Portal de Periódicos da CAPES

Active internalism and open dynamical systems

2011; Taylor & Francis; Volume: 25; Issue: 1 Linguagem: Inglês

10.1080/09515089.2011.569919

1465-394X

Jeff Yoshimi,

Plant and Biological Electrophysiology Studies

Abstract The question whether cognition is subserved by internal processes in the brain (internalism) or extends in to the world (active externalism) has been vigorously debated in recent years. I show how internalist and externalist ideas can be pursued in a common framework, using (1) open dynamical systems, which allow for separate analysis of an agent's intrinsic and embodied dynamics, and (2) supervenience functions, which can be used to study how low-level dynamical systems give rise to higher-level dynamical structures. Keywords: Active ExternalismDynamical Systems TheoryEmbodied CognitionInternalismSupervenience Acknowledgements I am grateful to Bill Bechtel, Rick Dale, Scott Hotton, Georg Theiner, an anonymous referee, and audiences at two conferences where earlier versions of this paper were presented—Embodied, Embedded, Enactive and Extended Cognition, at the University of Central Florida, and a meeting of the Central Valley Philosophy Association, at the College of the Sequoias. Notes Notes [1] I refer to this debate as the “embodiment debate.” I take “externalism” to be a covering term for positions that emphasize the importance of the environment in cognitive processes. “Externalism” in this sense covers active externalism, semantic externalism, embodiment, the enactive approach to perception, etc. When a more specific approach is in question, and the context does not make this clear, that position is referred to using appropriate terminology. [2] Recent treatments, with references to earlier work, are Chemero (Citation2009), Clark (Citation2008), Gibbs (Citation2006), Kiverstein and Clark (Citation2009), and Noë (Citation2004). [3] For overview of the internalist response, see Adams and Aizawa (Citation2008), and Rupert (Citation2009). The internalist position is also treated in detail in Clark (Citation2008). [4] Similar approaches exist in the literature. Clark (Citation2008) endorses externalism for much cognitive processing, but is also an internalist about consciousness (2009). Wilson describes a “moderate externalist view,” “a kind of pluralistic view of the mind vis-à-vis the debate over individualism, whereby individualistic and externalist views of cognition divide the mind between them” (2010a, p. 277) Wilson (Citation2001) also provides a set of tools for distinguishing different aspects of realization, which is similar to the set of tools described here. Bechtel (forthcoming) shows how mechanistic explanation often requires simultaneously considering internal dynamics and external couplings, at multiple levels of analysis. Also note that the tools I use—what I call “open dynamical systems” and “supervenience functions”—involve formalization of concepts associated with existing areas of research (more in section 3). [5] When I refer to a “system” or “object,” I mean a real thing or collection of things, something like a spatio-temporal particular. I denote systems (and their parts) by upper case letters. When I refer to “states” and “state spaces,” by contrast, I mean ways these systems can be. More specifically, I take state spaces to be a special kind of property set, and states to be members of such sets (Yoshimi, forthcoming). I denote state spaces by upper case bold letters. For example, if T is a total system, and A is an agent system “in” T (i.e., a mereological part of T), then we can talk about a state space S 1 for A, which is a set of ways A could be, and a separate state space S 2 for T, which is a set of ways T could be. As we will see, multiple state spaces can apply to the same system. [6] As Rupert says, “proponents of the extended view often rest their case on observations about dependence” (2010, p. 2; see also Rupert, Citation2004, section 2). [7] Also compare what Donald (Citation1991) calls “exograms” (e.g., written texts) which serve as external stores of memory (see also Sutton, Citation2010). [8] This was only labeled the “parity principle” later (Clark, Citation2008). Compare Adams and Aizawa (Citation2008) on “cognitive equivalence.” For critical discussion of the parity principle, see Menary (Citation2006); Sutton (Citation2010); and Wilson and Clark (Citation2009). The issues raised in these papers are addressed in section 4. [9] See Clark (Citation2008, p. 88). [10] For a helpful review, history, and analysis of internalism, see Bartlett (Citation2008). [11] One property of supervenience is that (roughly), if A supervenes on B then A also supervenes on any set of properties that in some sense “includes” B. For example, if conscious states supervene on brain states, then conscious states also supervene on world states (if changes in conscious states entail changes in brain state, they thereby entail changes in the world those brains are part of). Thus, it is sometimes stipulated that a given base set of properties be minimal, i.e. the “smallest” set of properties B that A supervenes on (e.g., for the case of conscious states, this minimal set might contain states of some subsystem of the brain). This concept of a “minimal subvenient base” has not, to my knowledge, ever been made precise. Throughout when referring to “subvenient bases” I have this kind of minimal subvenient base in mind. For example, when referring to the brain as the subvenient base of consciousness, I assume that there is actually some mereological part of the brain which is the minimal subvenient base of consciousness. [12] As Bartlett (Citation2008) says, there was no “organized defense” of internalism until recently; in fact, the position was initially formulated as a foil for critics to attack. [13] Among Greek and Roman scholars there was a debate between those who thought the heart was the seat of the soul (e.g., Aristotle), and those who thought the brain was (e.g., Plato). The Roman physician Galen is said to have settled the issue. In his most famous demonstration, he removed the laryngeal nerves from a pig, and observed that this eliminated vocalization but no other functions. Other experiments showed that damage to specific parts of the brain had a specific impact on cognition (e.g., inducing blindness), even if the sense organs were intact (Gross, Citation1988). [14] Clark (Citation2009) is critical of this type of evidence (wrongly I think), but I will not respond to his arguments here (though I will respond to related worries in section 4). Clark also describes a novel way of supporting a neural basis for consciousness. Citing an earlier suggestion by Chalmers, Clark notes that relatively high bandwidth communication channels within the brain (as compared with lower bandwidth channels connecting the brain to its environment) make it the “only adequate ‘vehicle’” for consciousness (2009, p. 983). [15] Compare Grush's (Citation2003) arguments concerning the control-theoretic functions of the nervous system. [16] On symbolic dynamics in relation to cognitive science, see Dale and Spivey (Citation2005); and Atmanspracher and beim Graben (Citation2007). On the relationship between recurrent neural networks and state machines, see Casey (Citation1996), which includes references to earlier work. On both topics see beim Graben and Potthast (Citation2009). These ideas are discussed in relation to the philosophical literature on supervenience in Yoshimi (forthcoming). [17] Also see Phattanasri, Chiel, and Beer (Citation2007). [18] Dynamical systems can be formally defined (see Hotton & Yoshimi, Citation2010, for a discussion of the formal definition), but we will not consider the formal definition here. Also note that there is a common equivocation at this point, between the formal mathematical structure and the real system being modeled (van Gelder, Citation1998). The equivocation is mostly harmless and will be ignored in what follows. [19] The term “orbit” is standard in dynamical systems theory. The term “path” is neutral between classical dynamical systems and open dynamical systems, and so I will prefer that term here. [20] I am using the phrase “phase portrait” in a somewhat non-standard way. The phrase usually refers to a visual depiction of selected orbits of a dynamical system (e.g., Figure 1). I also use it to refer to a logical rather than a graphical object—namely, the full set of orbits of a dynamical system, what is sometimes called an “orbit space.” [21] Of course, outside influences can be incorporated, but then they are taken to be part of a larger dynamical system which is not subject to external influences (a useful strategy, as we’re about to see). [22] As noted above, others have analyzed the dynamics of agents in environments. Our contribution is to formalize and generalize these ideas using an explicit mathematical framework, which in turn provides (among other things) for a more rigorous analysis of certain dynamical concepts, e.g., “hysteresis.” We also describe tools for analyzing some of the complex structures that occur in an open system. These points are elaborated in Hotton and Yoshimi (Citation2010), and Hotton and Yoshimi (forthcoming). In those papers we also differentiate open dynamical systems in our sense from open thermodynamic systems and open loop control systems. [23] In what follows, I distinguish open dynamical systems, which are compound systems like the one shown in Figure 2, from open systems or “embodied agent systems,” which correspond to one part of an open dynamical system, namely the agent system when it is embedded in an environment (see the bottom right panel in Figure 2). [24] The act of “placing” the agent system in the total space involves more formalism than this (Hotton & Yoshimi, forthcoming), but the intuitive idea should be clear. In this case, a dynamical system that consists of an infinite collection of copies of the agent system is defined on the total space. Placing the agent system in the total space then amounts to associating the agent system with one of these copies. [25] Technically, there is also a third classical dynamical system, corresponding to the embedding of the agent system in the total state space. [26] Compare Wilson's (Citation2010a) concept of “functional gain,” a property of integrated systems whereby they display novel functions compared to the behavior of any of their components considered in isolation. Also see Wilson and Clark (Citation2009). [27] In Hotton and Yoshimi (Citation2010), and Hotton and Yoshimi (forthcoming), techniques for dealing with this complexity are described (e.g., grouping paths into collections of paths that do roughly the same thing, and then color coding these collections). [28] See Yoshimi (forthcoming) for discussion of the relation between supervenience and dynamics, and for review of authors who distinguish diachronic causal process and synchronic determination relations. In the context of the embodiment debate, this kind of distinction has been invoked with aims similar to my own. Adams and Aizawa (Citation2001, 2008) distinguish coupling (which is dynamic) from constitution (which is static). Clark, drawing on such philosophers as Susan Hurley and Ned Block, distinguishes “instrumental dependence” (dynamic) and “constitutive involvement” (which is static; 2009, p. 965). I say more on this in section 4. [29] I have discussed elsewhere how standard formulations of supervenience imply this functional formulation (Yoshimi, Citationforthcoming). Varieties of supervenience such as strong, weak, and global (see McLaughlin & Bennett, Citation2005, for review) imply different functional formulations. For example, strong supervenience entails a single supervenience function prefixed by a modal parameter (so that base states determine the same supervenient states in all possible worlds), while weak supervenience entails a family of supervenience functions, one for each possible world. These differences can largely be bracketed here, though in a fuller treatment varieties of supervenience (including explicit modal parameters) could usefully enrich the overarching framework, with its emphasis on adaptability to different cases. [30] The function need not be defined on the entire domain; there are most likely brain states that don’t determine any conscious state, and similarly in other cases of supervenience. This introduces trivial complications, which I deal with elsewhere. [31] Note that, while the supervenient state set in this example is discrete, this need not be the case; it could be a continuum. [32] As Clark says, “inscriptions in the notebook figure as part of the physical supervenience base for certain standing beliefs of the agent” (2009, p. 966). [33] Matters are even more complex than this. Having specified a base space of interest, it is sometimes the case that a hierarchy of spaces supervenes on that base space. For example, physical states of the brain determine chemical states, which determine biological states, which sometimes determine psychological states. This situation has an interesting analysis in terms of the partitions described above, whereby as one moves up the hierarchy of supervenience functions, increasingly coarse-grained partitions are induced on the lowest level base space. [34] See Wilson (Citation2010a), who refers to “systems of various levels of durability and robustness over time and circumstance” (see also Wilson & Clark, Citation2009). [35] Recall that this was Haugeland's example in discussing “causal intimacy.” For review see Chiel and Beer (Citation1997). [36] Compare the analysis of dissociated cortical cultures (which allow isolated brain tissue to be studied), or for that matter just about any analysis of a biological system in vitro. [37] Also see Clark's (Citation2008) discussion of “coarse grained functional poise,” and Bartlett's (Citation2008) discussion of “core realizers.” [38] Clark counters this argument in a similar way; see (Citation2009, pp. 970–987). Also see the discussion of “Searle's objection” in Noë (Citation2004, p. 219). [39] On “soft assembly” and related concepts, see Bingham (Citation1988). [40] Pluralist approaches to cognitive science have been a subject of considerable recent discussion. See, e.g., Dale, Dietrich, and Chemero (Citation2009), McClelland et al. (Citation2010), and Spivey and Anderson (Citation2008). [41] Of course, the tools I’ve described are neutral with respect to the philosophical positions. In particular, they could be used by someone with strictly internalist or externalist leanings.