A Turing Tournament for Behavioral Economics

California Institute Of Technology, Pasadena CA

Investigators

Abstract

One fundamental goal of social science is to develop good models of human behavior. But, it is difficult to know when we have been successful -- there are many competing models and, using standard methods, it is difficult to compare them to each other. In this research we advance a methodology involving a computer-based Turing Tournament in which data from emulators (computer programs designed to ''act like human subjects'' in repeated 2 person games) and data from human subjects are compared by detectors (computer programs designed to separate human from emulator data). The winning detector is the one best able to sort humans from emulators. The winning emulator is the one best able to fool the best detector. Unlike other tournaments, this is not one program playing another to win a game. The idea, after all, is not to be better than humans but to be exactly like humans. One of the broad benefits from this activity will be the creation of a publicly available test-bed in which researchers can test their models of behavior against true human behavior, and in which researchers can test the efficacy of their methodologies to detect non-human behavior.