How Models Fail
|Table of Contents|
|2 The empirical failure of simulations of the evolution of cooperation|
|3 Justificatory narratives|
|3.1 Axelrod's narrative|
|3.2 Schüßler's narrative|
|3.3 The story of “slip stream altruism”|
|3.4 The social learning strategies tournament|
|4 Bad excuses for bad methods and why they are wrong|
|5 History repeats itself: Comparison with similar criticisms of naturalistic or scientistic approaches|
Although RPD simulations already fell out of fashion, I have myself published a book with RPD simulations as late as 2008. I felt uneasy about it at the time of writing the book and today I am even more convinced that the scientific method that I describe (but also criticize) in this book is fundamentally flawed. But the book was my PhD-thesis and I was not really given the free choice of topic - which is, of course, a widespread grievance of PhD-theses. So, I figured that the best I could make out of this situation was to follow the established pattern of research in this field, but also to examine it from an epistemological point of view and point out its deficiencies. The research pattern is that of producing a variant of an existing simulation model, finding “interesting” results and embedding them in a narrative that makes them appear “new”, “surprising” or at least somehow noteworthy.
In the series of population dynamical simulations of the RPD that I conducted, there are quite a few simulations where naive cooperators, i.e. strategies that cooperate but other than TIT FOR TAT do not retaliate when the partner fails to reciprocate, can still survive with a low share of the population or - even more “surprising” - come out on top, i.e. with larger population share than even the retaliating cooperators (Arnold 2008, 109ff.). I used the term “slip stream altruism” as a catch phrase to describe this phenomenon, because the simulations prove the logical possibility that unconditional altruism (which some moralists consider to be the only form of altruism that deserves its name) can develop in the “slip stream” of tough, reciprocating strategies.
But is this phenomenon really surprising and did we really need a series of computer simulations to get the idea? As mentioned earlier, with unrestricted modeling freedom and a volatile base model like the RPD, one is liable to find all kinds of phenomena. There are not really any surprises. And just as in Schüßler's case there is a simple explanation for the phenomenon: Unconditional cooperators can come out on top, if the conditional cooperators that drive the non-cooperators to extinction are badly coordinated so that they inadvertently hurt each other (Arnold 2008, 113). So, the phenomenon that my simulation series yields acquires the appearance of being interesting, surprising or relevant mostly by the narrative and the rhetoric of “slip stream altruism” in which it is embedded.
I never took the story of slip stream altruism very seriously and, as I said earlier, I was already convinced that the simulation method as practiced by Axelrod and his followers leads to nothing at the time when I wrote the book down. (See, for example, my talk at the Models & Simulations in Paris 2006, some time before I wrote down the book (Arnold 2006).) Given how strongly I criticize Axelrod-style simulations in the book, it may appear odd to the readers that I even bothered to conduct computer simulations of the same brand and describe them in the book. As mentioned earlier, this was a tribute that I had to pay to the circumstances. Somewhat to my distress I later found that some readers liked the simulation series much better than my criticism of the method (Schurz 2011, 344, 356). Others, at least, have understood that the main purpose of the book is a critical one (Zollman 2009). In my (biased) opinion, however, I believe that the criticism or, what amounts to the same, the deficiencies of the simulation method as practiced by the adherents of Axelrod have not yet been taken seriously enough.