KurisutinaDevelopmental Science 10(1):89–96, doi 10.1111/j.1467-7687.2007.00569.x. Author copy from the Harvard Laboratory for Developmental Studies (Spelke's lab). Read by researcher R4e (innate against learned) for research batch R4, 5 October 2026. Provenance: papers/base/spelke2007_core_knowledge.provenance.json.
What was read
- Read in full: every line of the pdftotext conversion of the 8-page author PDF: abstract, the whole text (it has no section headings beyond the introduction), acknowledgements and the reference list.
- Not read: nothing omitted. The paper is a short review: it reports no new data, no sample sizes and no effect sizes of its own. Every empirical claim below is the authors' description of cited work.
What it argues
- Neither of the two classic views is right. The mind is not one general-purpose learner (Locke, Hume, connectionism), nor hundreds of special-purpose modules (evolutionary psychology, "massive modularity"). It is built on a small number of separable core knowledge systems, on which new, flexible skills and belief systems build.
- Four systems are established, a fifth is proposed: inanimate objects and their mechanics; agents and their goal-directed actions; number (sets, ordering, addition, subtraction); places and the geometry of the layout; and, as a candidate, social partners and group members.
- How a system is identified: each centres on principles that individuate its entities and support inferences about them, and each has signature limits that let investigators find the same system across tasks, ages, species and cultures.
The evidence described, system by system
| System | Principles and signature limits | Infants and newborns | Other species | Adults; other cultures |
|---|---|---|---|---|
| Objects | Cohesion, continuity, contact; about three objects at a time | Some abilities present without visual experience (newborns; newly hatched chicks). Months of experience do not yield separate systems for foods, artifacts or non-object stuff (sand, liquids) | Monkeys: same constraints, set size limit of four | Adults track 3–4 moving objects that obey cohesion and continuity, and fail beyond that or with non-objects. The Pirahã (no number words beyond "two", per Everett) distinguish objects from non-objects and show the set-size limit |
| Agents | Goal-directed, efficient, contingent and reciprocal action; gaze direction; agents need no face | Newborns use gaze direction; infants do not read inanimate motion as goal-directed | Chicks, rhesus monkeys and chimpanzees are sensitive to what others can see; "mirror neurons" in monkeys | Mirroring in adults; goal-directed action representations guide intuitive moral reasoning |
| Number | Imprecise, with error growing with cardinal value (a ratio limit); abstract across modalities; addition and subtraction | Large-number discrimination with a ratio limit when continuous quantity is controlled; addition and subtraction | Monkeys discriminate and add large numbers | Precision improves from a ratio of 2.0 at 6 months to 1.15–1.3 in adults. The Munduruku (no verbal counting, no exact number words beyond "three", little schooling) discriminate large numbers as accurately as educated French adults, and they and uninstructed US preschoolers add and subtract approximate numbers |
| Geometry | Distance, angle and sense of extended surfaces; blind to colour or odour of the layout; reorientation separate from landmark association | Disoriented young children reorient by layout geometry, not by the geometry of object arrays | Disoriented animals do the same | Adults use landmarks far more, but fall back on geometry under verbal or spatial interference. Munduruku children and adults with little or no schooling use geometry, including in pictures |
| Social groups (candidate) | In-group preference; native language and accent as group cues | From birth, infants prefer the sound of their native language. 6-month-olds look longer at a woman who spoke their native language; 12-month-olds in Boston reach for the snack offered by an English rather than a French speaker. 3-month-olds look longer at own-race faces, but this depends on experience (below) | — | Any minimal grouping produces in-group preference in adults and children |
The experience-dependence the authors report themselves.
- Own-race face preference at 3 months depends on the family's race and the community's majority: Israeli infants of Caucasian families prefer Caucasian faces; Ethiopian infants of African families prefer African faces; Israeli infants of African families, living in a mostly Caucasian society, show no consistent preference.
- Infants look longer at faces of their primary caregiver's gender.
- The own-race preference is moderated by exposure to other-race faces; in adults, intergroup bias is moderated by certain kinds of contact.
What core knowledge does later (the authors' claims)
- A foundation for learning: word learning depends on pre-existing concepts; core geometry guides map understanding even in remote cultures without schooling; core number supports learning to count and symbolic arithmetic; the social system may guide "cultural learning", the acquisition of the skills and behaviours of one's own group.
- A source of errors: core principles fail at very small and very large scales (objects are not cohesive there; space is not Euclidean), so children and adults err about mechanics, non-Euclidean geometry and numbers outside the core domains. The authors single out the us-against-them system as the source of the most serious errors.
- It can be overcome: conceptual change in science, in children learning to count (their conception of number changes), and in children learning biology (their conception of agents changes).
Limits
- A short position piece from 2007. No study is presented in enough detail to judge size, sample or replication; some cited support is unpublished (a conference poster, an honours thesis).
- The fifth system is explicitly a candidate, and the evidence given for it is partly experience-driven.
- "Present in newborns" is claimed for some object abilities and for gaze use; for the other systems the evidence is from infants of a few months, which leaves room for early learning. The paper does not separate innate from early-learned within its systems.
- Cross-cultural evidence comes from two Amazonian groups (Pirahã, Munduruku) and a few tasks.
What it means for Amadeus (inference)
- A candidate list for what is shared by birth: representations of objects, agents, approximate number and layout geometry, with signature limits that are the same in infants, monkeys, schooled adults and unschooled Amazonian adults. These are the strongest candidates for the base's architecture or prior, not for its training data.
- The signature limits are testable. Set-size limits of 3–4 in object tracking, the ratio limit on number (adult 1.15–1.3), and reliance on layout geometry under interference are the kind of human signature that a frozen base battery (B9) could include.
- The social core is a learning device, not content. Infants come ready to sort people into "us" and "them" by language and familiar faces, but which people count as "us" is learned from the family and community. That splits the base from the slot cleanly: the disposition to form an in-group belongs in the base; who the in-group is belongs in the slot.
- Language rides on prior concepts. If word learning depends on pre-existing concepts, a base that learns its vocabulary from text alone has no such foundation. Whether that matters for a functional replica is open.