Parsing heuristic and forward search in first-graders game-play behavior

PAZ, LUCIANO; GOLDIN, ANDREA PAULA; DIUK, CARLOS; SIGMAN, MARIANO

Cognitive Science

Wiley Online Library

Año: 2014 vol. 39 p. 944 - 971

1551-6709

Seventy-three children between 6 and 7 years of age were presented with a problem havingambiguous subgoal ordering. Performance in this task showed reliable fingerprints: 1) a non-monotonic dependence of performance as a function of the distance between the beginningand the end-states of the problem, 2) very high levels of performance when the first move wascorrect and 3) states in which accuracy of the first move was significantly below chance. Thesefeatures are consistent with a non-Markov planning agent, with an inherently inertial decisionprocess, and that uses heuristics and partial problem knowledge to plan its actions. We applieda statistical framework to fit and test the quality of a proposed planning model (Monte-CarloTree Search (MCTS)). Our framework allows us to parse out independent contributions toproblem-solving based on the construction of the value function and on general mechanismsof the search process in the tree of solutions. We show that the latter are correlated with chil-dren?s performance on an independent measure of planning, while the former is highly domainspecific.