Optimised agent-based modelling of action selection

Abstract

Summary The problem of action selection has two components: what is selected? How is it selected? To understand what is selected, it is necessary to recognise that animals do not choose among behaviours per se; rather, behaviour reflects observed interactions among brains, bodies, and environments (embeddedness). To understand what guides selection, it is useful to take a normative, functional perspective that evaluates behaviour in terms of a fitness metric. This perspective can be especially useful for understanding apparently irrational action selection. Bringing together these issues therefore requires integrating function and mechanism in models of action selection. This chapter describes ‘optimised agent-based modelling’, a methodology that integrates functional and mechanistic perspectives in the context of embedded agent–environment interactions. Using this methodology, I demonstrate that successful action selection can arise from the joint activity of parallel, loosely coupled sensorimotor processes, and I show how an instance of apparently suboptimal decision making (the matching law) can be accounted for by adaptation to competitive foraging environments. Introduction Life is all about action. Bodies and brains have been shaped by natural selection above all for the ability to produce the right action at the right time. This basic fact leads to two observations. First, the neural substrates underpinning action selection must encapsulate mechanisms for perception as well as those supporting motor movements (Friston, 2009), and their operations must be understood in terms of interactions among brains, bodies, and environments. In other words, action selection mechanisms are embodied and embedded . Second, despite the generality of action selection mechanisms, it is unlikely that they can deliver optimal behaviour in all possible situations. Action selection models therefore need to integrate functional and mechanistic perspectives (McNamara and Houston, 2009), especially when observed behaviour departs from what appears to be optimal or ‘rational’ (Houston et al ., this volume). The goal of this chapter is to describe and illustrate a methodology – optimised agent-based modelling (oABM; Seth, 2007) – that accommodates both of these observations, and to contrast this methodology with standard techniques in ‘optimal foraging theory’ (OFT; Stephens and Krebs, 1986). The central idea is that the oABM approach provides a unified framework for modelling natural action selection, ‘rational’ and otherwise.