The Child as Hacker

Abstract

The scope of human learning and development poses a radical challenge for cognitive science. We propose that developmental theories can address this challenge by adopting perspectives from computer science. Many of our best models treat learning as analogous to computer programming because symbolic programs provide the most compelling account of sophisticated mental representations. We specifically propose that children's learning is analogous to a particular style of programming called hacking, making code better along many dimensions through an open-ended set of goals and activities. By contrast to existing theories, which depend primarily on local search and simple metrics, this view highlights the many features of good mental representations and the multiple complementary processes children use to create them.

Keywords: Computational modeling; Hacking; Language of thought; Learning and cognitive development; Program induction.

Figure 1:

Overview of the child as hacker hypothesis. a: Code can be changed using many techniques (x-axis) and assessed according to many values (y-axis). Standard learning models in machine learning and psychology (green region) tend to focus solely on tuning the parameters of statistical models to improve accuracy. Recent LOT models (red region) expand this scope, writing functions in program-like representations and evaluating them for conciseness and sometimes efficiency. Yet, the set of values and techniques used by actual hackers (and by hypothesis, children; blue region) remains much larger. b: A comparison of three families of developmental metaphors discussed in this paper—the child as scientist, the workshop and evolutionary metaphors, and the child as hacker—along three dimensions: the kinds of knowledge learners acquire, the primary objectives of learning, and the mechanisms used in learning.

Figure I:

Several classes of programs expressed as symbolic images: (a) blueprints; (b) assembly instructions; (c) musical notation; (d) knotting diagrams (e) juggling patterns; (f) graphical proofs; and (g) football plays.

Figure I:

Mapping kinship to code: (a) a family tree labeled by three kinship systems (circle = female, colors are different terms, child generation ignores gender); (b–d) Kinship in Prolog. Prolog expresses computations as Horn clauses called rules, Head :- Body.. Head is true if each term in Body is true (empty bodies are also true); (b) initial kinship data; (c) rules for inferring kinship relations, including new primitives parent, spouse, male, and female; and (d) a small set of rules such that (c) & (d) implies all of (b).

Figure II:

Mapping Mendelian inheritance to code: (a) an overview of Mendelian inheritance; (b–c) Mendelian inheritance in Church. Church expresses computations as parenthesis-delimited trees. (b) a list of individuals (a1, a2, b1, …), their parents, and phenotypes (YW = Yellow; GN = Green; SM = Smooth; WR = Wrinkly); (c) a list of traits (dominant followed by recessive) and part of a generative theory using Mendel’s laws and a uniform prior over unknown parents (i.e. random-gene draws a pair of alleles uniformly at random).