Coherent Dynamical Convergence as the Mark of the
Cognitive: a Non-representational, Goal-based
Demarcation of Cognition
Mark Andrew Flowers
Submitted 25/06/2018
Thesis for Masters’ degree in Philosophy Department of Philosophy, University of Amsterdam Supervisors: Dr. E. C. Brouwer, Dr. J. P. Bruineberg
Second reader: Prof. Dr. M. van Lambalgen Word count: 21,978
Coherent Dynamical Convergence as the Mark of the Cognitive: a Non-representational, Goal-based Demarcation of Cognition
Contents:
1. Introduction . . . 1
1.1. The extended cognition thesis . . . 1
1.2. The central issue of the extended cognition controversy is demarcating cognition . . . 3
1.3. Research question, aims of the thesis and roadmap . . . 5 2. Representational and deflationary accounts of cognition . . . 7
2.1. The underived content account . . . 7
2.2. Rowlands’ information processing account . . . 8 2.3. The reasons-having account . . . 10
2.4. Objections to representational approaches . . . 11 2.5. Deflationary approaches to cognition . . . 15
2.5.1. What work is there left for a mark of the cognitive to do? . . . 16 2.6. Summary . . . 17
3. Coherent dynamical convergence as the mark of the cognitive . . . 19 3.1. Trestman on implicit and explicit goal-directedness . . . 19
3.2. Coherent dynamical convergence as the mark of the cognitive . . . 22 3.2.1. Coherent dynamical convergence is gradable . . . 24
3.2.2. Patterns of coherent dynamical convergence are discoverable . . . 26 3.2.3. Coherent dynamical convergence is theory neutral . . . 27
3.2.4. The normativity of cognition . . . 28 3.2.5. Ownership . . . 32
3.3. Objections and replies . . . 33
3.3.1. Barely cognitive animals, almost cognitive plants . . . 33 3.3.2. The heart and car engines . . . 34
3.3.3. Group cognition . . . 35
3.3.4. Cognition and behaviour . . . 35
3.3.5. Scaling up the minimal accounts of normativity and ownership . . . 36 3.4. Summary . . . 37
4. Conclusion . . . 39
4.1. Coherent dynamical convergence and the extended cognition thesis . . . 39 4.2. The Clunky Robot Problem and Fluency . . . 40
1 1. Introduction
Cognition comes in many forms. Solving equations, planning future actions, imagining fantastical worlds, understanding language and perceiving the world around us are all aspects of human cognition. Remembering, learning and empathising also belong to the category. These phenomena are different from worldly processes: Sisyphus pushing a boulder up a hill is a very different process than the boulder rolling back down. While the former seems to involve cognition, the latter does not. The question of which features mark this difference is a central topic in the Philosophy of Mind: what makes learning, reasoning and so on distinctively cognitive? What demarcates cognition? An influential answer in the literature is that calculating, planning, imagining etc. are all functional states that a system can be in. Planning to go the beach, for example, is a matter of being in a state that plays a certain role in a person’s thought and behaviour. The state is what it is in virtue of the role it plays in prompting other thoughts and producing behaviour - as well as the ways the state itself is responsive to other thoughts and behaviours. Your belief that it will be sunny this weekend can prompt planning to go the beach, and the planning itself can prompt you buying sunscreen. This is the core idea of a family of theories about the mind and cognition that go by the name Functionalism (cf. Putnam 1967): cognitive processes are what they are in virtue of the role they play in a system’s cognitive economy. Thinking of cognition in this way suggests that extracranial resources - notebooks, calculators, maps - can be part of cognitive processing as well. Some resources from outside the brain might play the right functional role in the production of thought and behaviour to be considered part of some cognitive process (Clark and Chalmers 1998; Wheeler 2010; see section 1.1). That Functionalism implies cognitive process can be spread across intracranial and extracranial elements has become a controversy in the Philosophy of Mind (Adams and Aizawa 2001; 2008; Rupert 2004). What is at stake in this debate seems to be the nature of cognition, or what it is that marks cognition out as distinct from everything else (Walter 2010; Rowlands 2009). Functionalism in its raw form (see Putnam 1967) does not provide clear enough demarcation criteria to settle the issue. The aim of this thesis is to develop clear demarcation criteria for cognition. I will argue that coherent dynamical convergence (Trestman 2012; see also Nagel 1977) is a sufficient criterion of demarcation. A process exhibits dynamical convergence if it reaches the same outcome given a range of different circumstances. Cognitive processes differ from worldly processes because cognitive processes maintain their trajectory towards a certain outcome. The coherence condition demands that cognitive processes work together with other processes in order to realise patterns of dynamical convergence.
1.1 The extended cognition thesis
The extended cognition thesis (EC), in its most general form, states that some cognitive processes occur partly outside the brain. Remembering, believing, problem solving and other cognitive processes are not always activities performed solely by neural machinery. Although there is no single canonical text which EC originated from, the case of Otto and Inga from Clark and Chalmers (1998) was the jumping off point for much of the initial debate about EC. Otto is imagined to have a defective memory: he must rely on information stored in a notebook which is always on his person to guide his actions. Inga has a normally functioning memory and can
2 access the relevant information using only her neural resources. Both Otto and Inga have an interest in art, and an exhibition is taking place at the Museum of Modern Art tomorrow. They both go to the exhibition. In explaining Otto’s behaviour we have to refer to the information stored in his notebook concerning the date and time of the exhibition and the location of the Museum of Modern Art. All of this information is stored outside of Otto’s head but it influences his behaviour. Inga, meanwhile, remembers all the relevant information in the standard biological way and goes to the museum. Clark and Chalmers draw our attention to the fact that the information in Otto’s notebook influenced his behaviour in the same way that information stored in Inga’s biological memory influenced her behaviour. That is, externally stored information can play the same functional role as biologically stored information in the production of behaviour. To complete the argument they employ the parity-principle:
If, as we confront some task, a part of the world functions as a process which, were it done in the head, we would have no hesitation in recognizing as part of the cognitive process, then that part of the world is (so we claim) part of the cognitive process. (Clark and Chalmers 1998, 8)
Given the functional parity of Otto’s notebook and Inga’s biological memory, they reason that if Inga’s process involving biological memory is cognitive, then Otto’s process involving external memory is cognitive too. This view is controversial and objections to it are discussed in section 1.2.
Although much of the present debate around EC was initiated by Clark and Chalmers’ paper, similar ideas have emerged elsewhere in the literature. Varela et al. (1991) brought the notion of embodied cognition to the fore, emphasising the role that the body plays in our engagement with the environment - as opposed to accepting EC on Functionalist grounds. Van Gelder (1995) argued that cognition could be modeled in terms of the interactions between brain, body and environment using the mathematical framework of dynamical systems theory, leading him to conclude that, “since the nervous system, body and environment are all constantly changing and simultaneously influencing each other, the true cognitive system is a single unified system embracing all three.” The origins of EC stretch further back: Heidegger (1927/1962) and Merleau-Ponty (1945/2012) are often cited as progenitors of the movement, as is Gibson’s (1979) ecological approach to psychology - all three emphasise the reciprocal relationship between cognition and the environment. Just as cognition can bring about changes in the environment, the environment itself can influence cognition. Gallagher (2017) has argued that EC has historical roots in American Pragmatism: in the work of Peirce it is recognised that artefacts - like Otto’s notebook - can play a role in cognition (ibid, 53); similarly, Dewey anticipates ecological versions of EC in arguing that, when it comes to cognition, the unit of explanation is the organism and its environment (ibid, 54-7) rather than the organism taken alone. The genesis of EC is therefore multifarious and the concept in its various contemporary forms has variegated historical roots.
Currently, EC comes in many forms, not all of which are compatible. Disagreements often focus on the divide between Functionalist versions of EC and views that emphasise the integral role of
3 the body in cognition (see Clark and Kiverstein 2009; Kiverstein 2012). These differences can often be understood as disagreements about what the mark of the cognitive is. On Functionalist versions of EC all that is required for cognition is that the system in question has a certain functional organisation. This more functionally inclined view is developed in the work of Andy Clark (2003; 2008). Enactivist approaches have picked up the threads of Varela et al. (1991), and others, to develop a non-Functionalist, anti-representational version of EC. The central notion of this Enactivist theory is that of autopoiesis, the self-maintenance of living systems, and the emergence of cognition in autopoietic systems (Di Paolo 2005; Thompson 2007; Di Paolo, Buhrmann & Barandiaran 2017; Gallagher 2017). Cognition on this view is thought of as continuous with life, an idea typically referred to as the Life-Mind Continuity Thesis. Although it is not generally referred to as such, this is clearly a criterion of demarcation for cognition. Other ways of spelling out EC include: Richard Menary’s Cognitive Integrationism (see Menary 2007), which emphasises how sociocultural embedding and manipulation of the environment extends cognitive capacities; and Chemero’s (2009) ecological vision of radical embodied cognitive science. I mention these various theories within EC to make clear that there is a variety of views often collected under that rubric. In what follows EC refers to the very general thesis that not all cognitive processing occurs inside the head. On this, all versions of EC agree.
1.2 The central issue of the extended cognition controversy is demarcating cognition
The idea that not all cognitive processing occurs in the head has faced sustained opposition. In this section I survey the four most prominent lines of argument brought against EC. Among these is the mark of the cognitive problem: Adams and Aizawa (2001) argue that things like Otto’s notebook do not bear the mark that distinguishes cognition from everything else. What they think that distinctive mark is will be discussed in section 2.1. This argument puts the onus on advocates of EC to formulate a plausible alternative to Adam and Aizawa’s mark of the cognitive - an alternative on which extended systems like Otto and his notebook count as cognitive. The aim of this section is to explain, following Rowlands (2009) and Walter (2010), that other arguments made against EC can be solved if the mark of the cognitive problem is solved. I set out the most prominent arguments against EC and explain how each objection either assumes a mark of the cognitive (thereby begging the question of what it is for something to be a cognitive process), or can be satisfactorily responded to given a mark of the cognitive on which examples of EC count as cognitive processes.
Cognitive bloat has been discussed in the Philosophy of Mind literature for some time. Block (1978) invites us to imagine the population of China realising the functional organisation of the brain using a system of two-way radios - you can think of individuals as nodes in a neural network and the radios as connections between those nodes. Although the system made up of the population of China could realise the functional organisation of the brain, many philosophers have the intuition that this system cannot be a cognitive one. So this case is taken to be a problematic case for Functionalism, the view that cognitive states and processes are individuated by their functional organisation. Clark and Chalmers (1998) original argument for EC from the parity principle (see page 2) has a strong affinity with Functionalism, with some arguing that EC is in fact entailed by Functionalism (Wheeler 2010; Sprevak 2010). Hence the
4 problem of cognitive bloat arises for EC as well as Functionalism: where should defenders of EC draw the ‘cognitive line’ to rule out absurd cases like the above? Here the demand for a mark of the cognitive is clear: if an independently plausible mark of the cognitive can be given then the problematic cases of bloat can be ruled out or plausibly included. To say that a plausible mark of the cognitive cannot be correct if it includes the system consisting of the entire population of China is question begging - it assumes criteria for determining whether something is or is not cognition. In justifying some set of conditions as the mark of the cognitive you cannot appeal to the cognitive status of systems, states or processes.
The causal-constitutive fallacy (Adams and Aizawa 2001; 2008) is committed when a factor that merely plays a causal role in cognitive processing is said to be (partly) constitutive of that processing. Otto’s notebook, for example, aids his cognition in such a way as to allow certain information to inform his actions just as Inga’s biological memories inform her actions. While the information in the notebook clearly plays a causal role in the production of Otto’s behaviour it is another thing to say that it is constitutive of his cognition. An independent case for factors causally relevant to cognition being constitutive of cognition is required. This argument cuts both ways, however. Adams and Aizawa assume without argument that there is a meaningful distinction to be made between what constitutes a process and factors that causally influence the trajectory of the process, which is questionable (Kirchoff 2015). We have to agree on a notion of constitution for cognitive processes before Adams and Aizawa’s argument can be evaluated, and asking what constitutes cognitive processes is much the same as asking for a mark of the cognitive (cf. Rowlands 2009, 5).
Finally, the differences argument, posed by Rupert (2004) as well as Adams and Aizawa (2001; 2008; 2010), claims that EC cannot ground a unified science of the cognitive. If cognitive science individuates processes by their causal-structure1 (cf. Adams and Aizawa 2001, 51-2) then we cannot count Inga’s biological memory as part of the same category as Otto’s notebook memory (Rupert 2004, 407). The law-like regularities we use to generalise about biological memory will not apply to Otto and his notebook. The problem is broader than this, however: on EC the range of mechanisms included under ‘cognition’ will be so broad as to defy scientific generalisations. Hence Adams and Aizawa’s charge that, “systems consisting of brains coupled with tools would seem to form such a motley collection that they will not form the basis for any significant scientific theorising” (Adams and Aizawa 2001, 63). Rowlands’ response is that EC embraces the differences between Inga and Otto’s memory processes; there is no suggestion that the same scientific generalisations should apply to two such different processes (Rowlands 2009, 3-4). Instead, the claim is that the integration of neural resources and resources in the environment is itself an interesting object of study for cognitive science - regardless of whether this integrated system is functionally similar to, e.g., biological memory. Such integrations allow
1 This is the formulation Adams and Aizawa choose. Prima facie it seems to commit them to a kind of
Micro-Functionalism that would be unable to account for the multiple realizability intuitions discussed in Putnam (1967). Although I think the fine-grained notion of Functionalism their terminology suggests is implausible, I interpret the claim charitably in what follows - I take Adams and Aizawa to be claiming that there are some causal regularities by means of which we can individuate types of cognitive processes. Each token of these types need has exactly the same causal profile.
5 agents to complete tasks they would otherwise be unable to achieve (ibid; see also Menary 2007). Here Rowlands’ notes that defenders of EC must abandon the parity-principle as a basis of EC, in which case an independent case needs to be made for counting these integrated processes as cognitive (Rowlands 2009, 4). Once again, the crux of the issue is whether or not certain systems and processes bear the mark of the cognitive.
One line of reply to the differences argument is to explain that the parity-principle is an argumentative device used to eliminate ‘brain-bias’ in intuitions about which processes are cognitive (Clark 2007, 167). This has lead advocates of EC to reject the parity-principle in favour of the complementarity principle (Sutton 2010). This marks ‘second wave’ EC, where the emphasis is on how brain and environment come together not to replicate what goes in the brain, but in a complementary, integrated way that gives rise to novel cognitive capacities (see Menary 2007; 2010a). The foregoing suggests that the positive case for EC requires an independently plausible mark of the cognitive. Functionalism offers an alternative argument for EC. Wheeler (2010) argues that if we accept the ‘Martian intuition’ - that the familiar mental states of belief, desire and so on could be instantiated in entirely alien ways in the body of a martian - we should conclude that cognitive processes could be extended (see also Walter 2010; Sprevak 2010). The point here is that EC is entailed by the multiple-realisability intuition (the idea that one and the same cognitive process can have distinct physical instantiations, cf. Putnam 1967) that Functionalism accounts for: if systems as different from human brains as the Martian-brain can count as cognitive in virtue of their functional organisation, then systems consisting of brain-environment couplings can count as cognitive in virtue of their functional organisation. While this can be seen as a straightforward motivation for EC (Wheeler 2010), it can be taken as a reductio on Functionalism itself (Sprevak 2010) for reasons similar to the cognitive bloat problems discussed on pages 3-4. Unless Functionalism is supplemented by an additional mark of the cognitive the bloat cases that Sprevak, Block and others find problematic will constitute a reductio on Functionalism - and thus cut off Functionalism as a line of argument for EC (Walter 2010). It seems that both the defence of EC and the positive case for EC depend on the demarcation of the cognitive - that is, constructing a plausible mark of the cognitive. Rowlands (2009; 2010) accepts his own challenge and attempts to develop a mark of the cognitive (See section 2.2 for discussion).
1.3 Research question, aims of the thesis and roadmap
This thesis addresses the question: ‘can coherent dynamical convergence serve as the mark of the cognitive?’. My main aim is to set out desiderata on putative marks of the cognitive and put forward a positive proposal that satisfies these desiderata. This is accomplished in chapter 3. I argue that Trestman’s (2012) notion of coherent dynamical convergence (also called ‘implicit goal-directedness) satisfies all five desiderata, and can therefore serve as the mark of the cognitive. Throughout chapter 3 I pursue the secondary aim of differentiating the view I defend from other theories available in the literature. The desiderata, set out at the beginning of chapter 3, are extracted from the discussion of proposed demarcations of cognition already extant in the literature. I consider representational accounts (Adams and Aizawa 2001; 2008; Rowlands 2009; 2010; and Adams and Garrison 2013) and deflationary approaches that claim no detailed
6 demarcation is needed (Ramsey 2017; Allen 2017; Akagi forthcoming; see also Clark 2010b; Menary 2010a; Rowlands 2003). Chapter 2 surveys these proposals, arguing against the representational approaches and, contra deflationary approaches, argues that there is practical value in demarcating cognition beyond adjudicating the extended cognition debate. Chapter 2 therefore functions as both a literature review and essential set up for chapter 3 which is the most important substantive chapter of this thesis. A secondary aim of this thesis is to address the EC controversy which gave rise to much of the literature I engage with, which is accomplished in the conclusion. There I consider the implications the findings of chapter 3 have for EC.
7 2. Representational and deflationary accounts of cognition
In this chapter my main focus is surveying and evaluating existing proposals about the mark of the cognitive. I discuss representational accounts: the underived content account of cognition (Adams and Aizawa 2001; 2008), Rowlands’ (2009; 2010) information processing account, and the reasons-having account (Adams and Garrison 2013). Thereafter, recent deflationary attitudes to the mark of the cognitive are discussed (Ramsey 2017; Allen 2017; see also Akagi forthcoming). These positions are united by the conviction that there is little practical value to be had in defining cognition. Section 2.4 presents several arguments against the idea that trading in mental representations is the fundamental differentiating feature of cognitive processes. The representational views discussed in this chapter share a commitment to the idea that mental representations with underived intentional content are at the heart of cognition. Section 2.5.1 argues that, contra deflationary views, the mark of the cognitive has practical value beyond settling the EC debate.
2.1 The underived content account
Adams and Aizawa (2001) were the first in the recent EC literature to discuss the mark of the cognitive. Their motivation for demarcating cognition was to refute EC. For Adams and Aizawa, “cognition involves particular kinds of processes involving non-derived representations” (Adams & Aizawa 2001; 52-3). The relevant processes are differentiated into types according to their causal profile (ibid, 51-2): a belief and a desire may have the same content, say that there is beer in the fridge, but the two processes work differently in the sense of having different causal profiles. Underived representations have their content in virtue of processes or structures that are not themselves content-bearing. The idea has strong affinity with Brentano’s aphorism that ‘intentionality is the mark of the mental’. According to Adams and Aizawa, putative cases of EC (e.g. Otto’s notebook from Clark & Chalmers 1998) do not bear this mark, and therefore are not examples of extended cognition processes.
Adams and Aizawa (2001) cite Fodor’s (1987) asymmetric causal-dependency theory of content as an example of how this might work: tokenings of my HORSE concept have horses as their content just in case there is a law-like connection between tokenings of HORSE and horses, and any law-like connection between non-horses and tokenings of HORSE depend asymmetrically on the existence of the law-like connection between tokenings of HORSE and horses. The contents of beliefs, desires and other propositional attitudes is built up compositionally from representations like HORSE that have their content in virtue of a certain asymmetric causal-dependency. The causal-dependency itself involves no content-bearing structures or processes, so the content of the representation is not derived from any pre-existing content bearing structures.
The sense in which linguistic expressions have meanings (or content) is not underived in this way: sentences derive their meaning (contents) from the content-bearing thoughts they are conventionally associated with (Adams and Aizawa 2001, 48). Language can only represent the world because of its connection to thoughts which represent the world. Thoughts exhibit no such
8 dependence. The pen, paper, and system of rules for manipulating external symbols I use in mathematical calculations are not bearers of underived content, and therefore are at best resources or aids exploited by a neural process involving representations with underived contents. My use of the external tools is presumably mediated by some such genuine cognitive processes. Likewise, for Otto’s notebook. Perhaps unsurprisingly, Fodor (2009) follows the same line of argument against EC as Adams and Aizawa - extracranial factors are ruled out of cognition as they are not bearers of underived content.
2.2 Rowlands’ information processing account
Rowlands’ goal in constructing a mark of the cognitive is to show that even on the most traditional, individualist, representationalist conception of the cognitive, cases of extended cognition are still counted. As he says:
I am going to use the criterion of cognition to argue for the thesis of extended and embodied cognition, and so to defend the new science [of cognition] in general. So, in developing the criterion of cognition, I want to give the objector everything he or she could reasonably want. (Rowlands 2010; 114)
So instead of giving necessary conditions on cognitive process he aims to provide sufficient conditions which intra-cranialists like Adams and Aizawa will accept (see Rowlands 2010, 111). Here I discuss Rowland’s sufficient conditions for cognition independently of whether they support the extended cognition thesis (the interested reader should see Rowlands 2009, 12-4). With this aim in mind here are Rowlands’ sufficient conditions:
A process P is a cognitive process if;
(1) P involves information processing - the manipulation and transformation of information bearing structures.
(2) This information processing has the proper function of making available either to the subject or to subsequent processing operations information that was, prior to this processing, unavailable.
(3) This information is made available by way of the production, in the subject of P, of a representational state.
(4) P is a process that belongs to the subject of that representational state. (Rowlands 2010, 110-1)
The representational states in (3) are supposed to have underived content in Adams and Aizawa’s sense explained at the start of this chapter (Rowlands 2009, 9-10; 2010, 115-7). These conditions are extracted from cognitive scientific theories in the individualist, representationalist tradition of Chomsky and Fodor. Taking Marr’s theory of vision as paradigmatic of that tradition in cognitive science, we see that all of Rowlands’ conditions are satisfied (Rowlands 2010; 120-1). Visual perception, according to Marr’s theory, begins with the transformation of a retinal image - an information bearing structure - into a 3D object representation, satisfying (1). This process of transformation makes previously unavailable
9 information available for further processing (belief formation, inference, etc.), satisfying condition (2). Condition (3) is also satisfied as information is made available for, e.g. belief formation, via the production of a representational state. Hence Rowlands takes his conditions on the cognitive to be extracted from cognitive scientific theory. His next argumentative move is to show that these criteria are satisfied by processes distributed across brain and environment, but here I want to consider the plausibility of Rowlands’ mark of the cognitive itself - not whether it offers grounds for EC. Condition (4) is in place because, “whatever else is true of cognitive processes … such process always belong to someone or something” (Rowlands 2009, 15). Even if processes involving the entire population of China connected via walkie-talkies could satisfy (1)-(3) it is not clear who the subject, or owner, of the processing would be. This Cartesian notion that thought must have a bearer is meant to preempt cognitive bloat-style arguments applied to conditions (1)-(3) - in this regard Rowlands considers the case of a telescope, arguing that the internal processes of the telescope satisfy (1)-(3) (ibid, 14-5).
The most obvious problem with Rowlands’ view is the notion of the subject in condition (4). For his purpose (defending EC) it is fine that he has left the notion vague since he does not believe that internalists like Adams and Aizawa have a better account of subjects to whom cognitive processes belong than EC. Internalists cannot appeal to bodily containment (and a fortiori cranial containment) to explain ownership of cognitive processes. Rowlands (2009, 16-7) argues that bodily containment is a poor criterion for ownership of bodily processes like digestion. It is easy to imagine part of the work of digestion being carried out by an external machine connected appropriately to the body (ibid). If this is correct then there are no principled reasons why cognitive processes (taken as a subset of bodily processes) cannot occur outside of the body and still be owned in the sense required by (4). A better notion of ownership by a subject appeals to the extent to which a process is integrated into the (psychological) life of the subject (ibid, 17). Spelling this idea out further is not important for Rowlands’ argumentative aims, but it is relevant for our project of giving clear demarcation criteria for cognition.
Rowlands’ condition (4) seems hopelessly circular, as Adams and Garrison (2013) point out. The notion of a subject cannot be part of the mark of the cognitive since it is by the operation of cognitive processes that we have subjects at all (ibid, 345). Rowlands is not unaware of this problem, explicitly recognising the circularity (Rowlands 2010, 135-6), for it needs to be cognitive subjects that cognitive processes belong to - not just any object will count as a subject in the required sense. From chapter 6 onwards Rowlands (2010) is an attempt to explain what it is to own a process; if this account is plausible cognitive subjects are just subjects which own processes satisfying (1)-(3). I believe Adams and Garrison have misrepresented the move Rowlands makes here. He is arguing that cognitive processes and cognitive subjects emerge simultaneously, not sequentially, in isolation from one another. No antecedent notion of a cognitive subject is involved. So, contra Adams and Garrison cognitive processes do not give rise to cognitive subjects and Rowlands’ account of cognitive processes does not presuppose cognitive subjects. Rowlands’ account of ownership2 is what is really at issue then, not any
2 Or ‘instantiation’: the terms are used interchangeably by Rowlands: “the real problem is not, I think,
explaining what a cognitive subject is but rather explaining the sense in which a subject can own (or instantiate) its cognitive processes” (Rowlands 2010, 136).
10 apparent circularity in (4) - the question is perhaps best expressed as ‘what are cognitive processes, such that they are owned?’. In this sense the latter half of Rowlands (2010) can be seen as an attempt to explain how we can extract condition (4) from conditions (1)-(3). In section 3.2.5 I discuss the issue of ownership in relation to the mark of the cognitive proposed in chapter 3.
2.3 The reasons-having account
Adams and Garrison (2013) construct a mark of the cognitive based on the idea of having reasons for behaviour; if a system is amenable to explanations in terms of represented goal-states and strategies for reaching those goal-goal-states, then it is cognitive. The relevant reasons for Adams and Garrison are the beliefs and desires of familiar intentional explanation. When I get a beer from the fridge my behaviour can be explained by my desire for a beer, and my belief about the location of beer. In order for a system’s behaviour to be subject to this kind of intentional explanation, reasons must be represented in the organism or system itself. Adams and Garrison call these ‘system centered reasons’ (ibid, 347). These are distinguished from evolutionary reasons that explain why, e.g., plants grow towards sunlight. These kinds of evolutionary reasons belong to a species rather than an individual organism (see Dennett 2017). Since Adams and Garrison require the explicit representation of reasons, you could say that, contra Dennett, they require competence with comprehension. There is a reason why human cells have the properties that they do, but the system of my body does not have its own reason for its cells being that way. I, on the other hand, have reasons unique to me for doing all manner of things - reasons that do not apply to other humans. I go to the shops based on the contents of a specific fridge, and a specific set of cupboards. I visit London when I can in order to see a specific set of people. Adams and Garrison express the view like this:
evolution selects for a reasoning mechanism within the individual organism ... and the having of this mechanism constitutes the having of cognitive reasons for this animal’s behavior. We have perfectly good names for these mechanisms: beliefs, desires, plans, intentions, and reasons (ibid, 348)
The reason I associate Adams and Garrison’s view so strongly with the underived content view of Adams and Aizawa (2001; 2008) discussed in section 2.1 is that they cash out ownership of reasons by a system in explicitly representational terms: “in the types of reasons explanations we are featuring, the goal of doing B and the strategy for accomplishing B by doing A are represented within the system” (Adams and Garrison 2013, 347: emphasis added). The reasons-having account of the cognitive can thus be seen as a refinement of Adams and Aizawa’s earlier view: instead of saying any process with underived representational content is cognitive the view now states that any process which has an explanation in terms of system centered reasons is cognitive. Representations are introduced in the definition of system centered reasons. Adams and Garrison do not explicitly state that these reasons have to have underived content. Given Adams continuing opposition to EC, however, it seems we cannot interpret the reasons-having view in such a way as to allow the content of system centred reasons to be derived. To allow system centred reasons with derived content would make Otto
11 and his notebook a cognitive system since the words in the notebook are a paradigmatic case of representations with derived content (cf. Adams and Aizawa 2001). Adams would resist this conclusion, so I interpret Adams and Garrison as being committed to system centred reasons having underived content.
The views differ in the range of things they apply to. The original view from Adams and Aizawa (2001) applies to processes and states: rememberings, beliefs, desires and so on, whereas Adams and Garrison (2013) seem to have whole systems in mind. The former view is perhaps the more basic as the whole machinery of reason-having Adams and Garrison (ibid, 348) appeal to seems to presuppose a notion of representations with underived content. Like the reasons-having account, Rowlands’ information processing account explicitly takes the content of the representations involved to be underived in Adams and Aizawa’s sense (see Rowlands 2010, 115-6). In the next section all of these views are argued against together, given their shared commitment to this fundamental thesis.
2.4 Objections to representational approaches
I will develop four lines of argument against representational accounts of cognitive. Rather than arguing against the positing of mental representations in cognitive science, I argue against representational criteria for demarcating cognition.
Discoverability and difference-making
That a certain system is cognitive is something that we have to discover by looking at the system. It is unclear how we could empirically discover that the processes or states of a certain system have underived content in the sense required by Adams, Aizawa and Garrison. My objection here is that representational views of the cognitive presuppose a position in the debate about the causal efficacy of mental content. If they do not suppose that content is causally efficacious it is hard to see how we could identify cognitive systems, states or processes on the representational view. When we are discriminating the cognitive we have to look for differences, so underived content must be a difference maker if it is to be the mark of the cognitive. Given externalism about mental content (Putnam 1975; Burge 1979) many philosophers of mind have argued that content cannot be causally efficacious (e.g. Stich 1978; Fodor 1991; Kim 1982). There are, of course, responses to this in the literature (e.g. Crane and Mellor 1990; Dretske 1988; Segal & Sober 1991). Settling this debate would be an entire book in itself, so rather than taking a side in this debate I will simply point out that the Adams-Aizawa-Garrison view of the cognitive either, a) helps itself to the assumption that mental content is causally efficacious, or b) leaves it open that the cognitive is demarcated by undetectable epiphenomena. Menary (2010b) makes a similar point: in cognitive scientific theory, he claims, the distinction between derived and underived content is not always recognised. The function of cognitive scientific theorising and experimentation, as in in any science, is to discover and describe the causal ‘difference-makers’ for a certain set of phenomenon. If cognitive science can do this without recognising the distinction between derived and underived content then it stands to reason that underived contents do not play a unique, discoverable role in cognition.
12 Underived vs derived content
A perhaps more fundamental problem is that the distinction between derived and underived content is not conceptually clear (Clark 2005; 2010a; 2010b; Hurley 2010). In the first place, no theory of underived content enjoys broad assent in the Philosophy of Mind. So Adams and Aizawa (2001) make a controversial assumption when they appeal to Fodor and Dretske’s theories of mental content as examples of how one could spell out the notion of underived content. Without a plausible theory of this kind the distinction between derived and underived content lacks grounding. Moreover, Hurley and Clark both present examples in which agents have beliefs – a paradigmatically cognitive state – with content that appears to be derived from the content of other content-bearing structures. Hurley’s case is that of beliefs formed on the basis of testimony. When someone else tells you their birthday, or you read about the recent election in a newspaper, you form beliefs the content of which derives from the content of spoken and written words. Similarly, Clark (2010a) argues that the content of mental images of Venn diagrams derives from the content of Venn diagrams qua symbols in the language of mathematics.
Adams and Aizawa’s (2010) response is that Venn diagrams mean what they do by convention, and convention connects the external symbols to a natural meaning - where natural meaning is understood along the lines of Fodor or Dretske’s theories of mental content (see page 7). Mental images of Venn diagrams are also connected to this natural meaning in a way that is not mediated by the external symbols. Perhaps it is necessary to see the external symbols in order to form a concept of Venn diagrams, but once the concept is learnt it stands in the appropriate relation to the natural meaning. This response also applies to Hurley’s cases: although certain spoken or written words caused you to have a belief, the content of that belief is still determined by the relationship between the representations that are the object of belief and a natural meaning – even though this relationship was established by your belief forming processes interacting with someone else’s speech or writing. Again, it is tendacious for Adams and Aizawa to assume that the relationship between representations and natural meanings works along the lines Fodor and Dretske suggest. Given this assumption though, I agree that Adam and Aizawa’s response works (for more on Adams and Aizawa 2010 see Clark 2010b, 85-90). Adams and Aizawa’s strategy is not unproblematic. The general idea behind cases like Clark’s is to take words or symbols that have externally determined satisfaction conditions (cf. Burge 1979; Wittgenstein 1953), and then claim that the content of some thoughts derive from the content of these words or symbols. Adams and Aizawa respond by claiming that by learning the conventional use of the word or symbol certain mental representations come to stand in a relation to some natural meaning such that that natural meaning is their underived content. Now suppose a student looks at a Venn diagram with at least one element in each circle but an empty intersection and says, ‘the two sets represented by this diagram are identical’. If Adams’ and Aizawa are right the student might have a true belief when she says this, even though she says something false. The belief she expresses can be true if she has a warped notion of what Venn diagrams represent, since the conditions of satisfaction are determined internally and individually. At the same time the assertion is clearly false, since the meanings of words and
13 symbols are determined externally by convention. Adams and Aizawa’s strategy for maintaining the derived/underived mental content distinction comes at the price of allowing that there is no connection in principle between what people assert and what they believe. The student in my example in fact has a belief that she, on Adams and Aizawa’s view cannot possible report because the concept she associates with Venn diagrams is so different from the conventional meaning of Venn diagrams.
The gradability of cognition
Allen (2017) argues that giving a strict definition for cognition is undesirable because, inter alia, there are cases where it is unclear whether an organism is cognitive or not. Borderline cases like these studied by cognitive scientists include slime mould that can navigate through mazes and some bacteria that bias the direction of their tumbling so as to move towards nutrition. To support the case that these are indeed borderline cases of cognitive systems Allen points to empirical findings regarding structural and functional similarities between, “molecular binding mechanisms regulating communicative and adaptive capacities of bacteria … [and] ...the neuronal NMDA-receptors central to synaptic long term potentiation in learning and memory” (ibid, 4237). Put crudely, the difference between what bacteria are doing and human cognition is a distinction of scale - not an essential difference in kind. This gives us a reason to look for a mark of the cognitive that can recognise differences in scale - ‘degrees of cognitiveness’. Slime mould can navigate its way through a maze, leaving a trail of slime behind it to mark paths it has already taken. It seems natural to employ the concept of ‘memory’ to explain what the slime mould is doing. What the slime mould does seems to be very similar to what humans do when they leave a trail of paint behind them in a maze to mark paths they have already been down. Attributing cognition to organisms like slime mould implicitly assumes a mark of the cognitive, however, so I do not want to go as far as agreeing with Allen that certain minimally intelligent organisms are cognitive systems - that would beg the question. I want to make the weaker claim that the cases Allen draws our attention to provide empirical motivation for thinking about human cognition and the activity of much simpler life forms as points on a continuum. And importantly, these cases motivate the claim that similar concepts can be applied in explaining human behaviour and the behaviour of much less intelligent life forms (contra the differences argument, page 4-5). Representation does not admit of degrees - a state either represents or it does not. An on/off concept like representation can only capture points on a continuum of intelligent systems, not define the continuum. Having representational states (cf. Adams and Aizawa 2001) might appear at a certain point on the scale, while having reasons for acting (cf. Adams and Garrison 2013) appears at another point. The case I am making here is that systems which intelligently interact with the world to bring about certain outcomes (a category that includes bacteria and mankind) seem to constitute a category of living things distinct from passive plant life or inanimate objects like rocks. Moreover, similar concepts can be deployed in explaining the behaviour of systems of varying complexity within this category.
Another motivation for rejecting on/off notions as demarcating cognition comes from the development of human cognition on evolutionary and ontogenetic timescales. Cognitive
14 capacities develop gradually over both timescales and characteristically do not appear ‘overnight’ in either case. Present day human beings stand on one end of a continuum stretching back to ancestors that were presumably not cognitive in the sense we are today. As with other evolutionarily acquired traits, it does not seem plausible that there is a clear cut-off point where mankind became cognitive - instead, there is a continuum of evolutionary stages which includes borderline cases. The same considerations apply to cognition as it develops during a person’s lifetime. It is not clear how a capacity that trades in representations can be built up on an ontogenetic timescale – a process either involves mental representations or it does not. In order to explain how two people can have the same capacity to produce and consume representations to different extents requires the representationalist to avert to aspects of the capacity beyond the representations produced and consumed. You could reply that historical fuzziness is not the same as present fuzziness: while the process of evolutionary or ontogenetic development of a cognitive capacity is gradual, the capacity itself need not be gradable. We gradually grow into adulthood, for example, but being an adult is an on/off status (e.g. being over 18 years of age). Consider building a house. Construction is gradual and there will be stages where there is clearly no house. There are later stages of construction where the structure is habitable even though not everything in the architect’s plans has been added. The point of this example is that the structure can be used when it is in an intermediary, unfinished state. This is what distinguishes developing cognitive capacities - on either an evolutionary or ontogenetic timescale - from concepts like adulthood. The capacities can be used in their half-made forms. Someone’s adulthood is not there to be used while it is in development. A house, on the other hand, is there to be used while it is under construction. Similarly, evolutionary or ontogenetic progenitors of our present cognitive capacities were there to be used while they developed. Since representation is an on/off relationship other resources are required to explain this.
Representational accounts limit cognitive scientific theorising
Ramsey (2017) has argued that any demarcation of the cognitive that uses the concept of representations will unnecessarily restrict cognitive-scientific theory-crafting. There are several ways of expressing this problem, but I think the clearest is that a representational mark of the cognitive undermines the status of the representational theory of mind as an empirical hypothesis (Ramsey 2017, 4204-5) - instead it is made out to be a conceptual truth. Further, the central role of representations in cognition should be regarded as a working hypothesis in cognitive science, rather than as a fundamental truth about the science’s subject matter. Viewing it as such means we ought to view it, “tentatively with a willingness to abandon in the face of conflicting findings, methodological revisions and promising new alternatives” (ibid, 4002). Being closed off to alternative paradigms risks holding back the progress of a science. Representational accounts of cognition encourage this kind of attitude and might therefore be an impediment to the progress of cognitive science - a more neutral view would make space for existing representational theories of cognitive capacities as well as alternative non-representational explanations. Alternatives can then be judged empirically against experimental finding, rather than against a conceptual condition on cognitive scientific theorising.
15 Note that Ramsey is not explicitly targeting the Adams-Aizawa-Garrison view I have been discussing in this chapter. He is instead addressing assumptions about cognition that are prevalent within cognitive science. Applying Ramsey’s considerations on representational accounts of cognition to the Adams, Aizawa and Garrison view: the problem is not that they are defining cognition in a way that carries problematic theoretical commitments; the problem is that their mark of the cognitive has theoretical commitments at all (ibid, 4208). The Adams-Aizawa-Garrison view, like the prevalent attitude in cognitive science that Ramsey targets, makes it a the representational theory of mind a conceptual truth rather than an empirical hypothesis. In turn this puts anti-scientific constraints on cognitive-scientific theories: namely, they must appeal to representational mechanisms on pain of failing to be theories about cognition.
2.5 Deflationary approaches to cognition
Ramsey and Allen are united in claiming that the mark of the cognitive is not an issue that needs to be decided, at least from the perspective of cognitive science. While it might have interesting metaphysical consequences for philosophers of mind, it will have little bearing on cognitive science. Allen believes his considerations about empirically discovered borderline cases of intelligent behaviour, “call into question the utility of seeking [a mark of the cognitive] when so much remains to be learned about the variety and complexity of intelligent systems” (Allen 2017, 4243). Likewise, Ramsey concludes that, “[cognitive science] should be demarcated by the kinds of questions we are trying to answer, not by the sort of answer that is on offer” (Ramsey 2017, 4208). What is wanted is an open-ended notion of the cognitive free from theoretical commitments. The best way to get such a notion for Allen and Ramsey is for us to not think too much about it and get on with the work of cognitive science. I label these views of the mark of the cognitive ‘deflationary accounts’, although ‘accounts’ may be a misnomer here. (Akagi forthcoming, recommends an approach to the mark of the cognitive that aims at ‘ecumenical extensional adequacy’ which is similar in spirit to the requirement of theoretical neutrality demanded by Allen and Ramsey.)
Although both Allen and Ramsey want to remain agnostic about the mark of the cognitive, Allen, as quoted in the previous paragraph, mentions ‘intelligent systems’. It seems that here he has a very minimal notion of what cognitive science studies, one which is based on the notion of intelligence (which he may also think does not need defining). Ramsey also gives a soft endorsement to Rupert’s (2009) systems approach to defining cognition. Roughly, Rupert says that processes are cognitive just in case they are part of a mechanism that contributed to the production of cognitive phenomena (e.g. belief, desire, memory, problem solving etc.). Here there is a kind of ostensive definition of the cognitive that is used to form the questions Ramsey believes demarcate the subject matter of cognitive science. The idea of a deflationary or ostensive definition of the cognitive is not a new one, however.
Rowlands (2003), in work preceding that discussed in section 1.2, has given a ‘soft’ definition of the cognitive as follows:
16 … it does seem fairly clear that the notion of a cognitive process is defined, in part, in terms of the notion of a cognitive task. A cognitive process is one that plays a fairly central role in allowing a subject to accomplish a cognitive task. And the notion of a cognitive task is defined, in part, by ostension: by pointing to stereotypical examples of cognitive tasks. (Rowlands 2003, 161).
Menary (2010a, 230-1), quoting the same passage, explicitly subscribes to this view of the cognitive. Similarly, Clark (2010b, 93) believes, “we should individuate the cognitive by its characteristic effects, not by its characteristic causes.” His idea is roughly that processes are cognitive if they support intelligent behaviour. This view is also expressed to some extent by Hurley in her discussion of the enabling conditions of cognition (Hurley 2010). Again, Rupert’s systems account is in a similar vein. All of these ideas taken together form a motley assortment that I collect under the label of deflationary or ostensive accounts of the cognitive.
2.5.1 What work is there left for a mark of the cognitive to do?
All of the deflationary views have something in common. Namely, they make reference to what cognitive processes achieve or accomplish. With this result-oriented account of cognition, I believe the deflationists have picked up something essential about cognition, in my view the only essential thing about cognition: cognition is goal-directed. Going by the ostensive method of defining cognitive tasks almost any outcome or result can constitute a cognitive task. The range and variety of cognitive tasks is as broad as you can imagine. Moreover, Sisyphus can push a rock up a hill, but a terribly strong wind could do so as well. Although the same result is achieved there is a striking difference between the two cases. Having a certain result is clearly not enough. The difference is in the way the goal is achieved. In chapter 3 I develop a view of goal-directed behaviour that I believe distinguishes cognitive processes from processes in general.
Given the deflationary views of Ramsey and Allen, what is the utility of a mark of the cognitive? As noted, my project has its roots in the literature on EC - but is adjudicating the debate over EC the only work there is for a mark of the cognitive? Is it just a point of metaphysical interest for philosophers? A mark of the cognitive can, and should, do much more I believe. First, it has to explain the normativity of cognition: cognitive processes are the kinds of things that can succeed or fail. This can be on the very basic scale of attempting to navigate around an obstacle and bumping into it all the way up to getting the wrong answer to a complex calculation. The goal-directed account of cognition I develop in chapter 3 addresses this directly (see section 3.2.4). Second, related to normativity, the mark of the cognitive can provide the basis for a theory of what constitutes a cognitive agent (see section 3.2.5). Rowlands (2009; 2010) emphasises the idea of a system owning a cognitive process; ownership is important not only for individuating distinct cognitive systems but also for attributions of moral responsibility and epistemic credit. Although I do not address these issues directly, chapter 3 includes discussion of how cognitive systems come to be subject to these ‘higher’ dimensions of normative evaluation (see section 3.3.6).
17 A more ambitious project is using a mark of the cognitive to address the Clunky Robot Problem (Gallagher 2017, 21-2) and the problem of cognitive fluency (cf. Dreyfus 1992; 2007). The Clunky Robot Problem occurs when cognitive scientists construct models of linguistic processing, reasoning, theory of mind capacities and so on, but rarely do so in a way that acknowledges the fact that these capacities are part of entire coherent systems. Gallagher imagines departments being assigned to build the head of a robot, others the limbs, the visual system and so on; all the departments work in isolation, creating modules that function very well on their own but are not suitable for being integrated into one coherent system. The Frame Problem (Dreyfus 1992; 2007) is the problem of implementing open-ended responsivity in systems governed by explicit rules. The view I will develop in chapter 3 does not rely on explicit rules so I do not discuss the Frame Problem. There is a related problem to do with cognitive fluency, however. In a sense, the cognitive fluency problem is the reverse of the Clunky Robot Problem: when we look at the finished product, a functioning human being, we find a system that can flexibly and fluently respond to a wide variety of (changing) circumstances - the system knows when to engage or suppress certain modules. Fluent task switching and flexible responsivity to new circumstances require that the system is not a clunky robot in Gallagher’s sense. In chapter 4 I explain how the account of the cognitive I develop in chapter 3 represents a framework in which these problems can be addressed.
2.6 Summary
In this chapter I have presented arguments against a collection of views that put representations at the heart of cognition. The proposals of Adams and Aizawa’s (2001), Rowlands (2009; 2010) and Adams and Garrison (2013) differ in details, but all three approaches are committed to understanding cognition in terms of systems, states and processes that trade in representations with underived intentional content. Four arguments were presented against views that are committed to this thesis. First, it is controversial whether intentional content is causally efficacious; if it cannot be shown that intentional content is causally efficacious then it is mysterious how we could discover that any system is cognitive in the sense required by the representational views of cognition. Second, the notion of underived content itself is not clear, with no theory of underived content commanding broad acceptance; further, states like beliefs - which are paradigmatically cognitive for representational views - seem to sometimes have only derived content. Third, I argued that on/off concepts like representation are not suitable for modelling the development of cognition over evolutionary or ontogenetic timescales - a gradable notion is needed. Fourth, representational accounts of cognition undermine the status of the representational theory of mind and non-representational approaches to cognition as empirical hypotheses - these are made out to be conceptual issues rather empirical ones. Connected to this, representational accounts of cognition put undue restrictions on cognitive scientific theorising, conceptually ruling out certain types of explanation. In the next chapter desiderata for a mark of the cognitive are extracted from these objections and the discussion in chapter 2. The argumentative force of chapter 3 is that the mark of the cognitive I develop - one that takes goal-directedness to be characteristic of cognition - satisfies these desiderata. In section 2.5.1 I argued, contra deflationary accounts of cognition, that an explicit demarcation of cognition had
18 practical value to cognitive science in accounting for cognitive fluency (cf. Dreyfus 1992; 2007) and addressing the Clunky Robot Problem (Gallagher 2017, 21-2).
19 3. Coherent dynamical convergence as the mark of the cognitive
The discussion in chapter 2 suggests some desiderata on proposed demarcations of the cognitive:
1. Gradability. Cognition should be demarcated in a way that allows organisms and systems to be cognitive to a greater or lesser extent. (cf. section 2.4, pages 13-4, and Allen 2017)
2. Neutrality. The mark of the cognitive should not put unwarranted constraints on scientific theorising. (cf. section 2.4, pages 14-5; Ramsey 2017, Allen 2017)
3. Discoverability. The mark of the cognitive should enable us to empirically investigate whether some organism or system is or is not cognitive. (cf. section 2.4, pages 11-2) 4. Ownership. The mark of the cognitive should account for ownership of cognitive
processes. (cf. sections 2.2 and 2.4)
5. Normativity. The mark of the cognitive should make room for cognitive processes to make success and failure conditions. (cf. section 2.5.1, pages 16)
The aim of this chapter is to argue that goal-directedness - on a certain understanding - plausibly distinguishes cognitive processes from everything else and can satisfy the desiderata above. Gradability opens up space for the evolutionary development of cognition; and provides flexibility enough for us to recognise the ‘variety and complexity of intelligent systems’ (cf. Allen 2017). A theory-neutral mark of the cognitive should avoid the unnecessary limitations that representational accounts put on cognitive scientific theorising (cf. Ramsey 2017; Allen 2017); it should also avoid unnecessary philosophical commitments (e.g. to the causal efficacy of intentional content, see pages 11-2). Discoverability requires that the mark of the cognitive is not epiphenomenal; if the mark of the cognitive appeals to epiphenomena then its scientific and philosophical utility is destroyed. The ownership desiderata demands that a putative mark of the cognitive should allow us to distinguish one cognitive system from another. That is, say which cognitive processes belong to which systems. The normativity of cognition is a concept that the deflationary accounts of cognition discussed in section 2.5 do not deal with. The mark of the cognitive has to allow that cognitive processes can be successful or fail - in section 3.2.4 I distinguish the view I advocate from other views about the normativity of cognition. Section 3.3.6 discusses how we might scale-up the basic notion of normativity I develop in 3.2.4 to tackle higher level dimensions of normative evaluation - attributions of moral responsibility and epistemic credit, for example, Section 3.1 introduces Trestman’s (2012) notion of implicit goal-directedness which is the central concept in the view developed in this chapter. Section 3.2 explains how this notion satisfies my five desiderata. Section 3.3 defends the view against a range of objections.
3.1 Trestman on implicit and explicit goal-directness
The cornerstone of the view developed in this chapter is the concept of coherent dynamical convergence (a term to be defined shortly hereafter), which characterises what it is for an organism to be implicitly goal-directed. Trestman (2012) argues that there is an ambiguity in the
20 intuitive notion of ‘directedness’ between implicit directedness and explicit goal-directedness. Adams and Garrison (2013) appear to have a similar idea in mind when they talk about system-centered reasons in contrast to evolutionary reasons. According to Adams and Garrison I explicitly represent goal-states and strategies for achieving those goals when I take action. In contrast, a plant might, as an evolutionary function, direct its growth towards sunlight, even though it does not represent that state internally. This serves to isolate explicit goal-directedness. Evolutionary function is not what Trestman means by implicit goal-directedness, however. He believes that the goal(s) of an organism’s behaviour can be understood in terms of an organism’s present causal dynamics, without reference to represented goal states or historically determined factors like evolutionary function (Trestman 2012, 208-9). When you watch a dog run as a ball is thrown in the park you can immediately see where the dog’s behaviour is going. Hence the notion of implicit goal-directedness is a dynamic notion, different from both goals understood in terms of evolutionary function and explicitly represented goal-states. This is not to say that we can always immediately, unreflectively read the implicit-goal of an organism’s behaviour. Rather, there are cases where organisms ‘wear their goals on their sleeve’, as in the dog playing fetch example, but this need not always be the case. It might sometimes not be at all obvious where the present causal dynamics of an organism’s behaviour are headed.
The core idea in Trestman’s treatment of implicit goal-directedness is dynamical convergence defined as follows:
A system or process exhibits dynamical convergence if it is flexibly disposed to bring about a single result from any of a range of different starting conditions or despite any range of perturbations, in virtue of the way it is responsive to those starting conditions or perturbations. (Trestman 2012, 210)
Trestman draws our attention to the fact that points of dynamical convergence can ‘jump out at us’ when observing organisms (see, Trestman 2012, 217) in the sense that we can read where an organism’s behaviour is going off of its present behavioural dynamics. Again, this is not to say we can always see these patterns as easily as we can when we watch a dog playing fetch in the park. The important point is that patterns of dynamical convergence can jump out at us when observing the dog or Sisyphus pushing a boulder up a hill, but no such points jump out at us when we see the ball sitting idly on the ground or the boulder tumbling down the hill. The latter two processes clearly do not respond flexibly to the environment to maintain their trajectory towards a particular outcome. They are, in a sense, at the mercy of their circumstances. Sisphyus and the dog, on the other hand, do not passively suffer the circumstances they find themselves in. They can interact with those circumstances to maintain their trajectory towards a certain outcome. The word ‘responsive’ in Trestman’s definition seems problematic when using this notion to demarcate cognition. I think the notion is unproblematic as long as responsiveness is understood as counterfactual sensitivity. Likewise, ‘flexibly disposed’ should be understood in terms of counterfactual sensitivity. My strategy then is to define responsivity in terms of dispositions to interact with the environment, where these interactive dispositions are understood counterfactually. It is unclear whether Trestman himself would
