evolutionary leap that defines the human mind?

Elizabeth Spelke’s comparative analysis identifies two profound evolutionary innovations that distinguish Homo sapiens from all other creatures:

• The Combinatorial Power of Natural Language: While non-human animals possess rich, highly functioning core modules, those modules remain forever trapped in evolutionary isolation. An animal possesses a physical module to track stones, an agent module to predict a predator’s pounce, a geometric module to navigate terrain, and an approximate number system to assess shoal size—yet it cannot construct an informational bridge connecting these distinct computational domains. In animal minds, the modules are permanently encapsulated islands. Human beings, and human beings alone, evolved the neural circuitry for generative recursive syntax, providing an internal communication bus that permits concepts from different core modules to be combined into novel, flexible conceptual structures.
• Shared Intentionality and the Drive for Natural Pedagogy: While a chimpanzee may understand that a conspecific has a visual line of sight to a piece of fruit, it possesses virtually no evolutionary inclination to share its own mental states, point out unobserved affordances to a peer, or engage in cooperative teaching. Humans are uniquely equipped with an innate drive for joint attention, communicative cooperation, and pedagogical instruction. This pedagogical drive transforms individual cognitive discoveries into cumulative cultural artifacts.

This dynamic fuels what Michael Tomasello describes as the cultural ratchet effect. In animal populations, an individual organism may discover an ingenious mechanical trick during its lifespan, but that innovation typically dies with the individual, or spreads sluggishly through imperfect, noisy imitation. In human societies, the fusion of core knowledge via combinatorial language, combined with our innate pedagogical drive, ensures that conceptual innovations are instantaneously codified, shared, linguistically preserved, and transmitted down through the generations. The cumulative ratchet never slips backward; our species builds towering cultural, technological, and scientific superstructures precisely because our ancient animal core systems have been woven together by natural language.

11. Critical Counterpoints, Methodological Skepticism, and Theoretical Debates

11.1 The Perceptual-Cognitive Divide: The Haith Critique

Despite its profound impact, Core Knowledge Theory has generated intense theoretical and methodological controversy within developmental psychology. The most enduring methodological critique centers upon the validity of the Violation-of-Expectation paradigm itself, articulated most forcefully by developmental psychologist Marshall Haith. Haith and fellow skeptics raised the fundamental question of the perceptual-cognitive divide: Does prolonged looking time in pre-verbal infants truly reflect rich, abstract conceptual knowledge of unobserved physical principles, or does it merely register low-level, superficial perceptual processing heuristics?

Haith argued that developmental nativists such as Spelke and Baillargeon were guilty of “rich interpretations of lean data.” According to the skeptical critique, an infant may look longer at an impossible occlusion display not because they are computing abstract notions of “cohesion,” “continuity,” or “solidity,” but simply because the impossible condition happens to generate higher visual contrast, continuous retinal motion flicker, differing optical angles, or unfamiliar sensory configurations. Skeptics contended that looking time is an ambiguous metric that conflates sensory novelty, perceptual salience, and general attentional capture with high-level conceptual deduction.

To decisively dismantle this critique, Spelke and her colleagues executed rigorous, systematically counterbalanced experimental variations. They created test displays where the perceptual elements were identical, but the conceptual meaning was reversed through subtle changes in occluder positioning or initial framing. In these exquisitely controlled conditions, infants consistently looked longer at the conceptually impossible event, even when the physically possible event featured greater sensory novelty, higher visual complexity, and more dynamic retinal changes. Furthermore, the integration of high-density EEG, fNIRS, and predictive eye-tracking pupillometry confirmed that infant surprise is accompanied by specific neural signatures of conceptual violation (such as Nc ERP modulations), proving that VOE looking times index genuine cognitive processing rather than mere low-level perceptual heuristics.

11.2 Connectionist and Neo-Empiricist Challenges

A second major theoretical assault upon Core Knowledge Theory emerged from connectionist and neo-empiricist computational modeling. Researchers such as Mark Johnson, Denis Mareschal, and Jeffrey Elman demonstrated that deep artificial neural networks, trained purely on domain-general statistical learning algorithms without any pre-programmed physical or mathematical rules, can simulate many of the looking-time patterns observed in infant experiments.

These connectionist models suggest that what Spelke categorizes as “innate core systems” could actually be the emergent byproducts of rapid, high-density statistical learning occurring over visual inputs. If an infant observes millions of continuous spatial frames during their initial weeks of life, a generalized associative neural network can rapidly extract the high-probability statistical invariants that physical surfaces do not randomly disappear, that motion is continuous, and that objects typically do not pass through one another. Under this neo-empiricist view, domain-specific modules are not innate, pre-wired starting points; they are the emergent, developmental outcomes of domain-general statistical learning mechanisms operating over rich environmental sensory inputs.

In response, Elizabeth Spelke emphasizes the crucial problem of sample complexity and the severe poverty of the stimulus. While artificial neural networks require millions of training cycles, massive datasets, and brute-force backpropagation to learn elementary spatial regularities, human infants and newly hatched chicks demonstrate these conceptual constraints immediately upon their very first visual exposures, or with remarkably minimal training data. Furthermore, statistical associative models are notoriously fragile and prone to catastrophic failure when presented with completely novel out-of-distribution permutations that violate surface statistical patterns while preserving physical laws. The infant brain does not operate as an unconstrained statistical sponge; it operates as an active, hypothesis-generating engine constrained by pre-formatted, generative core inductive biases.

11.3 Neuroconstructivism and Annette Karmiloff-Smith’s Synthesis

Perhaps the most sophisticated and nuanced developmental critique of Spelke’s nativism was formulated by the late Annette Karmiloff-Smith through her framework of neuroconstructivism. In her seminal work Beyond Modularity, Karmiloff-Smith articulated a middle path between extreme Fodorian/Spelkean nativism and radical Piagetian constructivism. She challenged the foundational nativist assumption that the infant brain begins life with pre-packaged, fully specialized cortical modules.

Karmiloff-Smith argued that the human neonate brain possesses incredible plasticity. Rather than containing innate, informationally encapsulated modules, the neonate brain is equipped with subtle, domain-relevant attentional biases and processing predispositions. Through a process she designated as progressive modularization, the brain gradually sculpts its own modular architecture over the course of developmental time through dynamic, bidirectional interactions with the environment. In Karmiloff-Smith’s view, modularity is the developmental end-state of neurodevelopment, not its evolutionary starting point.

To substantiate her critique, Karmiloff-Smith pointed to developmental neurogenetic disorders such as Williams Syndrome and Autism Spectrum Disorder. In Williams Syndrome, individuals often exhibit hyper-fluent linguistic performance alongside severely impaired visuospatial and numerical capabilities. Nativists initially heralded this as clean genetic proof of a double dissociation between an intact language module and an impaired spatial/numerical module. However, Karmiloff-Smith’s micro-developmental analyses revealed that the linguistic syntax of individuals with Williams Syndrome is deeply abnormal and acquired through compensatory pathways, demonstrating that a developmental genetic mutation reverberates globally across the entire neural connectome rather than selectively disabling a single, neatly isolated core module.

This neuroconstructivist critique forced contemporary Core Knowledge Theory to refine and nuance its theoretical claims. Spelke has clarified that core systems should not be conceptualized as rigid, immutable, fully insulated Fodorian brain modules stamped into specific patches of cortex at birth. Rather, core systems are biologically endowed computational constraints and privileged informational pathways that guide early cortical development, ensuring that under standard developmental conditions, the brain predictably and reliably converges upon specialized, modular neural substrates dedicated to objects, agents, numbers, geometry, and social partners.

12. Contemporary Applications: Education, Neuroimaging, and Artificial Intelligence

12.1 Core Knowledge-Informed Pedagogical Interventions

The profound theoretical insights generated by Core Knowledge Theory have transitioned out of developmental laboratories into transformative real-world pedagogical interventions. By mapping the precise computational architecture of intuitive infant cognition, educational psychologists can now design preschool and primary curricula that deliberately harness, calibrate, and build upon children’s innate core systems, rather than fighting against them.

A premier example of this is the design of early mathematics curricula aimed at closing the socio-economic achievement gap in young children. Traditional preschool curricula often attempt to teach mathematics exclusively through rote memorization of arbitrary symbols and drill-based flashcards—an approach that frequently induces mathematical anxiety and fails to cultivate deep conceptual understanding. Interventions developed directly by Elizabeth Spelke and her educational collaborators (such as the Games for Learning initiative) take the opposite approach: they engage children in fast-paced, non-symbolic gameplay designed to exercise and calibrate the Approximate Number System and the Object Tracking System.

In these randomized controlled trials, young preschool children play games requiring them to rapidly estimate whether a blue cloud has more dots than a red cloud, or compare which animal received more crackers, without counting. Longitudinal results demonstrate that exercising the intuitive, non-symbolic Approximate Number System directly enhances children’s subsequent ability to grasp symbolic arithmetic, learn the meaning of Arabic digits, and solve formal algebraic problems years later. By grounding abstract symbolic mathematics in the intuitive, biological bedrock of the ANS and OTS, educators provide children with a robust, intuitive conceptual foundation that accelerates mathematical proficiency.

Similarly, in the realm of science education, Core Knowledge Theory illuminates the psychological etiology of persistent scientific misconceptions. When elementary and secondary students struggle to master Newtonian mechanics, gravity, or evolutionary biology, they are not exhibiting general intellectual deficits; they are colliding with their unyielding, persistent core physical and agentive systems. Pedagogical frameworks informed by core knowledge train science teachers to anticipate these persistent, intuitive core biases (such as the intuitive impetus theory or teleological biological design). Rather than attempting to erase these intuitions—which is neurologically impossible—educators teach students metacognitive strategies to consciously recognize when their core physical heuristics are misfiring, and use explicit linguistic and mathematical representations to construct formal, scientific models that successfully override their baseline intuitions.

12.2 Modern Neuroimaging Validation

In the contemporary era of high-resolution cognitive neuroscience, Core Knowledge Theory is receiving profound structural and functional validation through advanced infant neuroimaging technologies. Utilizing state-of-the-art awake infant fMRI and high-density, 128-channel infant electrophysiology, neuroscientists are directly mapping the localized neurobiological substrates that instantiate Spelke’s core computational systems.

Groundbreaking fMRI investigations conducted by researchers such as Rebecca Saxe and Ben Deen at the Massachusetts Institute of Technology have succeeded in scanning awake, active three- to eight-month-old infants as they view dynamic visual movies. The neuroimaging data reveals astonishing functional specialization: infants exhibit localized cortical activation within the superior temporal sulcus (STS) and medial prefrontal cortex specifically when viewing goal-directed animate agents, matching the neural topography observed in adult social cognition. Conversely, when viewing inanimate, mechanical objects colliding in space, the infant brain selectively activates the bilateral intraparietal sulcus (IPS) and regions within the ventro-temporal cortex.

Furthermore, neuroimaging studies examining the neural correlates of number have localized the Approximate Number System in pre-verbal infants to the deep banks of the intraparietal sulcus—the exact same anatomical location that computes numerical magnitude in adult mathematicians. High-density resting-state and diffusion tensor imaging (DTI) demonstrate that the structural white matter tracts connecting these specialized cortical regions—such as the arcuate fasciculus and the inferior fronto-occipital fasciculus—are already structurally formed and organized in the human infant connectome at birth. These imaging discoveries provide undeniable biological confirmation that the human brain is genetically and architecturally pre-wired with domain-specific computational networks designed to parse reality into the exact ontological categories delineated by Elizabeth Spelke.

12.3 Implications for Cognitive Artificial Intelligence

As the fields of cognitive science, computer science, and machine learning converge, Core Knowledge Theory has emerged as one of the most critical theoretical battlegrounds in the quest for Artificial General Intelligence (AGI). Modern deep learning architectures—such as large language models (LLMs) and massive convolutional neural networks—have achieved breathtaking benchmarks in pattern recognition, text synthesis, and game playing. Yet, despite their billions of parameters and petabytes of training data, contemporary artificial intelligence models remain fundamentally brittle, hallucination-prone, computationally inefficient, and completely devoid of basic commonsense reasoning.

Prominent artificial intelligence researchers and cognitive scientists, including Josh Tenenbaum, Yann LeCun, and Elizabeth Spelke herself, have forcefully argued that current AI architectures are failing precisely because they are pursuing a flawed, radical Lockean empiricist agenda. Contemporary AI attempts to learn everything from scratch through massive, unconstrained statistical associations across trillions of data tokens. A typical autonomous driving system requires millions of visual driving hours to recognize spatial obstacles, yet it can still be catastrophically confused by anomalous shadows or an unfamiliar road reflection—errors that a human teenager would navigate effortlessly.

To overcome this computational bottleneck, leading AI research laboratories are actively engineering core knowledge inductive biases directly into the foundational architectures of next-generation neural networks. Rather than expecting a deep network to discover the laws of physical mechanics or agent intentionality purely through trial-and-error pixel processing, computer scientists are hardwiring computational sub-modules that encode the Spelkean principles of cohesion, continuity, solidity, and rational efficiency.

Crucial benchmarks designed to evaluate machine intelligence are now constructed directly around Elizabeth Spelke’s empirical paradigms. For instance, the IntPhys (Intuitive Physics) benchmark challenges artificial intelligence systems to predict the physical trajectories of objects in synthetic 3D video environments, systematically exposing the networks to the exact same possible versus impossible physical events historically administered to human infants: disappearing objects, overlapping solid masses, and discontinuous teleportations. Contemporary deep neural networks consistently fail these benchmarks unless they are specifically augmented with object-centric representational architectures that track discrete physical entities through space and time. True artificial general intelligence will never be achieved through brute-force statistical curve-fitting alone; it will require artificial minds to be anchored upon the foundational, evolutionary building blocks of core knowledge.

Conclusion

The intellectual odyssey initiated by Elizabeth Spelke has fundamentally reconstructed our scientific understanding of the human mind, its evolutionary origins, and its developmental trajectory. By shattering the long-standing dichotomy between radical empiricism and unconstrained nativism, Spelke demonstrated that human infants enter the world neither as cognitively vacant slates waiting for environmental impressions, nor as fully fledged miniature adults carrying pre-formed cultural concepts. Instead, natural selection has endowed our species with a small, universal, and highly optimized toolkit of core knowledge systems: dedicated computational engines that parse the chaos of reality into bounded objects, goal-directed agents, approximate magnitudes, environmental geometry, and cooperative social partners.

These core systems are phylogenetically ancient, conserved across millions of years of vertebrate evolution, and biologically anchored within dedicated neural circuits. They emerge at the very dawn of postnatal life, operating as robust, permanent anchors that persist completely intact beneath adult conscious reflection. Most profoundly, Spelke resolved the great evolutionary enigma of human intellectual uniqueness: our species ascended to the heights of abstract mathematics, theoretical physics, literature, and cumulative cultural civil societies not through the wholesale invention of an entirely new brain, but through the evolutionary arrival of natural language as an open-ended combinatorial medium. Language serves as the computational tapestry that weaves our isolated, animal core modules into a generative, unified, and infinite cognitive whole.

As we gaze toward the future of cognitive science, the applications of Core Knowledge Theory will continue to reverberate across disciplines. In education, it provides the blueprint for pedagogies that calibrate children’s intuitive biological minds with abstract formal knowledge. In developmental cognitive neuroscience, it guides the mapping of the infant connectome with astonishing anatomical and temporal precision. And in artificial intelligence, it serves as the ultimate diagnostic and structural model for building resilient, commonsense-endowed machines capable of genuine understanding. Elizabeth Spelke’s legacy is nothing less than a unified, naturalized epistemology—an empirical testament to the enduring, beautiful architecture of the human mind.

References

• Baillargeon, R., Spelke, E. S., & Wasserman, S. (1985). Object permanence in five-month-old infants. Cognition, 20(3), 191-208. https://doi.org/10.1016/0010-0277(85)90008-3
• Carey, S. (2009). The Origin of Concepts. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780195367638.001.0001
• Chomsky, N. (1959). A review of B. F. Skinner’s Verbal Behavior. Language, 35(1), 26-58. https://doi.org/10.2307/411334
• Csibra, G., & Gergely, G. (2009). Natural pedagogy. Trends in Cognitive Sciences, 13(4), 148-153. https://doi.org/10.1016/j.tics.2009.01.005
• Dehaene, S. (2011). The Number Sense: How the Mind Creates Mathematics (Revised and Updated ed.). Oxford University Press.
• Fodor, J. A. (1983). The Modularity of Mind: An Essay on Faculty Psychology. MIT Press. https://doi.org/10.7551/mitpress/4737.001.0001
• Gergely, G., Nádasdy, Z., Csibra, G., & Bíró, S. (1995). Taking the intentional stance at 12 months of age. Cognition, 56(2), 165-193. https://doi.org/10.1016/0010-0277(95)00661-H
• Haith, M. M. (1998). Who put the cog in infant cognition? Is rich interpretation too costly? Infant Behavior and Development, 21(2), 167-179. https://doi.org/10.1016/S0163-6383(98)90001-7
• Hamlin, J. K., Wynn, K., & Bloom, P. (2007). Social evaluation by preverbal infants. Nature, 450(7169), 557-559. https://doi.org/10.1038/nature06288
• Hermer, L., & Spelke, E. S. (1994). A geometric module for surface orientation in young children. Cognition, 53(2), 107-133. https://doi.org/10.1016/0010-0277(94)90065-5
• Hermer, L., & Spelke, E. S. (1996). Modularity and development: The case of spatial reorientation. Cognition, 61(3), 195-232. https://doi.org/10.1016/S0010-0277(96)00714-7
• Karmiloff-Smith, A. (1992). Beyond Modularity: A Developmental Perspective on Cognitive Science. MIT Press. https://doi.org/10.7551/mitpress/1579.001.0001
• Kinzler, K. D., Dupoux, E., & Spelke, E. S. (2007). The native language of social cognition. Proceedings of the National Academy of Sciences, 104(30), 12577-12580. https://doi.org/10.1073/pnas.0705345104
• Piaget, J. (1954). The Construction of Reality in the Child. Basic Books. https://doi.org/10.1037/11168-000
• Pylyshyn, Z. W. (2001). Visual indexes, tracking, and object-based attention. Behavioral and Brain Sciences, 24(1), 127-158. https://doi.org/10.1017/S0140525X0100392X
• Riochet, R., Yildirim, I., McDermott, J., & Spelke, E. (2021). IntPhys: A framework and benchmark for visual intuitive physics reasoning. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(11), 8001-8016. https://doi.org/10.1109/TPAMI.2021.3109673
• Spelke, E. S. (1990). Principles of object perception. Cognitive Science, 14(1), 29-56. https://doi.org/10.1207/s15516709cog1401_3
• Spelke, E. S. (2000). Core knowledge. American Psychologist, 55(11), 1233-1243. https://doi.org/10.1037/0003-066X.55.11.1233
• Spelke, E. S. (2022). What Babies Know: Knowledge and Mind. Oxford University Press. https://doi.org/10.1093/oso/9780190088996.001.0001
• Spelke, E. S., Breinlinger, K., Macomber, J., & Jacobson, K. (1992). Origins of knowledge. Psychological Review, 99(4), 605-632. https://doi.org/10.1037/0033-295X.99.4.605
• Spelke, E. S., & Kinzler, K. D. (2007). Core knowledge. Developmental Science, 10(1), 89-96. https://doi.org/10.1111/j.1467-7687.2007.00569.x
• Tomasello, M. (2019). Becoming Human: A Theory of Ontogeny. Harvard University Press. https://doi.org/10.4159/9780674988651
• Vallortigara, G. (2012). Core knowledge of object, number, and geometry: A comparative and neural approach. Cognitive Neuropsychology, 29(1-2), 131-152. https://doi.org/10.1080/02643294.2012.654772
• Wynn, K. (1992). Addition and subtraction by human infants. Nature, 358(6389), 749-750. https://doi.org/10.1038/358749a0