We examine whether people evaluate the performance of other agents using only behavioral metrics (intuitive behaviorism) or by also taking into account the program that is driving an agent’s performance (intuitive cognitivism). In an online study, 200 participants, most without programming experience, learned to use a simple block programming language that controls a maze-solving robot. Participants then evaluated which of two programs was “better”, for 18 pairs of robots and programs. The programs varied in 3 metrics. Two of these, action efficiency and representation efficiency, were motivated by work in Reinforcement Learning and philosophy of mind, and the third, semantic generalization, is novel. We found that people’s judgements are in line with intuitive cognitivism, that people are sensitive to the program features, and that people intuitively evaluate and care about which other problems a program solves beyond the given task.

We apply Bayesian data analysis to a structured cognitivemodel in order to determine the priors that support humangeneralizations in a simple concept learning task. We mod-eled 250,000 ratings in a “number game” experiment wheresubjects took examples of a numbers produced by a program(e.g. 4, 16, 32) and rated how likely other numbers (e.g. 8vs. 9) would be to be generated. This paper develops a dataanalysis technique for a family of compositional “Language ofThought” (LOT) models which permits discovery of subjects’prior probability of mental operations (e.g. addition, multi-plication, etc.) in this domain. Our results reveal high cor-relations between model mean predictions and subject gener-alizations, but with some qualitative mismatch for a stronglycompositional prior.