Hurray! Epstein’s ‘generative’ social science is ‘recursive’ or ‘effectively computable’ social science!
I’m finding recent reading on agent-based modeling profoundly refreshing. I’ve been discovering a number of writers with a level of sanity about social science and computation that I have been trying to find for years.
I’ve dipped into Joshua Epstein’s Generative Social Science: Studies in Agent-Based Computational Modeling (2007), which the author styles as a sequel to the excellent Growing Artificial Societies: Social Science from the Bottom Up (1996). Epstein explains that while the first book was a kind of “call to arms” for generative social science, the later book is a firmer and more mature theoretical argument, in the form of a compilation of research offering generative explanations for a wide variety of phenomena, including such highly pertinent ones as the emergence of social classes and norms.
What is so refreshing about reading this book is, I’ll say it again, the sanity of it.
First, it compares generative social science to other mathematical social sciences that use game theory. It notes that, though there are exceptions, the problem with these fields is their tendency to see explanation in terms of Nash equilibria of unboundedly rational agents. There’s lots of interesting social phenomena that are not in such an equilibrium–the phenomenon might itself be a dynamic one–and no social phenomenon worth mentioning has unboundedly rational agents.
This is a correct critique of naive mathematical economic modeling. But Epstein does not throw the baby out with the bathwater. He’s advocating for agent-based modeling through computer simulations.
This leads him to respond preemptively to objections. One of these responses is “The Computer is not the point”. Yes, computers are powerful tools and simulations in particular are powerful instruments. But it’s not important to the content of the social science that the simulations are being run on computers. That’s incidental. What’s important is that the simulations are fundamentally translatable into mathematical equations. This follows from basic theory of computation: every computed program is equivalent to some mathematical function. Hence, “generative social science” might as well be called “recursive social science” or “effectively computable social science”, he says; he took the term “generative” from Chomsky (i.e. “generative grammer”).
Compare this with Cederman’s account of ‘generative process theory‘ in sociology. For Cederman, generative process theory is older than the theory of computation. He locates its origin in Simmel, a contemporary of Max Weber. The gist of it is that you try to explain social phenomena by explaining the process that generates it. This is a triumphant position to take because it doesn’t have all the problems of positivism (theoretical blinders) or phenomenology (relativism).
So there is a sense in which the only thing Epstein is adding on top of this is the claim that proposed generative processes be computable. This is methodologically very open-ended, since computability is a very general mathematical property. Naturally the availability of computers for simulation makes this methodological requirement attractive, just as ‘analytic tractability’ was so important for neoclassical economic theory. But on top of its methodological attractiveness, there is also an ontological attractiveness to the theory. If one accepts what Charles Bennett calls the “physical Church theory”–the idea that the Church-Turing thesis applies not just to formal systems of computation but to all physical systems–then the foundational assumption of Epstein’s generative social science holds not just as a methodological assumption.
This was all written in 2007, two years before Lazer et al.’s “Life in the network: the coming age of computational social science“. “Computational social science”, in their view, is about the availability of data, the Internet, and the ability to look at society with a new rigor known to the hard sciences. Naturally, this is an important phenomenon. But somehow in the hype this version of computational social science became about the computers, while the underlying scientific ambition to develop a generative theory of society was lost. Computability was an essential feature of the method, but the discovery (or conjecture) that society itself is computation was lost.
But it need not be. Just a short dip into it, Epstein’s Generative social science is a fine, accessible book. All we need to do is get everybody to read it so we can all get on the same page.
Cederman, Lars-Erik. “Computational models of social forms: Advancing generative process theory 1.” American Journal of Sociology 110.4 (2005): 864-893.
Epstein, Joshua M., and Robert L. Axtell. “Growing artificial societies: Social science from the bottom up (complex adaptive systems).” (1996).
Epstein, Joshua M. Generative social science: Studies in agent-based computational modeling. Princeton University Press, 2006.
Lazer, David, et al. “Life in the network: the coming age of computational social science.” Science (New York, NY) 323.5915 (2009): 721.