Working with very little: what do we get out of neutral (agent-based) models?

Carlos Santana

Most philosophical agent-based models (ABMs) are highly abstract, raising questions about their utility. Simple models may be useful because they are broadly generalizable, or easily comprehensible (Weisberg 2012), but it is far from guaranteed that a more complex model can’t better achieve these and other modeling aims (Evans et al. 2013). Moreover, given that philosophical ABMs tend not to be mere computational instantiations of well-confirmed physical theories, some of the leading accounts of the epistemology of simulation (Parker 2008; Winsberg 2009) are a poor fit for how simulations are used in philosophy.

Philosophers of science, of course, have developed accounts of how various types of highly abstract models work. These accounts typically focus on what sort of causal factors or mechanisms are central to the model. Minimal models1 “include only the core causal factors which give rise to a phenomenon” (Weisberg 2007). Since real-world systems tend to be complicated, minimal models are usually “thoroughgoing caricatures of real systems” (Batterman and Rice 2014), but by focusing on important difference makers these models can increase our (explanatory) understanding and contribute to effective intervention. A distinct class of hyper-simple models are how-possibly models which abstract to causal factors or mechanisms which may not be operating in nature (Bokulich 2014), yielding how-possibly explanations, which are weaker than how-actually explanations. Additionally, how-possibly models can refute impossibility claims.

Many highly abstract ABMs in philosophy are minimal models or how-possibly models, and their utility can be justified correspondingly. But other AMBs aren’t reasonably categorized as either because instead of abstracting to known core causes or uncertain possibly causes, they abstract to causes we know not to be the chief actual causes. (See Table 1) These models, which I’ll call neutral models, are undertheorized, and my aim is to give an account of their epistemic utility.

Model Type Highly abstract? What mechanisms are included?
Minimal model Yes Stuff we think are the actual core factors
How-possibly model Yes Stuff that may or may not be relevant
Neutral model Yes Stuff we think aren’t actual core factors

Given the philosopher’s penchant for working on the hardest problems, we often find ourselves using neutral models. To understand how neutral models can be useful, I look to some examples of successful(-ish) agent-based neutral models in other sciences. In particular, I look at neutral models of biogeography and biodiversity (Rosindell et al. 2012), models of language diversification (Cangelosi and Parisi 2002), and Schelling’s segregation model (1969). As O’Connor (2017) argues of the Schelling model, these neutral models can help us discover when interventions may be unsuccessful by helping map out the space of counterfactuals. But I identify two other benefits of neutral models. First, they force us to reevaluate our existing empirical evidence. The neutral models in ecology, for instance, call into question whether the data actually confirm the leading theories, since they show that the data match a broader set of possibilities than we had thought. Second, they help identify robust properties of complex systems. This goes beyond merely identifying when manipulations are unlikely to succeed or, conversely, are unnecessary. It also contributes to understanding those systems at a system level, and helps us classify systems into relevant kinds. The epistemic benefits of neutral models are thus surprisingly broad.

I then take these insights and apply them to three types of philosophical ABMs which sometimes (but not always) function as neutral models: signaling games (e.g. Skyrms 2010), scientific networks (e.g. Grim et al. 2013), and models of inequality (Bruner 2017). To the extent that these models are neutral models, we can now say something about what sort of lessons we can take from them. Just as importantly, we can say something about what sort of lessons we aren’t licensed to take. My conclusion is thus that the concept neutral model provides us both lessons and warnings about how to use ABMs in philosophy and other sciences.


  • Batterman, R. W., & Rice, C. C. (2014). Minimal model explanations. Philosophy of Science, 81(3), 349-376.
  • Bokulich, A. (2014). How the tiger bush got its stripes:‘How possibly’vs.‘how actually’model explanations. The Monist, 97(3), 321-338.
  • Bruner, J. P. (2017). Minority (dis) advantage in population games. Synthese, 1-15. Cangelosi A, & Parisi D (2002) Simulating the evolution of language (Vol. 1). London: Springer.
  • Evans, M. R., Grimm, V., Johst, K., Knuuttila, T., De Langhe, R., Lessells, C. M., … &Wilkinson, D. J. (2013). Do simple models lead to generality in ecology?. Trends in ecology & evolution, 28(10), 578-583.
  • Grim, P., Singer, D. J., Fisher, S., Bramson, A., Berger, W. J., Reade, C., … & Sales, A. (2013). Scientific networks on data landscapes: question difficulty, epistemic success, and convergence. Episteme, 10(4), 441-464.
  • O’Connor (2017) Modeling Minimal Conditions for Inequity. Unpublished manuscript
  • Parker W (2008) Franklin, Holmes, and the epistemology of computer simulation. International Studies in the Philosophy of Science 22(2), 165-183.
  • Rosindell, J., Hubbell, S. P., He, F., Harmon, L. J., & Etienne, R. S. (2012). The case for ecological neutral theory. Trends in ecology & evolution, 27(4), 203-208.
  • Schelling, Thomas C. 1969. “Models of Segregation.” The American Economic Review 59 (2): 488–93.
  • Skyrms, B. (2010). Signals: Evolution, learning, and information. Oxford University Press.
  • Weisberg, M. (2007). Three kinds of idealization. The Journal of Philosophy, 104(12), 639-659.
  • Weisberg, M. (2012). Simulation and similarity: Using models to understand the world. Oxford University Press.
  • Winsberg E (2009) Computer simulation and the philosophy of science. Philosophy Compass, 4(5), 835-845.



This term gets used in all sorts of ways in the literature, but I’m following the usage in Weisberg (2007)

Author: Research Group for Non-Monotonic Logics and Formal Argumentation

Created: 2019-02-26 Tue 09:56