Archie Fields III (University of Calgary)
Recently it has been argued that agent-based simulations which involve using the Prisoner’s Dilemma and other game-theoretic scenarios as a means to study the evolution of cooperation are seriously flawed because they lack empirical support and explanatory relevance to actual cooperative behavior (Arnold 2014, 2015). I respond to this challenge for simulation-based studies of the evolution of cooperation in two ways. First, I argue that it is simply false that these models lack empirical support, drawing attention to a case which highlights how empirical information has been and continues to be incorporated into agent based, game-theoretic models used to study the evolution of cooperation. In particular I examine the work of Bowles and Gintis and show how they draw upon ethnographic and biological evidence as well as experiments in behavioral psychology in their models of the evolution of strong reciprocity (2011). Ultimately, I take Arnold’s misdiagnosis of the empirical support and relevance of these models to result from too stringent standards for empirical support and a failure to appreciate the role the results of these models can play in identifying and exploring constraints on the evolutionary mechanisms (e.g. kin selection, group selection, spatial selection) involved in the evolution of cooperation. Second, I propose that a modified version of Arnold’s criticism is still a threat to model-based research in the evolution of cooperation: the game-theoretic models used to study the evolution of cooperation suffer from certain limitations because of the level of abstraction involved in these models. Namely, these models in their present state cannot be used to explore what physical or cognitive capacities are required for cooperative behavior to evolve because all simulated agents come equipped with the ability to cooperate or defect. That is, present models can tell us about how cooperation can persist or fail in the face of defection or other difficulties, but cannot tell us very much about how agents come to be cooperators in the first place. However, I also suggest a solution to this problem by arguing that there are promising ways to incorporate further empirical information into these simulations via situated cognition approaches to evolutionary simulation. Drawing on the dynamics of adaptive behavior research program outlined by Beer (1997) and more recent work by Bernard et al. (2016), I conclude by arguing that accounting for the physical characteristics of agents and their environments can shed further light on the origins of cooperation.
