The Dark Side of the Force: When computer simulations lead us astray and "model think" narrows our imagination. (revised version, October 2006)
|Table of Contents|
|2 Different aims of computer simulations in science|
|3 Criteria for explanatory simulations|
|4 Examples of Failure: Axelrod style simulations of the “evolution of cooperation”|
|4.1 Typical features of Axelrod style simulations|
|4.2 How Axelrod style simulations work|
|4.3 The explanatory irrelevance of Axelrod style simulations in social sciences|
|4.4 Do Axelrod style simulations do any better in biology?|
Let us look in more detail at a typical exponent of this tradition of simulation based research to see how Axelrod-style simulations work in practice. An in many respects good example for this tradition is provided by Rudolf Schüßler's “Kooperation unter Egoisten” (Schuessler 1997). Schüßler called into question Axelrod's assumption that continued interaction is a necessary precondition for the evolution of cooperation. Quite the contrary to Axelrod's thesis, Schüßler wanted to show that cooperation can even emerge on “anonymous markets”. In order to do so he set up is own Axelrod-style simulation where agents are free to break up the cycles of interaction whenever they want. This encourages a kind of hit and run tactic where agents do not cooperate in the last round of the interaction on their behalf and take away the benefit of single-sided non-cooperation without being punished. With the help of his computer simulation Schüßler could demonstrate that even in this case cooperative strategies could - under certain specific simulation conditions - outcompete the cheaters (Schuessler 1997, p. 78ff.). The reason for this astonishing phenomenon is quite easy to understand: When the interaction is broken up, the previous partners of interaction are forced to pick their new partner from the pool of free players. As the cooperative players tend to be bound in partnerships by other cooperative players, the pool is made up mainly of cheaters. Therefore a cheater has only a small chance to find a new partner that can be exploited.
As can be seen, Schüßler started off with some arbitrary and at best plausible assumptions about an “anonymous market” that are in no way related to any specific empirical situation (points one and two in the above list of features of Axelrod-style simulations). But Schüßler also had a deeper motivation for his simulation experiments, which brings us to the third point: the general conclusions that are derived from the simulation results. With his simulation that showed that cooperation could even emerge on “anonymous markets” Schüßler wanted to provide arguments against sociological normativism. Sociological normativism is by Schüßler understood as the thesis that social order cannot be upheld without social cohesion and the appeal to common norms. The classical proponents of sociological normativism are - among others - Ferdinand Tönnies with his distinction of “Gesellschaft” (society) and “Gemeinschaft” (community) and Emile Durkheim, who greatly emphasized the importance of social bonds. Schüßler's simulation is linked with the problem of sociological normativism in so far as it proves the “logical possibility” (Schüßler) of norm conformant behavior (if cooperation is taken as normatively desired in this case) even under the absence of authority or other previously fixed coordination mechanisms such as social cohesion. But does the proof of this “logical possibility” really establish a strong point against sociological normativism? This is not at all the case. The fact that something is logically possible does not even remotely imply that it is possible in reality. When sociological normativists speak for the importance of social bonds they usually do not mean to assert that it is by logical necessity that the social order requires some level of cohesion to function properly. Rather, they draw on the social character of human nature. Therefore, in order to refute them, one has to show why their conception of human nature is wrong or that the empirical support for their claims is inconclusive and could be interpreted otherwise. Claims about mere logical possibilities as they appear in the highly stylized and artificial setting of agent based simulations are notoriously weak arguments in sociological discussions. Not the least so because it would most probably be easy to draw up Axelrod-style computer simulations where under different but equally plausible boundary conditions cooperation is bound to break down when social ties are weakend.
To do Schüßler justice it must be mentioned that he is fully aware of the just mentioned explanatory limits of his computer simulations and that he discusses them frankly and with great intellectual honesty (Schuessler 1997, p. 91f.). It is only that doing so he makes the reader wonder why he filled a whole book with computer simulations that demonstrate so little. The same questions could be asked for many of the simulations that have been done on the topic of the “evolution of cooperation”. Most authors were, like Schüßler, more careful than Axelrod in drawing sweeping conclusions from their computer simulations. But if no conclusions can be drawn from them, the question inevitably arises what these computer simulations are good for after all. It is this question that has become crucial in the case of Axelrod style simulations. In order to answer it, let us see how Axelrod style simulations fare when it is attempted to employ them in the context of an explanation of some real world phenomenon.