10 Action-Oriented Models of Cognitive Processing A Little Less Cogitation , A Little More Action Please

This chapter considers action-oriented processing from a model-oriented standpoint. Possible relationships between action and cognition are reviewed in abstract or conceptual terms. We then turn to models of their interrelationships and role in mediating cognitively enriched behaviors. Examples of theories or models inspired by the actionoriented paradigm are briefl y surveyed, with a particular focus on ideomotor theory and how it has developed over the past century. Formal versions of these theories are introduced, drawing on formulations in systems biology, information theory, and dynamical systems theory. An attempt is made to integrate these perspectives under the enactivist version of the Bayesian brain; namely, active inference. Implications of this formalism and, more generally, of action-oriented views of cognition are discussed, and open issues that may be usefully pursued from a formal perspective are highlighted. Existing Schema for Action and Cognition Before considering the form and consequences of models that take an actionoriented view of cognition, it is worth considering how the relationship between these two processes has been described. Here, we consider four schemata which capture different notions of how cognition and action could be coupled (Figure 10.1). It is important to stress that these schemata are not models but From “The Pragmatic Turn: Toward Action-Oriented Views in Cognitive Science,” Andreas K. Engel, Karl J. Friston, and Danica Kragic, eds. 2016. Strüngmann Forum Reports, vol. 18, series ed. J. Lupp. Cambridge, MA: MIT Press. ISBN 978-0-262-03432-6. 160 J. Kilner et al. rather depictions of different views on cognition and action. Crucially, differences in the schemata do not refl ect a fundamental difference in the nature of the coupling but rather in how cognition and action are defi ned. At one extreme, in Figure 10.1a, the two processes can be considered to be entirely independent processes: action is simply the behavioral output of the cognitive process (e.g., in the classical sandwich conception of the mind). In this open-loop formulation, action does not, and indeed cannot, infl uence cognition. Although not explicitly stated, the scheme depicted in Figure 10.1a is implicitly assumed in many models of high-level human cognitive functions used in behavioral and imaging experiments: input is carefully manipulated and output (action) is kept to a minimum (e.g., key pressing) in order to focus on internal processes. Here, action is considered to be a necessary output to disclose internal operations. For example, in a typical experiment that investigates language processing, individuals are presented with written or spoken words/sentences and asked to make a key press decision about them. Such studies, which still constitute the majority of cognitive science and neuroscience studies of high-level cognition, are completely silent as to whether action might play a role. Alternative approaches (Figure 10.1b–d) consider how action and cognition depend on each other. In addition, action can be explicitly considered to be a subset of cognition or a largely overlapping process (Figure 10.1b and c). This commonly held view defi nes cognition as information processing à la Neisser. Accordingly, most, but not all, cognition relates to action: cognition infl uences action, and action infl uences cognition. This infl uence is typically, but not always, constructive: cognition and action can exist in harmony without collapsing into one another. They are heavily intertwined, but they are not the same, and none exists solely for the benefi t of the other. Just like attention, working memory, and consciousness, cognition and action are intimately related, yet (a)