Cognitive Systems Research 8 (2007) 15–27 www.elsevier.com/locate/cogsys Teleonomic functions and intrinsic intentionality: Dretske’s theory as a test case Action editor: Mark Bickhard Itay Shani Department of Philosophy, School of Social Sciences, University of the Witwatersrand, Johannesburg, Private Bag 3, Wits 2050, South Africa Received 19 May 2006; accepted 10 June 2006 Available online 24 August 2006 Abstract Fred Dretske’s theory of indicatory functions [Dretske, F. (1988). Explaining behavior: reasons in a world of causes. Cambridge, MA: MIT/Bradford; Dretske, F. (1994). A recipe for thought. Originally published as ‘‘If You Can’t Make One, You Don’t Know How It Works.’’ In P. French, T. Uehling, & H. Wettstein (eds.), Midwest studies in philosophy: Vol. 19. Reprinted in D. J. Chalmers (2002) (pp. 468–482).] is undoubtedly one of the more ambitious attempts to articulate a sound naturalistic foundation for an adequate theory of intentional content. In what follows I argue that, contrary to Dretske’s explicit intentions, his theory fails a crucial adequacy test – that of accounting for mental content as a system-intrinsic property. Once examined in light of the first-person perspective of an embodied psychological agent, I argue, it becomes clear that neither ‘indication’, nor ‘function’, as used by Dretske, can be consistently applied. Dretske’s theory of indicatory functions is, thus, doubly incoherent. It is then argued that the problems identified here stretch far beyond Dretske’s specific theory – covering the better part of contemporary attempts to naturalize content. I conclude by suggesting that these general problems of representation, exemplified so vividly in Dretske’s theory, also testify to the inadequacy of the quest to reduce teleological phenomena (function and purpose) to predominantly mechanistic variables.  2006 Elsevier B.V. All rights reserved. Keywords: Dretske; Epiphenomenalism; Function; Indication; Intrinsic intentionality; Self-organization; Teleology; Teleonomy Contemporary research on mental content is dominated by the persuasion that an adequate account of intentionality, that property of mental states whereby they represent conditions (objects, properties, events, processes, places, and situations – both real and imaginary) external to themselves, ought to be naturalistic. Call this the naturalistic criterion.1 Most workers in the field (though by no means all of them) adhere to an additional adequacy criterion: An appropriate naturalistic explanation of mental content is one that explains the intentionality of a mental state as an intrinsic property of the system in which it is embedded. In pain of explanatory regress, the idea goes, mental states cannot derive their intentionality from some external source (an interpreter, programmer, observer, and so forth); they must be intentional on their own right. Call this the intrinsicality criterion.2 E-mail address: shanii@social.wits.ac.za For our present purpose, it is enough if we understand the naturalistic criterion as requiring that intentionality be explained as an integral part of the natural world, without invoking entities or processes that cannot be so integrated. For an overview of naturalism emphasizing its integrative aspect see Hooker (1995); for a comprehensive overview of naturalistic approaches in epistemology and philosophy of science see Kitcher (1992). 2 On the distinction between intrinsic and derived intentionality see, for example, Haugeland (1981), Searle (1992, pp. 78–82), and Shani (2005). Adherents of the intrinsicality criterion include, among others, Bickhard (1993), Block (1990), Fodor (in Dennett, 1987, p. 288), Harnad (1990), and Millikan (1989). A notable critic of the intrinsicality criterion is Dennett (1987, 1996). 1. Introduction 1 1389-0417/$ - see front matter  2006 Elsevier B.V. All rights reserved. doi:10.1016/j.cogsys.2006.06.001 16 I. Shani / Cognitive Systems Research 8 (2007) 15–27 In his book Explaining Behavior (1988) Fred Dretske sets himself the ambitious task of articulating a theory of mental content that satisfies both criteria. In addition, Dretske hopes to satisfy another important requirement: that the theory will explain how the contents of mental states are relevant for behavior. Call this the ‘causal relevance criterion’. Dretske’s strategy is to explain the content of mental states in terms of their indicatory functions. The reduction of content to indicatory function is meant to achieve this tripartite goal of satisfying the naturalistic criterion, the intrinsicality criterion and the causal relevance criterion.3 Recently, Dretske’s claim to satisfied the causal relevance criterion has been intensively criticized (Baker, 1991; Bickhard, 2003; Block, 1990; Kim, 1991; Saidel, 2001; Stampe, 1990). The central charge against Dretske is that his theory identifies the properties that determine the contents of mental states with historical properties while mental causation, on the other hand, depends on presently effective properties. This unfortunate feature of the theory, the critics argue, yields the discouraging consequence that the contents of mental states are inert at the time intentional actions are taking place. The ultimate result, then, is that it is not in virtue of their contents that mental states are causally relevant for behavior – the content-constitutive properties of mental states are epiphenomenal. In this paper, I would like to pursue a path far less trotted, however, and to concentrate on Dretske’s failure to satisfy the intrinsicality criterion.4 While the epiphenomenalism charge targets the failure of Dretske’s theory to account for the causal efficacy of content, I will attempt to establish the claim that in addition, and contrary to Dretske’s explicit purpose, his theory falls short of sustaining intrinsic intentionality. Thus, I argue that neither Dretske’s notion of ‘indication’ (see Section 3), nor his notion of ‘function’ (Section 4), can do the work they are expected to do, namely, sustain a notion of mental representation that makes functional, and epistemic, sense from the firstperson perspective of a genuine cognitive agent. On close examination, then, it becomes clear that Dretske’s ‘‘recipe for thought’’ (Dretske, 1994) is, at best, a recipe for derived intentionality, and that it does not account for the possibility of intrinsically functioning, intrinsically informative, mental states. It is then argued (in Section 5.2) that the epiphenomenalism of Dretske’s theory is but a mirror image of this basic failure to model intentionality as a systemintrinsic phenomenon. In addition, I argue (Section 5.1) that the charges advanced against Dretske can be extrapo- 3 As mentioned in the next section, there exists yet another criterion Dretske aims to satisfy – the ‘misrepresentation criterion’. 4 This is not to suggest that the problem remained completely unnoticed, however, Bickhard (2003), for example, pays attention not only to the epiphenomenalism, but also to the intrinsicality problem immanent in Dretske’s theory. My discussion of the intrinsicality criterion and its significance owes much to Bickhard’s work, there and elsewhere. lated so as to apply, with equal force, to the majority of currently existing attempts to naturalize mental content. Dretske’s failure to satisfy the intrinsicality criterion is, then, but a special case of theoretical malfunctioning on a more general scale. More positively, I indicate (in Section 5.3) how these problems might be avoided by taking a novel approach towards the question of representation, in particular by taking representation to be an emergent aspect of dynamic self-governing. The paper concludes (Section 5.4) with the diagnostic suggestion that the failure to satisfy the intrinsicality criterion, as exemplified in Dretske’s theory, is related to a neglect to pay attention to the irreducible role played by self-organization in the construction of biological and mental functions. Far from being coincidental, such neglect, I argue, is shared by many and is motivated by a general metaphysical commitment to a mechanistic picture of reality, a commitment which leaves no room for genuine self-organization and self-governing, and, ipso facto, no room for genuine function and purpose (biological or mental). It follows that in order to solve the intrinsicality problem, and to secure a place for teleological phenomena within the general order of things, we need to take a fresh look at this deep-seated commitment. 2. Dretske’s theory of indicatory functions The theory offered in chapter 4 of Explaining Behavior is proposed as an improvement over Dretske’s earlier theory of mental content presented in his book Knowledge and the Flow of Information (1981, see his 1986 for an early version of the new account). In his earlier book, Dretske formulated the canonical account of what came to be known as ‘Information Semantics’ (IS). The basic insight of IS ties content individuation to information and information to lawful, or counterfactual supportive, indication. A signal s is said to carry the information that ‘a is F’ if, and only if, it (lawfully, counterfactually) indicates that ‘a is F’. Indication is, in turn, explained by recourse to the notion of reliable correlation: a condition (state, event) C1 indicates another condition C2, if, and only if, C1 is reliably correlated with C2 (typically via a causal connection). However, as early as his 1981 account Dretske was well aware of the limitations of this reductivist program. Reducing semantic content to strictly information theoretic terms yields a conception of meaning too far removed from ordinary mental content, as commonly conceived, to do justice to some of its core characteristics. In particular, Dretske came to realize that IS is not well poised to solve the problem of misrepresentation. Reliable correlation is a factive notion whereas representation is normative. A correlation may obtain or may not obtain, but it cannot obtain properly or improperly. By contrast, one could represent adequately as well as inadequately, one could, for example, say, or think, something that is wrong, provide an inaccurate description, assume a misguided assumption, etc. Consequently, the ability to explain the possibility of misrepresentation, of representational error, must be 17 I. Shani / Cognitive Systems Research 8 (2007) 15–27 considered an additional adequacy criterion for theories of intentionality (call this ‘the misrepresentation criterion’). For all its elegance, a pure information theoretic semantics is ill equipped to handle the problem of misrepresentation.5 Dretske’s solution to this chronic inability to account for misrepresentation within the confines of a pure IS theory, was to incorporate his (information theoretic) account of natural signs within a teleonomic, functional, theory of content. The basic idea is that what confers on a natural sign the status of a fully accredited representation (a ‘belief’, as Dretske puts it) is the fact that it has the function of indicating what it naturally indicates. More precisely, Dretske’s idea can be schematically represented as follows. An inner state C constitutes a belief to the effect that ‘o is F’ if, and only, if (i) C reliably indicates that o is F. (ii) There is a system S, of which C constitutes a functional proper part, such that C’s function in S is to indicate that ‘o is F’. An additional core assumption of this basic model is that what elevates C from a mere natural sign to the rank of a functional indicator is the fact that (a) C has been selected, via a learning process, to play a specific causal role in S. And (b) C was selected to play the role it does because of its indicatory properties (i.e., in virtue of being a natural sign). These basic assumptions, then, are already present in the theory defended in Knowledge and the Flow of Information, but they are developed to full maturity in Explaining Behavior.6 Unlike the earlier theory, Dretske’s latter theory is biologically oriented and its kernel consists of an articulated attempt to explain mental contents as a subspecies genre of biological functions. In a nutshell, Dretske’s proposal consists of the idea that the key towards a successful naturalization of mental content lies in identifying the content of mental states with their indicatory functions. Dretske argues that in order for an inner state to possess intrinsic content, to be causally relevant for behavior, and to satisfy the misrepresentation criterion, it is not enough that it will be indicative. Rather, the state must function as an indicator, and must do so for the system in which it is embedded. His concern with the causal relevance criterion leads Dretske to insist that ‘‘the fact that something has meaning’’ be ‘‘a causally relevant fact about that thing’’ (1988, p. 80). More precisely, Dretske deals with the causal relevance problem by postulating that what a mental state is doing, its current causal role in the system, must be such that it is causally explained, in part, by what the state indicates, i.e., by its semantic value. Suppose, then, that C, a mental state, causes some motor output M. If we ask what is the immediate, triggering, cause of M, says Dretske, our causal explanation will not refer to C’s semantic properties but, rather, to its physical, neurophysiological, properties. What, then, is the causal relevance of C’s semantic character? Dretske answers that its relevance lies in the fact that it partakes in a causal explanation explicating why C is wired in the system’s mental economy in such a way that, under some definite circumstances, it triggers M. In short, C’s semantic profile is causally relevant for behavior in the sense that it is a structuring cause of the C ! M connection: the fact that C indicates F is (partly) responsible for the crystallization of the C ! M triggering causal pattern (see Fig. 1). When this happens, when C gets a handle on the steering wheel of behavior and establishes itself as a triggering cause of M, and does so in virtue of the fact that it indicates F, then, Dretske argues, C acquires the function of indicating F, and thereby represents F. In this way Dretske also hopes to satisfy the intrinsicality criterion and the misrepresentation criterion. First, by acquiring an indicatory function within the system, S, in which it is embedded, C also acquires an intrinsic semantic significance. Second, since C’s function is to indicate F and not, say, G, a tokening of C in the presence of G would qualify as a misrepresentation. At this juncture, there are two things that need to be emphasized about Dretske’s interpretation of the elusive notion of ‘function’. First, according to Dretske, ‘function’ Triggering Cause C M 5 Even Fodor, who disapprove the appeal to biological functions as a means of solving the misrepresentation problem, had to thicken his own information-theoretic account of content with auxiliary assumptions (i.e., his celebrated asymmetrical dependency principle) in order to deal with the problem (see Fodor, 1987, 1990). 6 Dretske’s early solution to the problem of misrepresentation (1981, chap. 8) was abundantly criticized (a notable example is Fodor (1984)), and it was not before long that he himself renounced it (for a sympathetic appraisal, however, see Sterelny 1990, chap. 6). Dretske’s latter account (1988, 1994) differs from the account presented in Knowledge and the Flow of Information in two primary respects: it includes an improved account of the learning process involved in the acquisition of inner states with indicatory functions, and it explicitly identifies indicatory functions as a biological functions. Indication F Structuring Cause Fig. 1. Structuring cause (adapted from Dretske (1988)). 18 I. Shani / Cognitive Systems Research 8 (2007) 15–27 is a diachronic, rather than a synchronic, notion. Namely, what makes an inner state C function as an F indicator for a system S is not the fact that indicating F is C’s present causal role in S; rather, what makes C function as an F indicator is the fact that it has been selected for performing the causal role it currently performs in virtue of the fact that it indicated F. Dretske’s notion of function is thus selection dependent, in a way that resembles other teleonomic theories of content such as Millikan (1989). Second, Dretske concentrates on an ontogenetic, rather than a phylogenetic, modeling of function. That is, he is not concerned with the selection of traits on an evolutionary time scale, but, rather, with the developmental selection characteristic of individual learning processes. In this respect, his theory differs significantly from Millikan’s. It is a widespread phenomenon among plants and animals that a behavior M is triggered by an inner state C that was naturally selected for (in virtue of) indicating an external condition F. In noctuid moths, for example, there exists an evolutionarily established contingency pattern between bat sensing and bat avoidance behavior. Still, Dretske argues, such an evolutionarily shaped (genetically determined) pattern does not confer on C the status of a belief whose content is F. In order to qualify as a belief (hence as a genuine mental state), it must be the case that the fact that C indicates F will actually partake in an individual process of behavioral modification, culminating in the crystallization of C as a cause of M. Such a qualification rules out simple tropistic, and instinctive, behavior but it is perfectly satisfied by a relatively simple process of operant learning.7 Thus, consider a rat that learns to press a bar (M) when, and only when, a certain tone (C) is heard. The correlation between enacting M upon hearing C, and the rewarding experience of feeding, leads to the crystallization of a behavioral pattern in which hearing the tone becomes a cause of, or a switch for, the behavioral output. In this learning process, Dretske argues, C is recruited as a cause of M, because of what it indicates about F, the external condition on which the success of M depends. The learning process selects C as a cause of M, and does so in virtue of the fact that C indicates F. The indicatory profile of C, thus, becomes a structuring cause of the C ! M contingency pattern. C ‘‘gets a hand on the steering wheel’’ (Dre- 7 Dretske’s main reason for denying that natural selection confers on individual indicatory states the status of a belief seems to be this. If reasons (intentional attitudes) are to qualify as causes, they must be the causes of individual actions. But natural selection, Dretske argues, does not explain individual actions; it only explains why certain types of causal dependencies between inner states and behavioral outputs exist (they were selected). Thus, Dretske concludes, ‘‘one must look to systems whose control structures are actually shaped by the kind of dependency relations that exist between internal and external conditions. The places to look for these cases are places where individual learning is occurring, places where internal states acquire control duties or change their effect on motor output as a result of their relation to the circumstances on which the success of this output depends (1988, p. 95).’’ tske, 1988, p. 101) due to the fact that it indicates F; indicating F becomes its function. This, then, is how Dretske accounts for mental content in terms of indicatory functions. In the remaining parts of the paper I shall try to establish the conclusion that, Dretske’s contention notwithstanding, his theory does not satisfy the intrinsicality criterion. I shall argue that neither Dretske’s notion of function, nor his notion of indication, are consistent with a system-intrinsic conception of mental content. Dretske’s theory, then, is doubly incoherent. First, I argue that mental states cannot be intrinsically indicative (or, intrinsically informative) in virtue of being indicative in Dretske’s sense. Second, I argue that, contrary to Dretske’s proposal mental states cannot function as intrinsic indicators of external conditions in virtue of having been selected for so indicating. I call these problems the first, and the second, incoherence problems, respectively. 3. The first incoherence problem: mental states cannot be intrinsically indicative (or informative) in virtue of being Dretske-indicators As we have seen, Dretske accounts for indication in terms of reliable correlation. What makes C informative of F, then, is the fact that it ‘‘locks onto’’ to (encodes) F. The problem with this idea is that it rests on an irreconcilable third-person conception of ‘information’. From the first-person perspective of the psychological agent that owns C, the fact that this inner state corresponds to F is epistemically vacuous, it is simply insufficient to generate knowledge of F. No matter how reliable the correspondence between F and C is, S, the psychological agent, can only access F via C (or some other mental states). Unlike an external observer, S cannot observe the correspondence from both ends and use the fact that it obtains as an independent source of knowledge. In the absence of such knowledge, however, the situation is analogous to having an access to the symbol string ‘‘ -.’’ without knowing that it is the Morse code correspondent of ‘‘N’’: no knowledge of ‘‘N’’ can be miraculously gained merely in virtue of the fact that the correspondence obtains. If the fact that C corresponds to F is epistemically vacuous, however, if it yields no intrinsically available information to the effect that it is F that C stands for, then it cannot be taken as constitutive of C’s being F-informative. An analogous way to state the problem is this. The fact that C encodes F is an extrinsic fact about C in the sense that, in itself, it makes no difference to the internal causal structure of the representation.8 Thus, C would be exactly the same even if, instead of corresponding to F, it were to correspond to F 0 , or even to nothing at all (note that the ‘‘ -.’’ Morse code would be exactly as it is even if it were not paired with the character ‘‘N’’). If C’s indicatory profile 8 The relation between C and F is, as it were, an external relation (for more on the distinction between external and internal relations, and on its significance for theories of content, see Bickhard (2003)). I. Shani / Cognitive Systems Research 8 (2007) 15–27 makes no difference to its internal causal makeup, however, then – since whatever C does it does in virtue of its causal powers – the fact that C possesses this particular indicatory profile can bear no impact on the manner in which it interacts with other mental structures (C 0 , C00 . . .) in S’s cognitive space. But then the difference between C  F and, say, C  F 0 (where ‘  ’ stands for ‘reliably corresponds to’) is not a difference that can be detected elsewhere in the system, and hence not a difference that makes a difference to the ongoing flow of S’s cognitive activity. As before, the upshot is that the mere fact that C corresponds to F does not engender information that can be used, let alone consciously apprehended, from the first-person perspective of the system itself. Now, one may attempt to resist this ‘‘argument from extrinsicality’’ by holding that it misrepresents the idea behind Dretske’s appeal to the notion of a structuring cause. Recall that, according to Dretske, the indicatory properties of mental states are causally relevant for behavior not because they act as efficient, triggering, causes but, rather, because, and insofar as, they become structuring causes of established neuro-motor contingency patterns. What the argument from extrinsicality shows, the rejoinder goes, is that mental states cannot function as triggering causes in virtue of being Dretske-indicators, but since this point is conceded by Dretske right at the outset it can hardly be considered an effective criticism of his position. Moreover, the idea behind the assumption is that F can become a structuring cause within S’s mental economy is, precisely, that it cannot be arbitrarily replaced with other potential correspondents (F 0 , F00 . . .). F being a structuring cause of C ! M implies (a) that this contingency pattern has been selected due to the fact that past activations of it reliably corresponded with the presence of F; and (b) that such past correspondences proved rewarding enough to motivate the selection. Correspondingly, suppose F 0 is a non-nourishing obnoxious substance; then the activation of a C ! M contingency pattern in the presence of F 0 would result in a non-rewarding experience, which will, in turn, activate a feedback learning process selecting against future activation of this contingency pattern. Is it not a mistake, then, to maintain that C’s indicatory profile bears no impact on the manner in which it interacts with other mental states? The first objection, I believe, carries little weight. Regardless of Dretske’s intentions, the question in front of us is whether his theory succeeds in accommodating the first-person perspective of real psychological subjects, and the argument from extrinsicality suggests that it does not. If this failure is due to Dretske’s assumption that the only sense in which the semantic properties of mental states might be causally relevant is by virtue of acting as structuring causes, then so much the worse for the assumption. The second objection is, however, more serious, and it deserves a more thoroughgoing consideration. To repeat, the argument from extrinsicality purports to show that since corresponding to F is an extrinsic fact about C this 19 fact is not reflected in C’s causal makeup and, perforce, cannot, in itself, affect the manner in which C interacts with other mental states within the system’s cognitive network. It then drives at the conclusion that, since, from the firstperson perspective, information must be available in the form of discernible ‘‘news of difference’’ (Bateson, 1979), correspondence relations are insufficient to generate information that could be effective from such a perspective. The second objection challenges one of the premises of the argument, namely, the assumption that the fact that a C  F relation obtains cannot, itself, affect the manner in which C interacts with other mental states within S’s cognitive network. The apparent refutation consists in the fact that the existence of such an indicatory relation is, presumably, a structuring cause of C ! M. If ‘‘C is recruited as a cause of M because of what it indicates about F’’ (Dretske, 1988, p. 101), then, presumably, indicating F does translates into a specific effect on the manner in which C interacts with other components in the network. What the objection fails to notice, however, is the misleading nature of the suggestion that C is selected as a cause of M because of what it indicates about F. Recall that the selection process to which Dretske refers is a learning process, and, as such, a process in which the system itself is an active participant. It is the system’s ability to respond to signals (negative or positive, feedback or feedforward) with novelty – with novel neural configurations and novel dispositions for behavior – which makes learning possible, and which underpins the selective recruitment of some contingency patterns over others. But if the system itself mediates the selection it follows that, whatever it may be, that which causes C to be selected as a cause of M must be something about C that the system can sense and value, something that can motivate a selection. This is even more conspicuous given Dretske’s explicit assumption that selection, via operant learning, for particular causal roles is the key to the solution of the intrinsicality problem (see Section 2): if, as we now see, such a selection operates on variations that the system itself must be able to discern, from its own perspective, then the properties that are directly relevant for selection must be properties that can be so discerned. Yet, this is precisely what cannot be done when it comes to C’s property of being in perfect correspondence to F: nothing in this property per se can motivate internal selection. In order for there to be a selection favoring a systematic activation of C ! M, in correlation with F’s presence, it is not enough that C  F obtains, and that C gets to cause M when F obtains; rather, what makes such positive selection possible is the fact that activating C ! M in F’s presence yields rewarding internal outcomes, outcomes which the system can appreciate (i.e., recognize and evaluate) from its own perspective. To put it otherwise, although F is, in Dretske’s words, a condition ‘‘on which the success of M depends’’ (ibid.), knowledge of the successfulness of the act depends on the availability of internal outcomes delivering the good news. And since selecting M as a typical 20 I. Shani / Cognitive Systems Research 8 (2007) 15–27 behavior in F-infested environments causally depends on such knowledge it follows that it is the internal outcomes that are directly responsible for the selection. The point, then, is that, in order to affect the manner in which C interacts with other components in the network so as to produce successful adjustments (successful learning), correspondence is insufficient. For that, we need correspondence and internal interaction outcomes, and it is the outcomes, and not correspondence per se, which motivate C’s recruitment as a cause of M. Thus, considered on its own merits the fact that C reliably corresponds to F bears no traces which could be discerned elsewhere in the system, it yields no news of difference that make a difference, no information to work with. To recapitulate, what both arguments (the epistemic vacuity and the extrinsicality argument) show is that Dretske’s assumption that reliable indication is constitutive of semantic significance is untenable. If it were, than the mere fact that C is a Dretske-indicator of F would have been sufficient for making it intrinsically informative for S, provided that it is appropriately wired in S’s cognitive makeup. But what the arguments show, is that the one thing that the postulation of a symbol-world correspondence relation does not explain is how any symbol could function, intrinsically, as a representation in virtue of the fact that it stands in such correspondence relations to some external items – no matter how well it is wired in the system. This, then, is the first incoherence problem (for more on the epistemic incoherence of informational encodings see Bickhard, 2000b, 2003; Edelman & Tononi, 2000, chap. 11; Shani, 2005). The gist of the critique advocated here is also hinted in Piaget’s argument against ‘‘copy’’ theories of knowledge (1970, p. 15). 4. The second incoherence problem: mental states cannot function as intrinsic indicators of external conditions in virtue of having been selected for their indicative properties But the problems with Dretske’s proposal run deeper than the commitment to an epistemically untenable notion of indication. I shall now argue that not only is Dretske’s notion of indication unsuitable for the task of accounting for intrinsic intentionality, his notion of function is equally inept. The import of the first incoherence problem is that the notion of ‘‘indication’’ Dretske employs fails the intrinsicality criterion and that for this reason, and on Dretske’s own terms, it is ill suited to serve the purpose of articulating an adequate naturalistic account of mental content. This means that if a theory of content is to employ the term ‘indication’ as an useful explanatory construct (and why not? After all, representation presupposes some form of indication. . .) it must cast into it a different sense. An useful hint as to what such a concept may be can be found in Dretske’s stock example of the rat that learns to press a bar upon hearing a certain tone. When the rat in this experimental setting learns to press a bar in response to a sound stimulus, it learns to associate the stimulus, and the behavioral output, with anticipation of feeding. There is, then, a sense in which the stimulus, once so associated, indicates the prospect of feeding, indicates the likelihood of such an interaction outcome. Interpreting ‘indication’ in this sense is conspicuously opposite to Dretske’s own interpretation. As we have seen, Dretske uses the term to denote reliable covariance, of the sort exemplified by natural signs, where the signal invariably follows the external condition it is said to indicate. Under this interpretation, indication is reactive, consisting of an ‘‘upstream,’’ signified-item-to-sign, arrow. By contrast, on the alternative interpretation suggested here ‘indication’ is essentially proactive (or, enactive); it implies an anticipation of possible future interaction outcomes hence an opposite ‘‘downstream,’’ sign-to-signified-item, arrow.9 No doubt, a commitment to such an alternative notion of indication entails its own questions and problems. In particular, it raises the question how can the rich texture of our ordinary representations of the world around us be constructed out of such seemingly primitive, and thoroughly action-oriented, information (while Akins, 1996; is a skeptic, detailed attempts to provide a positive solution to the problem are suggested in Bickhard (1993) & Shani (in press)). However, insofar as the intrinsicality criterion is concerned, the alternative on offer carries a clear advantage: unlike the property of being in reliable correlation with a given external item, the property of anticipating an interaction outcome is internally accessible, and, as such, it can, and do, affect the internal selection of contingency patterns. Thus, for example, encounters with a nutritious food of type F yield different internal outcomes than, say, encounters with samples of the obnoxious substance G in the simple sense that, when completed, they leave the system in one final state rather than in another – say, X rather than Y. Suppose now that, as in the rat experiment, the possibility of arriving at X is associated with the occurrence of a certain stimulus C in the sense that, when C occurs, it becomes possible for the system to engage itself in an action that yields X (by doing M, for example). Before long, the system may come to anticipate X upon C’s occurrence, and (given X’s desirability) to engage itself in behavior conducive to X. In other words, C may come to proactively indicate X. It is worth noticing that in this case it is one internal state, C, that indicates another internal state, X; and since both states are, as it were, written in the flesh, the problem of accounting for the first-person significance of the indication becomes tractable. It might seem that proactive indications are limited to the system’s interior and, therefore, that they are ill suited 9 For more on the distinction between upstream and downstream signaling see Collier and Hooker (1999). For the distinction between reactive and enactive approaches to the mind see, for example, Newton (2000). I. Shani / Cognitive Systems Research 8 (2007) 15–27 to represent the external world, yet such a judgment is premature. Notice that X, the indicated interaction outcome, depends not only on the system’s actions but also on the environment: F-infested environments will support the possibility of arriving at X, while G-infested environments will fail to do so. In other words X, and the actions capable of yielding X, dynamically presuppose F’s existence: a successful arrival at X, via M, ontologically depends on the environment ‘‘cooperating’’ by manifesting the properties that constitute F (the availability of food), and an X-conducive behavior presupposes such a ‘‘cooperation.’’ There is a sense, then, in which X implicitly categorizes some environments as X-type environments, environments in which this interaction outcome is in fact possible.10 Thus, indicating the availability of an interaction outcome is also, indirectly, an indication that the environment supports this outcome: indicating X is indirectly, and implicitly, an indication of F. However primitive, such categorizations or predications, constructed within the system as structural changes brought about by interactions, provide the system with valuable information about its external surroundings; information that might be false, should the environment fail to ‘‘cooperate,’’ and whose falsity might be detected, should the interaction fail to achieve the expected outcome. With the emergence of complex webs of interconnected indications the power, articulation, and scope of the knowledge at the system’s disposal, and its ability to learn from its own successes and failures, may grow exponentially.11 Finally, note that, unlike Dretske’s passive correlates, the environmental conditions dynamically presupposed by proactive indications are internally (i.e., essentially) related to those indications: X could not be the internal outcome that it is if it were it not for F.12 Thus, while attending to the internal, first-person, components of representation, this alternative, proactive, model of indication seems well poised to address the question how representational knowledge relates to the external world. But, although substituting this action-oriented notion of indication for Dretske’s original proposal solves the first incoherence problem, Dretske’s theory of content faces another major obstacle, an obstacle that cannot be 10 For more on implicit indication and dynamic presupposition see Bickhard (2000b, 2003). 11 The idea that knowledge of the external world is literally constructed as structural effects brought about by interactions can be found, in one way or another, in the writings of otherwise diverse thinkers such as Damasio (1999), Edelman and Tononi (2000), Gibson (1979), Maturana and Varela (1980), and Piaget (1954). 12 This may seem unfair to Dretske given that he, too, is presupposing that the availability of F is necessary for C’s being what it is; however, the point is that, on Dretske’s account, C is F-informative because, and only because, of the correspondence between the two, and, as mentioned in Section 3, correspondence per se does not affect the properties manifested by the correspondents. By contrast, on the assumption that information about the external world is constituted by dynamic interactions it follows that the intrinsic properties of an environment are constitutive of the intrinsic properties manifested by the internal states representing that environment. 21 amended merely by inducing this (in my opinion) necessary substitution. For no matter what notion of indication you care to employ, no mental state can function as intrinsic indicator in virtue of satisfying the conditions that, according to Dretske, grant it a functional status. In other words, Dretske’s notion of function is a veritable blind alley on its own merit. As mentioned before, despite the fact that Dretske’s indicatory functions are modeled on a developmental, rather than an evolutionary, time scale the model he offers is teleonomic, or selection-dependent.13 This means, that what confers a functional status on an indicatory state, is not the causal role performed by that state, but the fact that it has been selected (via learning) for playing this role. Yet this selection-based explanation of indicatory functions yields a notorious circularity problem: as we shall see shortly, in order to be selected as an F-indicator, C must first function as an F-indicator; so, rather than constituting its status as an intrinsic functional indicator, C’s selection as an F-indicator presupposes such intrinsic functioning. Consider, again, Dretske’s example of the rat that learns to press a bar (to do M) upon hearing a certain tone (upon undergoing a perceptual state C). On Dretske’s account, C acquires the function of indicating F (the availability of food) only after the C ! M contingency pattern has been solidified. Yet, this gets things backwards. Recall that when a rat learns to press a bar in response to a sound stimulus it learns to associate the stimulus, and the possible behavioral output, with anticipation of feeding, an interaction outcome towards which, under normal conditions, the animal is motivated. Without the anticipation, there is no basis for motivational arousal, and without such motivational and emotional factors there is no basis for appraisals of failure or success, hence, no basis for learning.14 But if hearing the sound invokes anticipation of feeding, which in turn exerts control on the rat’s behavior, then it transpires that C already functions as an X-indicator, and derivatively as F-indicator, prior to, and in a way that is presupposed by, the successful termination of the learning process. Thus, C’s selection as an F-indicator presupposes a selection-independent notion of functional (functionally useful) indication. To reassure ourselves of the validity of this claim, I suggest we examine it in the light of two key parameters: causal efficacy and normativity. 13 Teleonomic theories of content are known by other names too, e.g., ‘teleological’, ‘etiological’, ‘historical’ or ‘proper function’ theories. My choice of the term ‘teleonomic’ is explained in the concluding section. 14 In recent years there is a growing acknowledgment of the essential role played by motivational and emotional factors in learning and other cognitive processes see, for example, Bickhard (2000c), Christensen and Hooker (2001), Damasio (1994), Edelman and Tononi (2000), Faw (2000), Mook (1996), Montague, Dayan, Person, and Sejnowski (1995). 22 I. Shani / Cognitive Systems Research 8 (2007) 15–27 • Causal efficacy: To begin, note that all the causal tasks performed by C in the post-selection period are essentially in place prior to the establishment of a standardized C ! M connection. C’s selection for the task of indicating whatever it is that it indicates is but a stamp of approval on a successful performance that takes place before the learning process achieves its closure, yet the performance itself remains essentially unaltered. But if performing all the causal tasks (including all the causally indicative tasks) of the post-selection period at the preselection phase is not enough to confer on C the status of a functional indicator it seems inevitable to conclude that the functional relations Dretske hypothesizes are epiphenomenal. • Normativity: Invoking functions in attempts to explain representation is a popular move among naturalists largely because it carries a promise of accounting for the normative dimension of representation, including, in particular, the possibility of misrepresentation. As we saw in Section 2, Dretske is no exception. Yet the claim that selection constitutes functionality, hence normativity, is rather dubious. Even prior to its selection as an invariant cause of M, there is a clear sense in which C’s causing M, thereby leading to interactions bent on yielding X, is good for S (the system): sure enough, consuming nutrients is essential for S’s survival and healthy functioning. By indicating that X is likely to be realized (hence indicating that F is about to obtain), and that M is likely to lead to such a realization, C contributes to S’s well being. Such a contribution carries normative significance for S, since in order to maintain its viability, its survival and ongoing self-maintenance, the system must make sure that the conditions on which its viability depends continue to hold, and whatever contributes to the satisfaction of those conditions is intrinsically good for the system. Nor will it do to maintain that understanding functions, and functional normativity, in terms of contribution to the maintenance of an organic whole illuminates only pragmatic aspects of normativity but that it fails to shed light on the alethic (truth-related) aspects which preoccupy Dretske. Rather, this selection-independent notion of function relates directly to the problem of misrepresentation. Recall that, on the proactive model, representational error occurs when the environment falls short of supporting the interaction possibilities indicated by a given intentional state, that is, when it does not manifest the properties that sustain the success conditions of the anticipated interaction outcome. Representational error, then, is, first and foremost, a defiance of expectation; it constitutes a hindrance to the system’s Sisyphic effort to make its way in the world; it constitutes a functional failure. Thus, misrepresentation is a specific form of malfunctioning in as much as the prospects indicated by a misrepresenting intentional state are ungrounded, making that state ill-equipped to contribute to the system’s collective effort to maintain itself, and to orient itself in its social and natural environments. Moreover, it is precisely because it constitutes a malfunction, because it defies expectations and hinders prospective self-regulation, that error can be detected via negative feedback, and that corrective measures utilizing such feedback might ensue. Finally, note that nothing in this explanation requires selection to account for the emergence of misrepresentation; representational error is totally constituted in current system states and in current dynamic patterns of system–environment interactions. Thus, our alternative account of functions, and of functional indication, enables us to concur with Dretske’s claim that representation is a specific form of function, an emergent sub species genera of, naturalistically explicable, biological phenomena. At the same time, it denies Dretske’s contention that indicatory functions are selection-dependent, thereby avoiding epiphenomenalism and the circularity inherent in the idea that selection constitutes functional relations. Taking a broader look at the problem in front of us – the incoherence of selection-based explanations of indicatory functions – we may note that the same lesson applies, mutatis mutandis, to selection-based theories of function in general, and, in particular, to the popular view that natural selection confers functional status on biological traits. According to selection-based theories, it is only after there has been a selection ‘‘for’’ a trait T that T can be considered functional; selection constitutes functionality, and functional normativity.15 But this obscures the fact that, in order to be selected, T must first contribute to the adaptability, or ecological competence, of certain individuals such that these individuals will perform, on average, better than other conspecifics and, as a result, will have an improved fitness rate. Such contribution to individual ecological performance is presupposed by selection and therefore, on pain of regress, cannot be explained as its outcome; and yet, it is functional and normative par excellence. First, note that, in this case too, all the causal capacities that the selection-based explanation ascribes to T in the post-selection period are already at play at this pre-selection stage. Consider an example. Some marine invertebrates (e.g., rotifers, barnacles, and bryozoans) developed an irreversible adaptive response to predation. Usually, that is, under normal conditions, they take the form of a typical morph but when exposed to nearby predators they can rapidly, and irreversibly, change their appearance into an alternative, atypical, morph, or to produce progenies with such non-standard appearance. The predator-induced morph lowers mortality rate in predator infested environments and thus has a higher fitness in those environments, but in predator free environments it is has lower fitness (Dukas, 1998). According to selection-based theories such as Millikan’s (1984, 1989) the structures responsible for these 15 The distinction between selection of a trait and selection for (i.e., in virtue) of a trait is due to Sober (1984). I. Shani / Cognitive Systems Research 8 (2007) 15–27 defense strategies in the tiny marine creatures has the function of protecting the creatures because they were selected for doing so. But this obscures the fact that in order to be selected ‘‘for’’ the task, the relevant causal structures must have already been at work – serving their owners by reducing mortality rate in predator-infested environments. The moral, then, is that the adaptive performance to be selected is already there in its entirety prior to the culmination of the selection process (first the performance, then the reward. . .). But, if performing all the causal tasks of the post-selection period at the pre-selection period is not enough to confer on T the status of a functional trait, then, as in Dretske’s case, epiphenomenalism seems inevitable (cf. Christensen & Bickhard, 2002a; Saidel, 2001). Second, since in this pre-selection stage T is useful to the organisms in which it is embedded it already carries normative significance for those organisms. Thus, consider the manifestation of, and reaction to, alarm signals such as tail splashing in beavers’ populations, or various vocalizations in vervet monkeys. These signals, and their characteristic modes of usage, were selected because they proved to be useful, ecologically competent, patterns of behavior, contributing to the survival and stable sustenance of individuals and populations. But the usefulness of such signals in their contexts of application constitutes a normative dimension that is, again, selection-independent. There is a clear sense in which it was good for beavers to splash their tail and for vervet monkeys to make their calls even before these behaviors were selected across the populations. It follows that selection presupposes normative relations, and, insofar as selection-based accounts appeal to selection as the putative source of norms, they are inconsistent.16 It transpires, then, that the problems of epiphenomenalism, and of normative inconsistency, that haunt selectionbased theories of biological functions stem from a neglect of a crucial fact about biotic evolution. Natural selection operates on variability in individual (or group) performance between systems that already possess a degree of functional organization, and that are already equipped with inner states capable of making some contribution to the incessantly self-preserving (i.e., functional) causal organization of their owners. To put it otherwise, in order to be a participant in the game of natural selection you have to be able to reproduce, maintain homeostasis, and compete for resources; but a physical system, which must, and can, maintain itself via resource acquisition, self-recuperation and self-reproduction – an autonomous agent – is already a clear exemplar of a functional system.17 16 Various authors have made the claim that selection based theories of function presuppose a more fundamental, selection-independent, notion of function. Examples that may be cited here are Bigelow and Pargetter (1987), Bunge and Mahner (1997, chap. 4), Christensen and Bickhard (2002a, 2002b), McIntosh (2001), and Stotz and Griffiths (2002). 17 For some accounts of autonomous agency see Bickhard (2000a), Christensen and Hooker (2000), Gibson (1994), Kauffman (2000), Smithers (1995), and Ulanowicz (1986). 23 Defenders of selection-based theories of function might respond by arguing that such functional systems are ultimately assembled by the operation of natural selection on simple, non-functional, template replication mechanisms, hence that, in the final analysis, selection does generate functions. Yet, significant developments in the last decades in the study of biological systems as complexly organized dynamical systems put this standard neo-Darwinian dogma to doubt. As is intimated in the works of Eigen (1971), Kauffman (1993, 1995), Margulis (e.g., Margulis & Sagan, 1986), Maturana and Varela (1980), and others, the very emergence of life presupposes holistic self-maintenance of the sort exemplified by collectively autocatalytic macromolecules. The point is that such self-maintaining systems – predating the emergence of the double helix – were already functionally organized and capable of prebiotic evolution. On this view, then, self-organization plays an essential, irreducible, role in the construction of biological order. As Kauffman puts it ‘‘[M]uch of the order in organisms, from the origins of life itself to the stunning order in the development of a newborn child from a fertilized egg, does not reflect selection alone. Instead, much of the order in organisms, I believe, is self-organized and spontaneous. Self-organization mingles with natural selection in barely understood ways to yield the magnificence of our teeming biosphere. We must therefore expand evolutionary theory (2000, p. 2).’’ To conclude, the upshot of the arguments advanced in this section is that selection does not, and cannot, constitute the ultimate explanation of functional organization and functional normativity. But if the emergence of functions, and of functional normativity, cannot be attributed solely to selection processes, then, as a special case, it follows that mental states cannot acquire indicatory functions merely in virtue of having been selected for their indicatory properties. This, then, is the second incoherence problem. 5. Some theoretical implications It is time to take stock. The discussion that follows examines the conclusions that can be derived from the arguments advanced in the last two sections. In addition, a special emphasis is given to some broader implications that might be drawn, using extrapolation and further analysis, from these more direct conclusions and that seem to be theoretically significant for the general project of explaining representational phenomena. These include insights into existent pitfalls, and hints at the prospects for a brighter future. 5.1. Intrinsicality and the incoherence arguments: extrapolating beyond Dretske’s theory I have argued that Dretske’s attempt to satisfy the intrinsicality criterion fails. The reasons for the failure, 24 I. Shani / Cognitive Systems Research 8 (2007) 15–27 however, go well beyond Dretske’s own theory of indicatory functions. The moral of the second incoherence argument is that there is a problem with the very idea that functions, and functional normativity, are selection-dependent. If the argument is cogent, then selection-based theories of function presuppose a more fundamental, selection-independent, notion of function, and the consequences for the thriving industry of explaining functions, and, a fortiori, representational functions, in terms of selection are dire. As for the first incoherence argument, the implications are even broader. If the argument is sound, it not only shows that information semantics is ill equipped to deal with the problem of intrinsic intentionality, it also casts a shadow on the entire enterprise of accounting for content in terms of correspondence, or encoding, relations, thereby taking to task almost all of the contemporary naturalistic semantics. 5.2. Intrinsicality and causal efficacy Moreover, we are now in a position to observe the connection between the popular charge against Dretske to the effect that his theory implies epiphenomenalism, and our own findings. From this vantage point of view, the epiphenomenalism of Dretske’s theory is a mirror image of the basic failure to explain intentionality as a system-intrinsic phenomenon. The link between intrinsicality and epiphenomenalism is intuitive enough. For a representation to be system-intrinsic is for it to be capable of functioning as a representation for the system in which it is embedded. That is to say, if C is a representation embedded in a system S, and if C’s content is P, it must be the case that C can function as a representation for S in virtue of its content; C’s content must be functionally available to S and it must be capable of making a difference, a causal difference, to S’s thought and action. A theory whose prescriptions for content individuation yield contents that can be neither accessed nor used by their owners is a theory whose content assignments are necessarily causally inert, hence epiphenomenal. Conversely, a theory whose prescriptions for content individuation yield causally inert contents necessarily fails to explain intentionality as a system-intrinsic property. 5.3. A hint on how to approach a solution to the incoherence tangles The popularity of teleonomic, selection-based accounts of function and representation stems, to a large extent, from the fact that many scientifically minded thinkers (e.g., Dawkins, 1976; Dennett, 1987; Pinker, 1997) believe it to be the only respectful way whereby the question of purposeful behavior may be approached. Similarly, the popularity of encoding-based accounts of content is rooted in the persuasion that this is the only way whereby the question of representation may be approached. But if the arguments presented here are along the right track, we had better look for alternatives. A not too careful reading between the lines of this critical essay reveals that it already contains the seeds of a possible alternative. For it offers, in passing (a) thinking of function in terms of making systematic contribution to the maintenance of an organic whole instead of in terms of selective history, and (b) thinking of indication (hence representation) in terms of anticipation of (possible) interaction outcomes rather than in terms of reliable correlation. Nor is this alternative a mere hypothetical program. Proponents of a dynamic systems approach to mental phenomena called ‘interactivism’ (Bickhard, 1993 and elsewhere; Christensen & Hooker, 2000) have developed, in considerable detail, a theoretical account of representational content incorporating these basic insights (along with significant theoretical tenets borrowed from complexity theory, developmental psychology, ecological psychology, pragmatism, phenomenology and more). On a more general scale, it may be mentioned that contemporary cognitive science witnesses a steady growth in the popularity of embodied, and action-oriented, theories of mind manifesting a significant degree of approximation to the ideas defended here (Lakoff & Johnson, 1999; Varela, Thompson, & Rosch, 1991; are but two of the more familiar examples). Although it is my conviction that this alternative way of looking at the problem of mental representation possesses a decisive advantage over more popular theories such as Dretske’s in that it offers a coherent solution to the intrinsicality problem, the paper’s modest aim was simply to show how, and why, Dretske’s own solution fails. Therefore, within the confines of the present discussion, I refrained from making a fully systematic attempt to explain, and defend, the interactive program. 5.4. Intrinsicality and self-organization: or, why attempts to reduce teleology to mechanistic causation are bound to fail Having concentrated on the difficulties enfolded in Dretske’s position, I would like to conclude with a general observation regarding the connection between Dretske’s failure to respect the intrinsicality criterion and his neglect to take note of the constitutive role played by self-organizing dynamics in the construction of functions, and of functional indications. In discussing the two incoherence problems, I argued that the failure to model either indication, or function, as system-intrinsic phenomena stems from a neglect of their inherently dynamic, self-organizing, character. Functions, I argued, ought to be understood as contributions to the self-organizing dynamics of an autonomous organic whole, and indications ought to be explained as functions of a special sort whose contribution to autonomy consists of environmentally sensitive, anticipatory, action-guidance. It is I. Shani / Cognitive Systems Research 8 (2007) 15–27 with this picture in mind, I concluded, that we may hope to overcome the difficulties faced by the traditional approach to function and representation, of whom Dretske is, without a doubt, a particularly well-spoken representative. In the remaining pages I shall argue that there is a reason, a deeply seated metaphysical reason, behind the reluctance of the traditional approach to take advantage of self-organization in attempting to account for biological, and mental, phenomena. As before, I argue that Dretske’s theory can be used as a telling example. As we shall see shortly, Dretske’s appeal to a selectiondependent account of function, and of functional indication, reflects a deep theoretical commitment to the idea that an adequate naturalistic explanation of teleological phenomena must conform to a mechanistic outlook of reality. A consistent adoption of the mechanistic image of reality, however, leaves no room for genuine self-organization and, as a result, no room for explaining functions in general, and representational functions in particular, as system-intrinsic phenomena. Indeed, the commitment to the idea that the only naturalistically acceptable kinds of explanations are explanations that refer to mechanistic modes of production and becoming (or, at the very least, that presuppose strictly mechanical processes at the relevant level of ‘‘implementation,’’ or ‘‘realization’’) has the inevitable effect that, in the final account, all teleological phenomena are rendered illusionary. Attempts to redeem our teleological intuitions by making telos conform to the mechanistic framework are plenty and here again, as we shall see, Dretske is a loyal representative, but the rift between telos and mechanism is such that these attempts manage to salvage no more than a faint apparition of genuine function and purpose. I therefore propose, that by taking seriously the idea that function and representation ought to be explained in predominantly self-organizational terms, we are obliged to reconsider our all too sweeping commitment to the mechanistic view of the world and to take a fresh look at the role of telos in nature. The theory presented in chapter four of Explaining Behavior constitutes a deliberate attempt to provide a naturalistic foundation for a viable account of mental content. The hub of the theory is the idea that the key towards a successful naturalization lies in identifying mental representation as a kind of biological function, which function is in turn explained by reference to selective history. Theories of content that hinge on this idea are often referred to as ‘teleological’ (e.g., Papineau, 1991), but a closer examination reveals that, for reasons that are far from trivial, the euphemism ‘teleonomic’ is a more appropriate choice. The term ‘teleonomy’ was proposed by Pittendrigh (1958, p. 394) as a substitute for the traditional, more familiar, term ‘teleology.’ Teleology, as commonly conceived, is the study of ends or final causes – the explanation of phenomena by reference to goals, or purposes. As such, it carries with it connotations of the Aristotelian worldview which was repudiated with the advent of modern science. On the mechanistic worldview extracted from classical 25 dynamics there is no place for final causes; and purposive, or seemingly purposive, behavior must ultimately be reduced to efficient, mechanistic, causation. It is often maintained that one of Darwin’s remarkable achievements was that his theory of the evolution of species by way of natural selection made such a reduction feasible. Darwin’s theory, the idea goes, provided the means for explaining the purposive behavior, and design-like organization, found in nature in terms of the mechanisms governing mutation, variation, and selection. The seeming teleology of biological phenomena could now be explained in terms compatible with the mechanistic modes of explanation characteristic of the physical sciences.18 The upshot of such a reductive explanation is that the appearance of purposefulness in nature is exactly that – an appearance, the result of nature’s laborious and opportunistic blind tinkering. The term ‘teleonomy’ – especially as adapted by Monod (1971) and Mayr (1992) – was meant to cover precisely this type of explanation, namely, to account for apparently purposive structures, functions, and behaviors as evolutionary adaptations, which could, in the final analysis, be analyzed into their ultimate mechanistic components. In the words of Richard Dawkins, ‘‘in effect, teleonomy is teleology made respectable by Darwin (1982, p. 294).’’ It is clear that Dretske’s theory is teleonomic in spirit, if not in its letter. As mentioned throughout the paper, the underlying working assumption of Dretske’s account is that what confers a functional status on indicatory states is the fact that they were selected for their indicatory properties. Since such an account essentially reduces intentionality to indicatory function, the implication is that the apparent purposefulness of intentional states is simply the result of nature’s cunning blind tinkering – the same logic underlying Pittendrigh’s introduction of ‘teleonomy’ as a substitute for the debunked term ‘teleology.’ The epistemic incoherence of Dretske’s teleonomic theory of content, and especially the fact that it presupposes an untenable explanation of the emergence of systemintrinsic functions (representational or otherwise), gives grounds for suspecting that the problem might be symptomatic of a more general fault, namely, that it might have to do with the idea that teleology could be reduced to teleonomy. In more than one place, Dretske compares the task of explaining purposeful behavior and design-like organization in nature to that of explaining a work of engineering (1988, pp. 96–97; 1994). The comparison is illuminating. To begin, note that there is an obvious disanalogy between engineered, or otherwise manufactured, artifacts and naturally constructed biological systems: all naturalists believe that only the formers are the products of intelligent design. Nevertheless, advocates of teleonomy insist 18 Ironically, it is in contemporary physics itself that the mechanistic paradigm ultimately breaks down, a fact that seems to have gone beyond the radars of those who espouse its application to biological, and intentional, phenomena. 26 I. Shani / Cognitive Systems Research 8 (2007) 15–27 also on the existence of a manifest analogy: both are made to appear purposeful, both are structured as if they possess a telos. The word ‘made’ is revealing; it implies that the functional organization of the system is dictated by an external agency. Such an external agency might be intelligent, but it need not be so. So long as the end product is secured it matters not whether the designing agency is intelligent or blind, natural or unnatural, final or efficient. Now, clearly an artifact is a system whose organization is shaped from without; but the interesting point is that so is the case with a mechanistically tinkered-together contraption. Indeed, one of the defining characteristics of a mechanistic explanation is that it accounts for the behavior of the system under scrutiny completely in terms of external agencies – the impact of other bodies, the operation of forces, etc. (cf. Bohm, 1957; Prigogine & Stengers, 1984; Rosen, 1991; Ulanowicz, 2000). The bottom line is that a mechanical device – whether naturally formed or artificially contrived – does not partake in the making of its own organization. It transpires, then, that the communality between intelligent design and teleonomy is that both of them presuppose external formation while excluding self-organization. Proponents of mainstream reductionism often assume that the only alternatives to mechanism are vitalism or, worse still, supernatural intelligent design. Yet, the dilemma is fabricated in that it ignores the possibility of a self-organized purposive dynamics. This is not merely a logical quibble. The recent decades have seen a rapid advancement in the study of complex, dynamically non-linear, systems of various levels of manifested complexity, and a remarkable increase in our understanding of the self-organized aspects characterizing their emergence and behavior. Such developments leave a place for more than a shred of optimism regarding the prospects of accounting for function and purpose in primarily self-organizational terms. One of the fascinating features of this newly emerging theoretical approach to the study of function and purpose in nature is that it offers a getaway from the need to choose between the Scylla of intelligent design and the Charybdis of a mechanistic ‘‘natural design.’’ While the intelligent design solution has been amply criticized for implying a supernatural interference in the order of things, it has often been overlooked that a mechanistic solution implies an almost equally unsettling conclusion. For, as mentioned above, on a strictly mechanistic explanatory framework biologically evolved function and purpose are, in the final analysis, mere appearances – they are, as it were, mere as if phenomena. But if function and purpose could be explained by reference to the process dynamics characteristic of open, self-organized, systems, then there might be room for a less suspicious approach towards the prospect of explaining genuinely telelological phenomena as an integral part of nature. Such integration would amount to the reestablishment of telos within the natural order of things, in a way that implies neither supernatural manipulation, nor eliminative reductionism. Needless to say, a more elaborate exploration of the bearings of these recent developments in complex systems research on the potential rehabilitation of teleology deserves a separate treatment. Acknowledgements I thank Amir Horowitz and the audience at the 2005 annual colloquium of the Israeli Philosophical Association in Haifa, where an early draft of the paper has been read. References Akins, A. (1996). Of sensory systems and the ‘‘Aboutness’’ of mental states. The Journal of Philosophy, 22–362X, 337–372. Baker, L. R. (1991). Dretske on the explanatory role of belief. Philosophical Studies, 63, 99–111. Bateson, G. (1979). Mind and nature: A necessary unity. NY: E.P. Dutton. Bickhard, M. H. (1993). Representational content in humans and machines. Journal of Experimental and Theoretical Artificial Intelligence, 5, 285–333. Bickhard, M. H. (2000a). Autonomy, function and representation. Communication and Cognition – Artificial Intelligence, 17(3–4), 111–131, Special issue on: The contribution of artificial life and the sciences of complexity to the understanding of autonomous systems.. Bickhard, M. H. (2000b). Information and representation in autonomous agents. Journal of Cognitive Systems Research, 1(2), 65–75. Bickhard, M. H. (2000c). In: R. D. Ellis & N. Newton (Eds.), Motivation and emotion: An interactive process model. Bickhard, M. H. (2003). Process and emergence: normative function and representation. In J. Seibt (Ed.), Process theories: crossdisciplinary studies in dynamic categories. Dordrecht: Kluwer Academic. Bigelow, J., & Pargetter, R. (1987). Functions. Journal of Philosophy, 84, 181–196. Block, N. (1990). Can the mind change the world? In G. Boolos (Ed.), Meaning and method: Essays in honor of Hilary Putnam (pp. 137–170). Cambridge: Cambridge University Press. Bohm, D. (1957). Causality and chance in modern physics. New York: Harper & Brothers. Bunge, M. A., & Mahner, M. (1997). Foundations of biophilosophy. Berlin: Springer. Christensen, W. D., & Bickhard, M. H. (2002a). The process dynamics of normative function. Monist, 85(1), 3–28. Christensen, W. D., & Bickhard, M. H. (2002b). Function as design versus function as usefulness. Unpublished manuscripts. Christensen, W. D., & Hooker, C. A. (2001). Self-directed agents. In: J. McIntosh (Ed.), Naturalism, evolution and intentionality. Canadian Journal of Philosophy, Special Supplementary Volume. Christensen, W. D., & Hooker, C. A. (2000). An interactivist-constructivist approach to intelligence: Self-directed anticipative learning. Philosophical Psychology, 13(1), 5–45. Collier, J. D., & Hooker, C. A. (1999). Complexly organized dynamical systems. Open Systems and Information Dynamics, 6, 241–302. Damasio, A. R. (1994). Descartes’ error: Emotion, reason and the human brain. NY: Grosset-Putnam. Damasio, A. R. (1999). The feeling of what happens: Body and emotion in the making of consciousness. NY: Harcourt Brace. Dawkins, R. (1976). The selfish gene. New York: Oxford University Press. Dawkins, R. (1982). The extended phenotype: The gene as the unit of selection. Oxford: Freeman. Dennett, D. C. (1987). The intentional stance. Cambridge, MA: MIT. Dennett, D. C. (1996). Kinds of Minds. NY: Basic Books. Dretske, F. (1981). Knowledge and the flow of information. Oxford: Blackwell. I. Shani / Cognitive Systems Research 8 (2007) 15–27 Dretske, F. (1986/93). Misrepresentation. In A. I. Goldman (Ed.), Readings in philosophy and cognitive science. Cambridge, MA: MIT/ Bradford, Reprinted. Dretske, F. (1988). Explaining behavior: Reasons in a world of causes. Cambridge, MA: MIT/Bradford. Dretske, F. (1994). A recipe for thought. Originally published as ‘‘If You Can’t Make One, You Don’t Know How It Works.’’ In: P. French, T. Uehling, & H. Wettstein (Eds.), Midwest studies in philosophy: vol. 19. Reprinted in D. J. Chalmers (2002) (pp. 468–482). Dukas, R. (1998). Evolutionary ecology of learning. In R. Dukas (Ed.), Cognitive ecology: The evolutionary ecology of information processing and decision making. Chicago: University of Chicago Press. Edelman, G. M., & Tononi, G. (2000). Consciousness: How matter becomes imagination. London: Penguin Books. Eigen, M. (1971). Molecular self-organization and the early stages of evolution. Quarterly Reviews of Biophysics, 4(4), 149. Faw, B. (2000). In: R. D. Ellis & N. Newton (Eds.), Consciousness, motivation and emotion: Biopsychological reflections. Fodor, J. A. (1984). Semantics, Wisconsin style. Synthese, 59, 231–250. Fodor, J. A. (1987). Psychosemantics. Cambridge, MA: MIT. Fodor, J. A. (1990). A theory of content and other essays. Cambridge, MA: MIT. Gibson, J. J. (1979). The ecological approach to visual perception. Hillsdale, NJ: LEA. Gibson, E. J. (1994). Has psychology a future? Psychological Science, 5, 69–76. Harnad, S. (1990). The symbol grounding problem. Physica D, 42, 335–346. Haugeland, J. (1981). Semantic engines: An introduction to mind design. In J. Haugeland (Ed.), Mind design. Cambridge, MA: MIT. Hooker, C. A. (1995). Reason, regulation and realism: Toward a naturalistic regulatory system theory of reason. Albany, NY: State University of New York Press. Kauffman, S. A. (1993). Origins of order: Self-organization and selection in evolution. New York: Oxford University Press. Kauffman, S. A. (1995). At home in the universe: The search for the laws of self-organization and complexity. NY: Oxford University Press. Kauffman, S. A. (2000). Investigations. NY: Oxford University Press. Kim, J. (1991). Dretske on how reasons explain behavior. In B. McLaughlin (Ed.), Dretske and his critics (pp. 52–72). Cambridge: Basil Blackwell. Kitcher, P. (1992). The naturalists return. Philosophical Review, 101, 53–114. Lakoff, G., & Johnson, M. (1999). Philosophy in the flesh: The embodied mind and its challenge to western thought. NY: Basic Books. Margulis, L., & Sagan, D. (1986). Microcosmos. New York: Summit. Maturana, H., & Varela, F. (1980). Autopoiesis and cognition. Dordrecht: Reidel. Mayr, E. (1992). One long argument: Charles Darwin and the genesis of modern evolutionary thought. London: Allen Lane. McIntosh, J. S. (2001). Function, malfunction and intentional explanation. Unpublished manuscript. Millikan, R. G. (1984). Language, thought and other biological categories. Cambridge, MA: MIT/Bradford. Millikan, R. G. (1989). Biosemantics. Journal of Philosophy, 86(6), 281–297, Reprinted in R. G. Millikan 1993. 27 Monod, J. (1971). Chance and necessity: An essay on the natural philosophy of modern biology. Trans. from French: A. Wainhouse. New York: Alfred A. Knopf. Montague, P. R., Dayan, C., Person, P., & Sejnowski, T. J. (1995). Bee foraging in uncertain environments using preditive Hebbian learning. Nature, 377, 725–728. Mook, D. G. (1996). Motivation: The organization of action (2nd ed.). New York: W.W. Norton. Newton, N. (2000). Conscious emotion in a dynamic system: How i can know how i feel. In R. D. Ellis & N. Newton (Eds.), The Caldron of consciousness: Motivation, affect and self-organization. Philadelphia: John Benjamins. Papineau, D. (1991). Teleology and mental states. Aristotelian Society Supplementary, LXV, 33–54. Piaget, J. (1954). The construction of reality in the child. NY: Basic Books. Piaget, J. (1970). Genetic epistemology. NY: Columbia. Pinker, S. (1997). How the mind works. New York: Norton. Pittendrigh, C. S. (1958). Adaptation, natural selection and behavior. In A. Roe & G. Simpson (Eds.), Behavior and evolution. New Haven: Yale University Press. Prigogine, E., & Stengers, I. (1984). Order out of chaos. New York: Bantam. Rosen, R. (1991). Life itself: A comprehensive inquiry into the nature, origin and fabrication of life. NY: Columbia University. Saidel, E. (2001). Teleosemantics and the epiphenomenality of content. In J. McIntosh (Ed.), Naturalism, evolution and intentionality. Canadian Journal of Philosophy, Special Supplementary Volume. Searle, J. R. (1992). The rediscovery of the mind. Cambridge, MA: MIT/ Bradford. Shani, I. (2005). Computation and intentionality: A recipe for epistemic impasse. Minds and Machines, 15, 207–228. Shani, I. (in press). Narcissistic sensations and intentional directedness: How second-order cybernetics helps dissolving the tension between the egocentric character of sensory information and the (seemingly) worldcentered character of cognitive representations. Cybernetics and Human Knowing. Smithers, T. (1995). Are autonomous agents information processing systems? In L. Steels & R. A. Brooks (Eds.), The artificial life route to ‘Artificial Intelligence’: Building situated embodied agents. Hillsdale, NJ: Lawrence Erlbaum Associates. Sober, E. (1984). The nature of selection. Cambridge, MA: MIT/Bradford. Stampe, D. W. (1990). Desires as reasons – Discussion notes on Fred Dretske’s ‘Explaining Behavior: Reasons in a World of Causes’. Philosophy and Phenomenological Research, 50, 787–793. Sterelny, K. (1990). The representational theory of mind: An introduction. Oxford: Blackwell. Stotz, K., & Griffiths, P. E. (2002). Dancing in the dark: Evolutionary psychology and the problem of design. In S. Scher & M. Rauscher (Eds.), Evolutionary psychology: Alternative approaches. Dordrecht: Kluwer. Ulanowicz, R. (1986). Growth and development: Ecosystems phenomenology. NY: Springer Verlag. Ulanowicz, R. E. (2000). Life after Newton: An ecological metaphysics. In: D. R. Keller & F. B. Golley (Eds.), The philosophy of ecology. Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience. Cambridge, MA: MIT.