In What Sense is “the Prisoner's Dilemma Game” a Dilemma for a Human or a Programmed Agent?

Abstract

In this paper, we compare the experimental results of human agents with that of programmed agents using the double-bind prisoner's dilemma game, in which an ordinary prisoner's dilemma game is nested into another dilemma; that is, a player has to decide firstly if s/he will play prisoner's dilemma or not. Comparing human-agent experimental results with programmed agent experimental results, we found remarkable similarities and differences as follows: (1) when cooperation is chosen in the second stage, both human and programmed agents show similar distributions of the first choices, (2) in the case when agents choose defection in the second stage, human agents have a tendency to play an ordinary prisoner's dilemma game, (3) in the case when agents choose defection in the second stage, programmed agents show almost equal distribution. These results suggest that programmed agents may be cooperators and get close to the Pareto optimal equilibrium in an appropriate setting.