Technical Program

Paper Detail

Paper: PS-1A.38
Session: Poster Session 1A
Location: Symphony/Overture
Session Time: Thursday, September 6, 16:30 - 18:30
Presentation Time:Thursday, September 6, 16:30 - 18:30
Presentation: Poster
Publication: 2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania
Paper Title: Human Priors in Hierarchical Program Induction
Manuscript:  Click here to view manuscript
Authors: Mark Ho, Sophia Sanborn, Fred Callaway, David Bourgin, Tom Griffiths, UC Berkeley, United States
Abstract: People impose structure onto other agents' sequential problem-solving behavior. That is, they interpret actions in terms of a \textit{likely program} that the observed agent was executing to solve a problem. But what prior expectations do people have about these programs? For example, in both cognitive science and computer science, \textit{shortest description length} has been proposed as a general principle for inducing a program. However, there may be other criteria that bias how people reconstruct others' solutions: That they are symmetric, balanced, or organize child and parent processes in particular ways. Here, we report preliminary experiments and models that investigate peoples' priors on others' problem-solving programs. We first present a novel experimental paradigm in which participants were given examples of how a problem was solved and needed to reconstruct the program that generated the solution. Then, we discuss the application of our model of human program priors to these data. We find that shortest description length inadequately explains how people reconstruct others' problem solving programs.