Computer Models in the Social Sciences

In their book Growing Artificial Societies: Social Science from the Bottom Up (MIT, 1996, pp. 1-2), Joshua Epstein and Robert Axtell highlight various serious methodological difficulties that have particularly beset the social sciences:

  • Social systems aren't "neatly decomposable into separate subprocesses", despite the institutional separation of social science departments – Economics, Demography, Political Science etc. – that attempt to treat them separately.
  • Controlled experimentation of hypotheses in the social sciences is typically impossible, for a range of obvious practical (and moral) reasons.
  • Some methodologies – notably in Economics – attempt to impose order by relying on the fiction of the perfectly "rational actor" (perfectly informed and with infinite computing capacity), but this is hugely unrealistic.
  • Even where that fiction isn't prevalent, standard social science models are forced by practical considerations to suppress real-world heterogeneity (e.g. through "representative agent" models in macroeconomics, or aggregative models).
  • Heterogeneity can be crucial in social contexts, yet until the rise of agent-based computer models, "there has been no natural methodology for systematically studying highly heterogeneous populations".
  • There has likewise been no natural methodology for studying dynamic changes, and hence "social science, especially game theory and general equilibrium theory, has been preoccupied with static equilibria, and has essentially ignored time dynamics".

Epstein and Axtell go on to promote a notion of what they call Generative Social Science, based on interpreting the question, "can you explain it?" as asking "can you grow it?" within an agent-based model. The NetLogo system provides a widely accessible framework for such models, including the following:

Growing Artificial Societies, Epstein and Axtell

Growing Artificial Societies, by Joshua M. Epstein and Robert Axtell