Knowledge Vault 5 /11 - CVPR 2016
What Babies Know
Elizabeth Spelke
< Resume Image >

Concept Graph & Resume using Claude 3 Opus | Chat GPT4o | Llama 3:

graph LR classDef infants fill:#f9d4d4, font-weight:bold, font-size:14px classDef objects fill:#d4f9d4, font-weight:bold, font-size:14px classDef agents fill:#d4d4f9, font-weight:bold, font-size:14px classDef concepts fill:#f9f9d4, font-weight:bold, font-size:14px classDef implications fill:#f9d4f9, font-weight:bold, font-size:14px A[What Babies Know] --> B[Infants gain knowledge
rapidly, flexibly 1] B --> C[Abstract ideas fundamental
to thinking 2] B --> D[Infants abilities differ
from animals 3] B --> E[Diverse studies provide
cognitive insights 4] A --> F[Babies organize scenes
into objects 5] F --> G[Infants expect object
solidity, continuity 6] F --> H[Infants infer object interactions,
struggle with features 7] F --> I[Chicks suggest innate
object knowledge 8] F --> J[Infants physical knowledge
improves through learning 9] A --> K[Infants expect agents
spontaneous motion 10] K --> L[Chicks expect agents
motion, causality 11] K --> M[Infants view reaching
as goal-directed 12] M --> N[Mittens enhance goal-directed
reaching understanding 13] K --> O[Infants expect efficient
goal pursuit 14] O --> P[Infants judge costlier
goals as valuable 15] A --> Q[Infants grasp objects,
agents intuitively 16] Q --> R[Unclear specifics of
infant concepts 17] Q --> S[Object individuation improves
by 12 months 18] S --> T[Labels unify objects
into categories 19] T --> U[Infants expect similar
properties within categories 20] A --> V[Infants excel at
learning, intelligence 21] V --> W[Infants may have
innate cognitive mechanisms 22] V --> X[Early cognitive systems
persist lifelong 23] X --> Y[Infant, adult cognition
mutually informative 24] V --> Z[Animals share infant
object, agent representations 25] A --> AA[AI-psychology partnership could
elucidate human mind 26] AA --> AB[Infant intelligence informs
human-like AI 27] AA --> AC[Infant origins key
to understanding mind 28] AC --> AD[Infant knowledge shapes
modern adaptation 29] AA --> AE[Infant capacities model
flexible, learning AI 30] class A,B,C,D,E,V,W,X,Y infants class F,G,H,I,J,Q,R,S,T,U objects class K,L,M,N,O,P agents class Q,R,S,T,U concepts class AA,AB,AC,AD,AE implications


1.- Infants gain knowledge rapidly and flexibly, allowing them to become competent in any human society.

2.- Abstract ideas that can't be seen or felt are fundamental to infants' thinking, from math to morality.

3.- Infants' cognitive abilities seem qualitatively different from animals, despite similar perceptual and motor skills.

4.- Behavioral studies on infants and comparisons across age, culture, species and levels of analysis provide insight into infant cognition.

5.- Babies organize visual scenes into objects despite occlusion, using motion and spatial arrangement cues.

6.- Infants extrapolate object motion behind barriers, expecting objects to move on connected, unobstructed paths (solidity principle).

7.- Infants infer interactions between objects upon contact but struggle to use object features alone to track identity.

8.- Controlled rearing studies with chicks provide an existence proof that some object knowledge could be innate.

9.- Between 3-10 months, infants' knowledge of object support, gravity, and solidity improves, likely through learning.

10.- Infants expect agents to spontaneously cause their own motion and action, unlike inert objects.

11.- Newly hatched chicks also expect agents to spontaneously move and cause state changes in other objects.

12.- Infants represent reaching actions as goal-directed, expecting agents to pursue goals efficiently.

13.- With sticky mittens, 3-month-olds interpret reaching as goal-directed and expect efficient action, earlier than without mittens.

14.- Infants represent the cost of agent's actions and expect them to take the most efficient means to goals.

15.- 10-month-olds judge that agents value goals more if they take costlier means to achieve them.

16.- Infants have early concepts of objects (naive physics) and agents (naive psychology) to understand the world.

17.- It's unclear if infant object concepts rely on ideas of forces/masses vs. bodies/motions, or goal concepts rely on costs/rewards vs. actions/goals.

18.- At 10 months, infants struggle to use object features to individuate objects, but this improves by 12 months.

19.- Infants hearing common labels for objects use this to unify them into categories despite perceptual differences.

20.- By 12 months, infants expect same-named objects to have similar causal powers and kind-based affordances.

21.- Young infants are currently the best "learning machines" we have as models of intelligence.

22.- There may be innate mechanisms in infants for representing objects, agents, number, geometry, and language.

23.- Infants' early cognitive systems aren't discarded with age, but persist in children and adults.

24.- Adult and infant cognition can be mutually informative; what's present in infants is likely still at work in adults.

25.- Object and agent representations in infants have parallels in other animals that can be studied comparatively.

26.- A future partnership between computation and psychology could build human-like AI and elucidate the human mind.

27.- Understanding infant intelligence could help build AIs that construe the world similarly to humans.

28.- Studying the origins of intelligence in infants is important for understanding the human mind more broadly.

29.- The knowledge infants gain shapes how we adapt to the rapidly changing modern world.

30.- The cognitive capacities of human infants offer a model for creating flexible, rapidly-learning, and useful artificial intelligence systems.

Knowledge Vault built byDavid Vivancos 2024