What cognition is about

To achieve the goals of this chapter, I need a view of behaviour – indeed, of cognition in general – that is consistent with my premise, that is, one in which behaviour and cognition can be conceived of as thoroughly historical, relational, constructive, and contingent processes. In other words, I need a view of behaviour and cognition that takes the changing animal-in-its-environment as the fundamental unit of analysis. This will not be the dominant, representational-computational view of cognition. The cognitivist view is based on the representational theory of the mind (Fodor, 1975). It starts with the theoretical suggestion that cognition involves the operation of mental processes on mental states. Mental processes refer to activities such as thinking, remembering, comparing, judging, and inferring. Mental states refer to thoughts, beliefs, desires, perceptions, memories, and so on. These states are said to have intentionality because they are about

or refer to things other than the states themselves. For example, the perception of a red apple may be about a fruit that exists in the world, and a memory state might be about an event which occurred some time ago. The representational view of cognition explains intentional states in terms of representations, and explains their intentionality in terms of the semantic properties of representations (Pitt, 2017).

In its contemporary, computational version, mental representations are taken to be analogous to computer data structures, and mental operations are taken to be analogous to computational algorithms (Thagard, 2014). Mental representations, like data structures in the computer, are considered to be information-bearing structures (Pitt, 2017). That is, they carry information (sensu what-it-is-we-can-lean-from-it) about that which they represent. For this reason, mental or cognitive processes are, like computational processes, commonly characterised in terms of the ‘processing of information’ (although it would be more coherent to say processing of data). Most behavioural biologists work from this perspective and commonly conceive of learning in terms of the acquisition of task-relevant ‘information’, and social learning as a process of ‘transmission of

70 Chapter 4 information’ from one animal to another (Boyd & Richerson, 1985; Mesoudi, 2011; Whiten, Horner, Litchfield, & Marshall-Pescini, 2004). In these cases, the term ‘information’ is used somewhat loosely to mean, in fact, ‘representation’, possibly as a shorthand for ‘information-bearing structure’. It is therefore intimately associated with the notions of (mental) representation and computation.

4.2.1

Changing the basic metaphor: from computation to dynamics

In this thesis I reject representational-computational views of cognition as a starting point. However, my intention in the rest of this section is not to argue against it but simply to contrast it with an alternative which is non-representational and non-computational and therefore consistent with the approach I am developing.

In a paper titled “What might cognition be, if not computation?” Tim Van Gelder (1995) famously discussed the operation of the governor of steam engines, developed by Scottish engineer James Watt, to argue that cognition can be understood in dynamical rather than computational terms (see also a relevant discussion of this paper in Chemero, 2009, chapter 4). The answer Van Gelder gives to the question in his title is the following. “Rather than computers, cognitive systems may be dynamical systems; rather than computation, cognitive processes may be state-space evolution within these very different kinds of systems.” (p. 347).

I will follow this suggestion and side with the alternative view known as radical embodied cognitive science. I use this term after Chemero (2009) to refer to two traditions. One is the ecological psychology developed by James J. Gibson and Eleanor J. Gibson, and followers (E. J. Gibson & Pick, 2000; J. J. Gibson, 1966, 1979/2015; Kelso, 1995; Lee, 2009; Turvey, Shaw, Reed, & Mace, 1981) The other is the enactive approach initially proposed by Varela et al. (1991), following the insights from the theory of autopoiesis which Varela developed with his former professor Humberto Maturana (Maturana & Varela, 1973, 1980, 1992). Although the two approaches developed as different traditions, it has been increasingly recognised that they share more similarities than differences, and that it might be possible to bring them closer together conceptually (Chemero, 2009; McGann, 2014; Thompson & Varela, 2001).

The flow of behaviour 71 I am less interested in what sets these two traditions apart than in what makes them natural allies and for the most part I will treat them together under the banner of radical embodiment, still making distinctions where appropriate for clarity. This is how Chemero (2009, p. 160) summarises the approach:

Here, then, is radical embodied cognitive science: animals are active perceivers of and actors in an information-rich environment, and some of the information in the environment, the information to which animals are especially attuned, is about affordances. Unified animal-environment systems are to be modelled using the tools of dynamical systems theory. There is no need to posit representations of the environment inside the animal (or computations thereupon) because animals and environments are taken, both in theory and models, to be coupled.

This is, indeed, consistent with my purpose. The use of the term information in the context of radical embodiment is different from its use in the computational approach. I will return to this point below. For now, I want to consider a few issues to illustrate how the dominant approach (computational, representational, cognitivist) and the alternative approach (non-computational, non-representational, ecological, enactive, radical embodiment) diverge. According to the cognitivist view, perception involves forming meaningful representations from the meaningless stimuli that arrive at the sensory organs by means of computational processes operating on them. In contrast, according to the radical embodiment alternative, perception occurs in the dynamics of the coupled animal- environment system and it is therefore unnecessary to posit representations of the world inside the animal. In saying this I do not mean to be providing an explanation for how perception occurs but only to clarify how the different approaches suggest different starting points for theoretical and empirical studies of perception.

In the cognitivist approach, action (overt behaviour) is taken to be the expression of motor programs, the output of rule-based manipulations of representations both obtained in perception and retrieved from memory. This implies that the form of overt behaviour – how the animal moves in space and time during a period of time – is somehow given in advance in the posited motor program prior to its appearance in actual bodily movements. The alternative perspective rejects this version of preformationist thinking. The form of behaviour cannot possibly be given in the nervous system prior to its

72 Chapter 4 appearance because, in addition to the nervous system, other components – on either side of the skin – play causal roles in determining the flow of behaviour in real time.

The suggestion that behavioural form is already encoded in a previous motor program prior to its appearance in the actual behaviour is analogous to, and as problematic as, the notion that phenotypic form is already encoded in a genetic program prior to its appearance in ontogeny (Oyama, 1985/2000). From a developmental systems perspective, biological form – including behavioural form – is neither pre-existent nor ever finished. Rather, form appears and persists under transformation as the result of historical, constructive interactions going on among the components of the organism’s developmental system, as ontogeny unfolds. In a nutshell, behaviour does not appear in the runtime execution of an already computed algorithm, but in the real-time engagement of the animal with its environment.