Dynamical systems and embedded cognition

Abstract

The conceptual frameworks that we bring to our study of cognition can have a tremendous impact on the nature of that study. They provide a set of filters through which we view the world, influencing our choice of phenomena to study, the language in which we describe these phenomena, the questions we ask about them, and our interpretations of the answers we receive. For much of the last fifty years, thinking about thinking has been dominated by the computational framework, the idea that systems are intelligent to the extent that they can encode knowledge in symbolic representations which are then algorithmically manipulated so as to produce solutions to the problems that these systems encounter (see Chapter 4 of this volume). More recently, the connectionist framework forced an important refinement of the computational framework, in which representation and computation could be distributed across a large number of loosely neuron-like units (see Chapter 5).Beginning around the mid 1980s, just as the popularity of connectionism was rising, another conceptual framework appeared (or, as in the case of connectionism, reappeared) on the scene. This framework, which, for want of a catchier label, I will call the situated, embodied, dynamical (SED) framework, focuses on concrete action and emphasizes the way in which an agent's behavior arises from the dynamical interaction between its brain, its body, and its environment. In this chapter, I will attempt to trace some of the history of the individual intellectual threads of situated activity, embodiment, and dynamics that underlie the SED approach. I will particularly focus on the years 1985-1995. Although there were important precursors to the SED approach (some of which I will briefly mention), and work in this area has grown rapidly in recent years, many of the pivotal ideas were first given their modern form during this ten-year period. Situated activityThe first intellectual thread making up the SED approach is situated activity. Roughly speaking, situated activity stresses three ideas that have been traditionally neglected in AI and cognitive science.

Dynamics of cognition

“…[17][18][19][20][21] Dynamical systems were sometimes framed as replacements for computational-representational accounts, 8,19 as distinct but complementary, 22 or even as consistent with information processing given the implicit computations performed by dynamical systems. [23][24][25] Whether and how dynamical systems accounts might be reconciled with traditional accounts, they provide a fresh perspective on many foundational problems in cognitive science, 1 including perception-action, 12,14,26 memory, 27,28 word recognition, [29][30][31][32] decision making, [33][34][35] learning, [36][37][38] problem solving, 39 and language.…”

On the Cross-Disciplinary Nature of Guided Self-Organisation

“…(Lizier et al 2008b;Crutchfield 2009;Egri-Nagy and Nehaniv 2011;Dini et al 2013), • dynamical systems, e.g. (Gros 2008;Williams and Beer 2010;Der and Martius 2012;Beer 2013Beer , 2014, • statistical mechanics and thermodynamics, e.g. (Friston 2009;Prokopenko et al 2011Prokopenko et al , 2013Still et al 2012;Wissner-Gross and Freer 2013), and • graph theory, e.g.…”

Action Switching in Brain-Body-Environment Systems

“…The dynamical hypothesis proposes a unifying philosophical stance in cognitive science, which insists that cognitive systems are dynamical systems, that they are best understood as dynamical systems, and that therefore scientists should thrive for dynamical explanation of such systems. This stance brings with it a certain understanding; it influences the questions asked, the analyses performed, and the interpretation of results (Beer 2007). Many separate lines of research have come upon this same line of reasoning, and have employed dynamical explanation of cognitive behavior.…”

“…Coupling that flows from the environment to the agent is called "sensory," and coupling that flows in the opposite direction is called "motor". The agent's behavior is defined by its trajectory of motor outputs (Beer 2007). Cognitive capacities such as memory, learning, attention, and recognition are predicted to fall out of this description if such an ideal model is obtained, not as intrinsic properties of a system but as patterns that emerge from the dynamical trajectories of the system.…”