Phenomenology and the Cognitive Sciences

, Volume 12, Issue 1, pp 1–26

Making it mental: in search for the golden mean of the extended cognition controversy


    • Department of PhilosophyKyung Hee University

DOI: 10.1007/s11097-012-9273-z

Cite this article as:
Shani, I. Phenom Cogn Sci (2013) 12: 1. doi:10.1007/s11097-012-9273-z


This paper engages the extended cognition controversy by advancing a theory which fits nicely into an attractive and surprisingly unoccupied conceptual niche situated comfortably between traditional individualism and the radical externalism espoused by the majority of supporters of the extended mind hypothesis. I call this theory moderate active externalism, or MAE. In alliance with other externalist theories of cognition, MAE is committed to the view that certain cognitive processes extend across brain, body, and world—a conclusion which follows from a theory I develop in “Synergic Coordination: an argument for cognitive process externalism.” Yet, in contradistinction with radical externalism, and in agreement with the internalist orthodoxy, MAE defends the view that mental states are situated invariably inside our heads. This is done, inter alia, by developing a novel hypothesis regarding the vehicles of content (in “Extended cognition without externalized mental states”, and by criticizing arguments in support of mental states externalism (in “Reflections and objections”). The result, I believe, is a coherent theoretical alternative worthy of serious consideration.


Cognitive engagementIntrinsic contentInstantiative vehicles of contentMental states externalismModerate active externalismParity principleProcess externalismRadical externalismSynergic coordinationTransformative vehicles of content


The debate concerning the boundaries of cognition constitutes one of the liveliest areas of controversy in contemporary philosophy of mind and theoretical cognitive science. Over the last 15 years or so, the pros and cons regarding the hypothesis of extended cognition (HEC)1 have been explored rigorously and enthusiastically, leading to increased sophistication on both sides of the fence. There is now an impressive array of arguments for, and versions of, extended cognition (e.g., Clark 2008; Clark and Chalmers 1998; Gibbs 2001; Haugeland 1998; Hurley 2010; Hutchins 1995; Menary 2006; Noë 2004; Rockwell 2005; Rowlands 2003; Silberstein and Chemero 2011; Sutton 2010; Theiner 2011; van Gelder 1995; Wilson 2004), and these are being met by ever more articulated counterarguments, issuing from advocates of the internalist cognitive orthodoxy (e.g., Adams and Aizawa 2001, 2008; Block 2005; Grush 2003; Horgan and Kriegel 2008; O’Brien 1998; Preston 2010; Prinz 2006; Rupert 2004; Weiskopf 2008). Consonant with the nature of such long-standing controversies, there are also some voices urging pluralism and proclaiming the need to acknowledge the truth in each of the opposing parties (Yoshimi 2012). It is therefore not without some trepidation that one may venture to offer yet another contribution to this debate. For it may legitimately be asked what is there to be said which needs to be pronounced but has not yet received proper articulation?

This being said, it is my conviction that the perspective which I am about to advocate here is both novel and worth voicing. No doubt, various components of the view I shall present share significant degrees of overlap with previous arguments emanating from the work of major figures on both sides of the divide. Likewise, the reconciliatory spirit of the present paper resembles in its tone those scant voices calling for a compromising, golden mean assessment of the strengths and weaknesses of the opposing camps. Nevertheless, the argument as a whole presents a picture which, despite its plausibility, has not, to my knowledge, received the attention it deserves. Part of this neglect, I suppose, has to do with the natural aversion of active participants in the debate to concede that the program they advocate suffers from certain structural limitations which can only be amended at the cost of attenuating its imperialist aspirations. Ditto, it may also have to do with an understandable distaste for wishy-washy, middle of the road conceptual alternatives. But, if I am right, an important reason for the fact that the view I am about to present has not been championed earlier is anchored in the complete absence of a crucial conceptual distinction—namely, the distinction between instantiative and transformative vehicles of content—which, as I shall argue (see “Extended cognition without externalized mental states”), is of paramount importance for the debate. Correspondingly, my argument for the position I favor, which may be described as moderate active externalism (MAE), rests heavily on this distinction.

I argue that each of the two opposing parties to the debate is correct with respect to a crucial tenet denied by the other party and that once this fact is sufficiently appreciated a synthesis suggests itself which promises to incorporate the best in both. Supporters of HEC, I claim, are right to maintain that at least some cognitive processes are constituted, in part, of trans-cranial (that is, bodily and environmental) variables. I call this thesis process externalism or PE for short. In defending PE, I appeal to an argument I call the argument from synergy which, I believe, effectively captures the rationale behind many previous arguments motivating the externalization of cognitive processing. But while in support of PE, I deny another major thesis which most active externalists endorse, namely, the view that mental states may be similarly externalized. In denying mental states externalism (MSE), I side with the internalist orthodoxy.

The result is a synthesis which emphasizes the division of labor between inner and outer. Only the inner (i.e., intracranial) components of cognition, I argue, can act as bearers of content, but these bearers are systematically facilitated, transformed, and enabled by external variables to which they are coupled in a reciprocal dynamics of synergic coordination. In rendering this idea explicit, I introduce the distinction between instantiative and transformative vehicles of content (see “Extended cognition without externalized mental states”), arguing that this is a crucial conceptual distinction which helps clarify many issues and is sorely missed in the literature. I conclude by addressing some of the major avenues available to the radical externalist in response to my assault on MSE, offering reasons for deep skepticism with respect to these channels of refuge.

Synergic coordination: an argument for cognitive process externalism

Process externalism: what it is and why it is a problem

Arguments for active externalism do not always aim at establishing the same goal. Sometimes, they are meant to establish the thesis that cognitive systems are extended (e.g., Haugeland 1998; van Gelder 1995), at other times they are framed as favoring the extended status of cognitive processes (e.g., Menary 2006), and still on other occasions the emphasis lies on establishing the externality of mental states (e.g., Clark and Chalmers 1998). To be sure, many supporters of HEC advocate all three extension theses simultaneously, but the theses are nevertheless mutually distinct. Yet, while the distinction between system externalism and process externalism is universally accepted, the distinction between process externalism and (mental) state externalism is sometimes met with puzzlement and skepticism.2 This skepticism is, I think, unfounded and can be easily mitigated yet I shall postpone the relevant discussion to the next section, which is dedicated to a critical examination of active externalism with respect to mental states. In the present section, I focus on process externalism, that is, on the question do cognitive processes extend beyond the cranium and into the body and the environment of a cognitive agent? I shall not, in this paper, discuss the issue of system externalism, largely because the observations I would like to make with respect to the controversy surrounding HEC have more to do with the other two active externalist theses.3

Process externalism (PE) occupies center stage in the debate. Supporters of HEC argue that some (arguably many) cognitive processes extend beyond the cranium and are distributed across brain, body, and world. This being the case, it follows that trans-cranial variables are literally constitutive of such processes, making cognition an extended affair. This claim has provoked much alarm among critics, giving rise to a host of complaints and accusations, chief among which are the following two. First, there is the so-called coupling-constitution fallacy (CCF for short), namely, the claim that arguments advancing PE involve an illicit leap from the relatively innocent premise that body and world play a myriad of indispensable causal roles in support of cognitive processing to the conclusion that they literally constitute such processing (see, e.g., Adams and Aizawa 2001, 2008; Block 2005; Prinz 2006). Second, there is the accusation that PE compromises certain definitive marks of the mental, admitting as cognitive components process parts which clearly should not qualify as such.

The latter allegation is, in turn, driven by two distinct types of arguments. First, there are arguments to the effect that externally located cognitive “processes” or “sub-processes” do not adequately resemble well-established and well-studied cognitive processes, thereby failing the empirical test of matching up with the classificatory kinds of contemporary cognitive science (see, e.g., Adams and Aizawa 2008, chap. 4; Rupert 2004). Second, there is the more philosophical argument according to which PE runs afoul of a fundamental fact about cognitive processes, namely, that such processes are repositories of intrinsic content (see Adams and Aizawa 2001, 2008, chap. 3).

It is hard not to get lost in the thick canopy of arguments and counterarguments surrounding this issue, in particular since these are often ramified into distinct and even competing varieties (notice, for example, the tension between so-called first-wave and second-wave arguments for PE).4 On the good side, this abundance of information provides us with ample materials with which to approach the problem. My own view is that we are already in command of enough evidence to reach a plausible verdict and that this verdict must be in favor of PE, namely, in favor of the view that many cognitive processes extend beyond the brain and are constituted, in part, of trans-cranial variables.

Nevertheless, in order to show exactly why PE is the order of the day, work still needs to be done on both the theoretical and the meta-theoretical levels. At the meta-theoretical level of the debate, there is, I think, a need to better articulate the background conception of cognition against which the very idea that cognitive processes extend beyond the cranium is rendered plausible and even probable. It will be noticed that while some background conceptions of cognition are hospitable to PE, others are inhospitable, and that this fact plays a crucial role in fixing one’s intuitions when faced with the task of assessing whether external variables are literally constitutive of a certain cognitive process P or, alternatively, are merely causal factors relevant for, yet non-constitutive of, P. Given that opposite intuitions concerning such matters are recurrent in the debate, it seems worthwhile to attempt to articulate the background conception against which PE best makes sense.5

On a more straightforward theoretical level, there is still a need to articulate, as best as possible, the major and most general reason in support of the conclusion that an impressive array of cognitive processes extend beyond the brain proper. It may be the case that there is no one major reason of this sort but, rather, a plurality of arguments applicable to different process types in different contexts of application, yet I am inclined to think that this is not so and that underlying most, if not all, of the good arguments for PE is a single deep-structure logic. If such a master argument exist, then surely we are to benefit from attempting to latch onto it, while, if it does not, then our failed attempt would nevertheless constitute a good exercise.

Consequently, my major goal in the present section is to articulate what I see as the best and most general reason in favor of PE. The argument bears recognizable similarities to arguments already present in the literature, in particular to second-wave, integrationist arguments for extended cognition (Menary 2006; Sutton 2010), as well as to one of Andy Clark’s defense lines of HEC (Clark 1998, 2008, 129–131). Nevertheless, as we shall soon see, it is inspired by different sources. More importantly, I believe that it enjoys greater explanatory generality than alternative extant proposals.

Some meta-theoretical musings on the viability of PE

I mentioned earlier that PE is rendered plausible, let alone probable, only against certain background assumptions regarding the nature of cognition. Thus, for example, if the conception of cognition operative at the back of one’s mind is, say, that of conscious mental deliberations, or perhaps the manipulation of inner symbolic codes, then one would struggle to come to terms with the idea that trans-cranial variables are constitutive of cognitive processes and would be strongly inclined to interpret the contribution of such variables as one or another form of external (viz., non-constitutive) causal influence. The question thus arise what background conception of cognition renders PE a viable, perhaps even a natural stance to take, and is this conception compelling in its own right?6

The answer which I prefer, and which also seems to be in sync with much of the literature on extended cognition, is that PE is groomed within a conception of mind which views cognition as a dynamical activity of an adaptive character, a form of active engagement with the world. Notice that the point here is not merely to restate the consensual assertion that cognition enables, or sustains, or contributes to successful engagements, but rather, more radically, to give expression to the view that cognition literally is a form of engagement. Virtually any theory of cognition affirms that cognition is relevant for action but the claim that it is, itself, a form of active engagement is far from trivial, and it is this latter conception which, I believe, renders PE a more probable thesis than it would otherwise be. Cognitive engagements, such as perceiving, debating, remembering, or foraging, are laden with meaning—they are guided by meaningful information and involve the search for, and the acquisition, manipulation, transformation, dissemination, gauging, production, co-production, etc. of such information, recurrently—but they are active engagements nevertheless.7 Cognitive processes are, on this view, simply the processes through which cognitive engagements unfold.

From here, only a very small step is required to see why this conception of cognition is hospitable to PE: if it turns out that some cognitive engagements unfold as (i.e., take the form of) coordinated process configurations in which intracranial and trans-cranial variables operate in tandem, then we have every reason to conclude that the processes which substantiate such engagements—ergo, the relevant cognitive processes—are, in part, extended (the general logic of the argument resembles, in part, that of Rowlands 2006, chap. 3).

Noticeably, nothing here necessitates prior commitment to HEC. The background assumptions which motivate a conception of cognition as active engagement with the world seem to follow from a certain understanding of what it means for cognition to be embodied, situated, and geared towards action.8 In particular (in my view at least), this outlook is tied to the assumption that cognition is an active mode of adaptation—an emergent form of self-governance aimed at making a living in the world by way of transforming interactive situations (that is, transforming the patterns of interrelationships between the agent and its environment). Attempting to articulate precisely how this is brought out is beyond the pale of the present paper9; the important point, however, is that in the current intellectual climate where embodied, situated, and action-oriented conceptions of cognition are widespread such a stance is plausible enough to be taken seriously, and this is all that supporters of PE need in order to proceed to the task of articulating concrete arguments for their view.

The argument from synergy

Although, from a psychological perspective, no human cognitive activity should, in the final analysis, be considered in complete isolation from the body and the physical and social environments, some cognitive engagements seem to unfold purely internally (e.g., daydreaming or mental calculation). However, many other engagements—from spatial reasoning, to visual perception, to navigation, and to countless other activities—are executed as coordinated ensembles comprising both intracranial and trans-cranial elements. Such cross-boundary process configurations constitute a heterogeneous lot, yet if instead of immersing ourselves in the sea of concrete details, contexts, and circumstances we step back in search for “the pattern which connects” (as Gregory Bateson would put it), I think we will find that there is indeed a characteristic pattern to the manner in which internal and external factors bind together to form a coherent extended cognitive process.

When looking at cognitive engagements such as gesture-mediated spatial reasoning, visual perception, or navigation, we find that such engagements reach their closure (that is, are carried out or fulfilled) through a close coupling of internal and external variables. More informatively, we find that the coupling takes the form of a cyclical causal pattern, a feedback loop, in which intracranial, bodily, and environmental variables are synergically coordinated, with the result being that, through their reciprocal causal influence, they sustain and reenact (recursively generate) the very cyclical causal pattern which binds them together. Since it is the cyclical process configuration as a whole—call it C—which instantiates the relevant cognitive engagement we are, on the present view, justified in taking C in its entirety to be a cognitive process. And since C is constituted in part of trans-cranial variables, there seems to be no escape from the conclusion that cognitive processes such as C are extended.

Let me illustrate this idea with recourse to visual perception. I focus first on demonstrating how perception spread to the body, but will also say something about its potential spread to the environment. Imagine that a cognitive agent S is engaged in visually tracking the movements of a prey. Naturally, part of achieving this task involves the activation of a set of intracranial processes culminating in a temporally extended sequence of visual experiences. It involves, in other words, processes inside the eye, the optic nervous, and the brain (in particular LGN and the visual cortices). Call this set of intracranial processes X. However, the act of perceptual tracking is equally dependent on a set of bodily skills such as orchestrated eye movements, head movements, limb movements, and postural adjustments. Call this set Y. Now, the point is that X and Y are coupled to each other in a feedback loop of synergic coordination, repetitively modulating, reenacting, and reinforcing each other: orchestrated motor outputs facilitate and regulate the flow of optical information, which in turn guides further motor outputs leading to still more visual inputs, and so on.

In contrast with the classical stimulus–response schema (the so-called reflex arc), there is no obvious beginning or end to this process: it consists all the while of a coordinative dynamics whose constitutive factors (sensory, motor, and, we may add, intermediary states) operate in tandem. In short, the process of seeing, through which the task of tracking the movements of the prey is satisfied, is shown to be a dynamically holistic process which depends on Y as much as it depends on X, or to put it better: it consists of the entire functional cycle through which X and Y are synergetically linked.

Essentially, this is the view that Dewey was defending back at the end of the nineteenth century. In his famous article The Reflex Arc Concept in Psychology (1896) Dewey argued that perception is a dynamical interactive process, an act of inquiry, enacted as a sensory–motor coordination circuit in which sensory inputs and motor outputs modulate each other and are constantly involved in each other’s functioning. To describe the feedback loop through which such coordination is affected, he coined the term functional cycle. Several decades later, the German physiologist Jakob von Uexküll (1934/2010) used the same term in his own description of the cyclical process dynamics underlying visual perception (although, unlike Dewey, Uexküll’s functional cycle contains explicit reference to inner representations, dynamically interpolated between the sensory–motor components).10 Still later, J.J. Gibson (1966, 1979) characterized perception in similar terms, describing it as an integral “achievement of the individual” (1979, 239) ensuing from the synchronized operation of a hierarchy of perceptual systems each of which embodies feedback loops of sensory–motor coordination. We can see, then, that this idea has a history of endorsement and a tradition of thought behind it.11

Nor is this type of reasoning restricted in any way to perceptual processes. On the contrary, it seems to generalize quite promisingly. To mention but one of the more well-investigated cognitive processes cited by supporters of HEC, a similar pattern of synergic coordination binds inner thought sequences to gestures, culminating in a functional cycle which constitutes an extended process of spatial reasoning as studied, for example, by Goldin-Meadow (2003; see also Clark 2008).

So far, our argument for PE touched only on the extension of cognitive processes to the bodies of cognitive agents. We have not said anything yet about extension into the environment. Yet, even in its current limited form the argument lends support to the constitution hypothesis, namely, to the claim that some cognitive processes are literally constituted, in part, of trans-cranial variables. This, in itself, is already a radical departure from traditional lore, even if one is reluctant to go further and acknowledge, as most supporters of HEC do, that cognitive processes extend not only beyond the brain but also beyond the skin.

But in fact, there are good reasons to go further. As Uexküll stressed a long while ago, functional cycles range, in part, over portions of one’s surroundings; they involve transformations of elements in the agent’s environment (which elements may either be primary targets of interaction or means to an end) and these transformations, in turn, induce changes in the agent’s sensory–motor apparatus and neural organization, leading to further alterations in the environment and so on and so forth. In other words, functional cycles typically integrate not only a set X of intracranial processes and a set Y of bodily skills but also a set Z of environmental variables. Effective engagement is achieved through a skillful coordination of all three aspects.

J.J. Gibson and the ecological psychologists who followed in his footsteps perfected these insights by arguing that agents encounter their environments as ecological niches full of affordances. Since a niche, understood along such lines, specifies the potentials for interaction available to the embedded agent the idea that the affordances it presents play an integral role in the unfolding of cognitive engagements seems to come naturally. In a recent account of extended cognition along Gibsonian lines, Silberstein and Chemero (2011) provide a nice illustration of an X–Y–Z dynamical feedback loop integrating environmental affordances, sensory–motor couplings, and endogenous neural activity (see Fig. 1).12
Fig. 1

Animal environment system [from Silberstein and Chemero 2011, with kind permission of the authors]

The significance of environmental variables to the making of cognitive processes is especially conspicuous when it comes to human beings, as many advocates of HEC rightly emphasize (e.g., Donald 1991; Hutchins 2010; Rowlands 1999; Sutton 2010). Human beings are able to manipulate their environments to a degree unmatched by any other species, and they can design their environments so as to make them conducive to the unfolding of various cognitive engagements. Moreover, the environments of humans are, in many ways and on many levels, social environments; a fact which opens up vast riches of possibilities for interpersonal coordination of cognitive engagement. Many of the examples favored by supporters of HEC are testimony to the uniquely human social and technological aspects of cognitive engagements: from reliance on exograms as memory aids (Clark and Chalmers 1998; Donald 1991; Sutton 2010), to the restructuring of the environment in support of a cognitive task (Hutchins 1995; Tribble 2005), to transactive memory (Wegner et al. 1991).

Taking stock

Earlier, I ventured to express my belief that the argument from synergy (or something near enough) captures the general pattern behind many of the best examples of extended cognitive processes. If I am right, then external variables play a constitutive role in a cognitive process P when P, an instantiation of a cognitive engagement, takes the form of a coordinative effort whose coherence and operational closure depend, in part, on the regulatory contribution of such variables. To my understanding, this is what happens not only in such processes as perception, gestured spatial reasoning, or navigation but also in a variety of other cognitive engagements involving online manipulation of the environment (e.g., Tetris), utilization of external symbolic storage spaces (e.g., the celebrated notebook of Otto, the alzheimer patient), or even ongoing systematic dependence on other people’s mental processes (e.g., subcontracting parts of one’s memory tasks to a personal aid). However, that the argument succeeds in so generalizing is, at the moment, only hypothesis which I shall not attempt to prove here (but see “External symbolic storage devices as transformative vehicles of content” for further discussion). Instead, I would like to examine where this proposal stands with respect to other proposals similar in spirit and also to comment on an intrinsic feature of the argument which has significant implications for the debate concerning extended cognition.

As mentioned before, the argument from synergy resembles certain other arguments in support of PE. One such argument is advanced by Clark (2008, 131) as a response to the challenge of CCF. In his response, Clark helps himself to a useful analogy. He points out that, in a turbo-driven engine, exhaust flow, produced by the engine, is used to initiate the process of compressing the air flowing into the engine which, in turn, allows the engine to generate more power. Thus, the exhaust flow, a product of the mechanical work generated by the engine, is at the same time a catalyst for further work, with the result being that the entire work cycle—from engine to exhaust and back—becomes a single, extended power generating process. Clark then argues that the same moral applies to cognitive processes: in cases where a cognitive process (for example, spatial reasoning) creates products (e.g., gestures) which, once produced, “drive the cognitive process along… any intuitive ban on counting inputs as parts of mechanisms seems wrong” (Ibid).

Essentially, the underlying logic of Clark’s argument is the same as the one I advocate here, which is not terribly surprising given that functional cycles in biological and psychological systems simply are specialized work cycles (Kauffman 2000). However, as it stands, Clark’s argument lacks sufficient generality precisely because it is designed as a response to CCF. Recall that CCF challenges the inference from “A is coupled to B” to “A is a constitutive part of B” (where B is a configuration of intracranial processes and A is a set of external variables) which, in turn, presupposes that prior to the coupling B is already a coherent cognitive process. Clark’s response to the challenge shares this supposition, while insisting that certain forms of coupling do indeed justify an inference to constitution. However, there is little reason to doubt that many cognitive processes do not conform to a picture according to which B is already a coherent cognitive process (of the relevant type) which, by assimilating A, becomes extended but, rather, are such that they emerge only as a result of the coupling. This being the case, the perspective from which Clark’s argument is envisaged seems to be too narrow.

The idea that cognitive processes (or at any rate, those processes which concern supporters of HEC) are intrinsically coupled and extended is championed in particular by Richard Menary (e.g., 2006). Menary defends a version of extended cognition which he calls cognitive integration (CI), a view according to which cognitive processes are the holistic outcome of the integration of “internal and external vehicles and processes” (Ibid, 329). He then uses CI as a platform from which to muster a defense of HEC in the face of the CCF challenge. The latter, he argues, is based on a distorted picture which assumes that the coupling of A and B (as per above) results in A becoming a constitutive part of an already fully formed B. But, if cognitive integration is the norm, then this assumption must go, dragging CCF along with it.

For my part, I am inclined to think that this is going a bit too far. I see nothing outrageous in the idea that there are certain circumstances in which an already well-formed, purely internal cognitive process gradually turns into an extended process dedicated to the same cognitive task. Imagine that you return to a city you once knew well. At first you try to negotiate your way around using your memory but eventually, after staring defeat in the face, you turn to the map or the GPS. You are engaged in navigation all the while but the process gradually extends through sensible coupling to external aids. To deny that such extension through coupling ever takes place seems to me an overreaction.

Nor is there a need for such extreme measures. The argument from synergy does not discriminates between (1) situations in which B is already a well-formed cognitive process, which then gets extended via the coupling to A, and (2) situations in which neither A nor B are veritable cognitive processes in their own right but, being coupled, they give rise to an integrated cognitive process C. What matters is that A and B are coupled to each other through synergic coordination: in the first scenario such synergy makes A a part of (the now extended) B; in the second scenario it makes A a part of C. It is, I think, a virtue of the argument from synergy that it does not force us to exclude either scenario a priori.

Regardless of such differences, there is an important similarity between the proposal I advocate here and Menary’s cognitive integrationism, as well as other works corresponding to what Sutton (2010) describes as second-wave extended mind theories.13 On all of these accounts, extended cognitive processes are portrayed as the result of an integrative dynamics in which the contribution of external variables is complementary to that of internal factors. The process as a whole is seen as a unity in diversity, presupposing a functional division of labor between brain, body, and world. To illustrate, consider again our stock example of visual perception. While motor adjustments and environmental affordances coalesce with neural dynamics to form an extended process of seeing they do not replicate, or mimic this neural dynamics; rather, their contribution to the process of visual perception is complementary to the contribution of inner sensations, percepts, or memories.

As we shall see in the next section, there is a tension between this emphasis on complementarity and another popular strand within the extended cognition movement which emphasizes the functional equivalence of inner and outer variables. Whether this tension is real or apparent is one of the topics with which we will need to grapple.

Parity and mental states externalism

Parity and complementarity

Our argument for PE ended on an interesting note, namely, with the observation that extended cognitive processes seem, in general, to presuppose a functional division of labor according to which internal and external components play complementary roles to each other. I also mentioned, more generally, that an emphasis on complementarity is a characteristic feature of second-wave extended cognition theories. In contrast, one of the earliest and most influential arguments in favor of the extended mind is driven by a rather different rationale, stressing the functional isomorphism of inner and outer variables rather than their complementarity. I am referring, of course, to Clark and Chalmers’ (1998) celebrated appeal to a parity principle, according to which:

“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).

Clark and Chalmers discuss the hypothetical case of Otto, who suffers from Alzheimer. Due to his condition, Otto is using a pocket notebook to store and retrieve information on which he systematically and reliably depends while being engaged with daily tasks which normal people accomplish through the use of biological memory. Thus, for example, where Inga, a normal subject, finds her way to the Metropolitan Museum of Modern Art by relying on her recollections, Otto has to write down the relevant information as an entry in his notebook and then negotiates his way to the museum by consulting his little book when needed. Using parity, Clark and Chalmers contend that Otto’s memory is extended into his notebook and that we may justifiably treat the entries in the notebook as non-occurrent beliefs.

Clearly, by relying on the parity principle, Clark and Chalmers base their argument for extended cognition on the functional equivalence of inner and outer parameters. Otto’s memory is extended (the idea goes) because the entries in his notebook play for him the same role that neurally implemented non-occurrent beliefs play for normal subjects. This emphasis on functional isomorphism seems to be pulling in the opposite direction from our previous emphasis on complementarity: in appealing to parity one relies on the similarity between the inner and the outer constituents of cognition, while an appeal to synergy utilizes their differences.

This tension is not lost on Clark (1998), as well as on other supporters of HEC (e.g., Menary 2006; Sutton 2010), but, for the most part, it is considered more apparent than real.14 Thus, Clark argues that parity-driven arguments ought to be seen as articulating a polemically virtuous, sufficient condition for cognitive extension rather than a necessary one. While Sutton, in a similar spirit, argues that parity-based and complementarity-based arguments for HEC are not utterly incompatible and that the former can be seen as a special case of the latter. If I understand Sutton correctly then the idea is this: Otto’s notebook plays for him the same role that neurally ingrained dispositional beliefs play for Inga (hence parity) but, at the same time, the notebook plays a complementary role to Otto’s own impaired biological memory (hence complementarity). Notice, however, that insofar as one relies on complementarity alone one is no more entitled to conclude that Otto’s scribbling constitute beliefs anymore than she is entitled to conclude that gestures constitute thoughts. The only permissible inference is that the entries constitute a part in Otto’s relevant cognitive process and that this part is complementary to the workings of Otto’s biological memory. It is parity that accounts for the stronger conclusion that the notebook contains beliefs. In short, even if parity is compatible with complementarity, it adds independent content to HEC, and this content ought to be assessed independently.

Mental states externalism

The parity principle adds to the hodgepodge a more radical input. For, while synergic coordination gave us reasons to uphold PE parity leads to a stronger thesis, it leads to mental states externalism (MSE), namely, to the view that some mental states—beliefs, desires, etc. —literally reside outside the head. I shall promptly explain why, and in what sense, MSE is stronger than PE, but first, in order to prevent misunderstanding, I shall have to say in what sense I consider mental states to be different than cognitive processes and, correspondingly, why I consider MSE to be distinguished from PE.

By cognitive process, I mean a process, or process configuration, which realizes a cognitive engagement such as visual tracking, imagining, reasoning, memorizing, etc. (see “Some meta-theoretical musings on the viability of PE”).15 The realization of a cognitive engagement often involves a constellation of interlocked components operating in tandem across brain, body, and world—which is the reason why, in consonance with other supporters of HEC, I endorse PE, the view that some cognitive processes extend beyond the brain and even the skin of a cognitive agent. In referring to mental states, on the other hand, I follow the standard textbook procedure according to which they are being identified with such things as sensations, percepts, imaginations, feelings and emotions, moods, thoughts, propositional attitudes (e.g., beliefs, desires), and volitional states (e.g., intending, willing).

It can be readily verified that, when this conceptual distinction is borne in mind, there are visible differences between mental states and cognitive processes. In particular, it is not merely that a mental state is but a phase, or a temporal slice, of a temporally extended cognitive process, in the manner in which a single thought relates to a thought sequence, but rather, that extended cognitive processes contain much more than mental states since they contain such things as motor adjustments, gestures, navigation instruments, and great many other things which, while not being mental states in themselves, are nevertheless suitably coupled to such states so as to generate coherent cognitive engagements.

It follows that MSE must be distinguished from PE since it insists not merely that some cognitive processes are, as a whole, extended but, rather, that some cognitive processes are such that they are extended (solely or in part) in virtue of the fact that they contain mental states which are themselves extended. Clearly, MSE is stronger than PE in the straightforward logical sense that it asymmetrically entails the latter thesis. This difference in logical strength corresponds to the fact that MSE is philosophically more radical: for while PE says that there is more to cognition than the workings of inner mental states, MSE implies that there is more to mental states than inner mental states! Not surprisingly, some of the strongest reactions against HEC are reactions against the parity principle and in particular against the implication that such things as the entries in Otto’s notebook realistically qualify as genuine beliefs.

One of the major sources of resistance to the contention that Otto-style “beliefs” qualify as genuine beliefs is that they fail to be bearers of intrinsic content (see, in particular, Adams and Aizawa 2001, 2008).16 My own ordinary beliefs need not, as a general rule, be interpreted by myself; they are intrinsically meaningful to me; they constitute an immediate integral part of the sense (and nonsense) making agent that I am, and it is through such beliefs (as well as other mental states such as feelings) that I come to comprehend and interpret external events, situations, texts, and so on. In contrast, marks on paper enjoy no such intrinsic meaningfulness. They are meaningful only insofar as a cognitive agent (or a community of such agents)—be it a writer, a reader, a critique, a translator, or an exegetic scholar—animates them with meaning. In the absence of such animation, the text is bound to remain obscure (as it would appear to the illiterate) or completely meaningless (as it would be for an ant crawling on an open book). But of course, animation presupposes the capacity to assign the written words an internally comprehensible meaning and this, in turn, presupposes the existence of a source of semantic significance more primitive than the text.17 This, in essence, is what is meant when it is asserted that the association of written texts with content is derivative, rather than intrinsic.18

More generally, the concern with respect to MSE can be stated as follows. Mental states as traditionally conceived (in particular propositional attitudes such as beliefs, desires, and their ilk) appear to be loci of intrinsic content. Moreover, the possession of such content is what makes a mental state, say the belief that foxes are wily, the kind of mental state that it is—it is what endows it with a unique psychological significance. And, of course, mental states as traditionally conceived are exclusively in the head. MSE implies that some mental states are located outside the head—inter alia, in such items as pocket notebooks. But, by so doing, MSE also implies that some mental states are devoid of intrinsic content. If so, then the thesis threatens to strip the concept of mental state of that which gives it vitality in the first place—it throws the baby out with the bathwater.19

It would seem, then, that MSE runs into conflict with the idea that mental states are, invariably, bearers of intrinsic content. To add to the problematic, the thesis is unintuitive and its defense requires complicated maneuvers. For example, it requires the maintenance of a delicate balance between, on the one hand, insisting that Otto-style “beliefs” and “memories” are sufficiently like ordinary beliefs and memories, while, on the other hand, acknowledging (as we must) that, in some important respects, they are not at all alike. This, by all means, is no easy task, and the thesis has been under constant pressure since its inception (for some criticisms, see Adams and Aizawa 2008; O’Brien 1998; Preston 2010; Rupert 2004; Weiskopf 2008).

Nevertheless, supporters of HEC, of various different persuasions, tacitly presuppose that they must defend MSE at all costs. This they do, I suspect, because they hold on to a conception of the vehicles of content which seems to encourage, if not entail, such radical externalism. I turn next to a critical discussion of this highly charged concept.

Extended cognition without externalized mental states

Vehicles of content: a critical discussion

In the previous section, I introduced a challenge to MSE based on the claim that genuine propositional attitudes are, invariably, loci of intrinsic content and that externalized “mental states” do not satisfy this condition. In the characteristic terminology of the debate concerning extended cognition, one would expect that I would formulate the challenge to MSE by reference to the concept of vehicle of content, namely, as the claim that externalized “mental states” fail to be vehicles of intrinsic content. Unfortunately, this cannot be done without first introducing an important refinement of our terminology. For, as it stands, the term “vehicle of content” does not distinguish between two significantly distinct senses in which a physical structure may be reasonably described as a vehicle of content. Worse still, in the absence of this important distinction, the notion of vehicle of content serves to confirm MSE all too easily and begs the question against the argument from intrinsic content.

It is customary to distinguish between the content of a representation and the vehicle through which a representation with such content is represented in one’s mind (Dennett 1991; Hurley 1998; Millikan 1991). Advocates of extended cognition often describe their position as vehicle externalism (Hurley 1998; Rowlands 2003), stressing the fact that theirs is a more radical form of externalism than the content externalism pioneered by Kripke (1972); Putnam (1975), and Burge (1979). Vehicle externalism, however, is a rather general thesis. In essence, it is merely a restatement of active externalism, or of the extended mind thesis, namely, of the claim that the mind and the cognitive processes that constitute it extend beyond the brains and the skins of cognitive agents (see Wilson 2010). Thus, when used in such a broad sense, the term “vehicle” applies indiscriminately to the supervenience base of either a cognitive system, or a cognitive process, or a mental state, and it asserts with respect to all of them that their supervenience base is wide rather than narrow. In itself, then, the concept of vehicle externalism is too broad to be directly useful for our current purpose.

What we need is a more exact terminology with which to express the idea that the vehicles corresponding to content bearing mental states spread beyond the boundaries of brain and skin. At first glance, one would think that the term “vehicles of content” suitably serves this purpose but, unfortunately, that is not the case. One problem with this notion is that advocates of HEC sometimes use it in a way which gives room for suspicion that they refer indiscriminately both to tokens of cognitive processes and to tokens of mental states. Still, some are careful enough to distinguish the two. Mark Rowlands (2006, 33), for example, uses the term vehicles of cognition to refer to tokens of cognitive processes while reserving the term vehicles of content to refer to tokens of mental states. Yet, even this is not enough since, as commonly used, the concept of “vehicle of content” is notoriously ambiguous.

The term “vehicle” can be used in two different senses. In the first sense, to be a vehicle of X is roughly synonymous with being a carrier, or bearer, or a host of X. But in another sense, to be a vehicle of X is to be a conveyor or transmitter of X. Sometimes, these two senses converge. A car, for example, is both a host and a conveyor of the person who uses it. But such convergence is not universal. There are countless processes in which a property P is groomed throughout the process as a whole—the result of various contributing factors—but is nevertheless instantiated locally, at a specific phase or part (often the terminal point) of the process. For instance, although the process of getting sunburned crucially involves interaction with the Sun it is the skin, and not the sun (or our sun-exposure behavior, for that matter), which is the bearer of sunburns. The Sun and our careless behavior drive the process along, but sunburns are nevertheless skin-bound. Ditto, while a speedometer depends crucially on the workings of various structures, chief among which is a rotating magnet, it is the needle and the dial which instantiate the property of actually indicating the car’s speed to our eyes.

I think that something rather similar also obtains when it comes to the analysis of vehicles of content. Thus, in one sense, to be a vehicle of content is to be a bearer of content, i.e., to be the casual substrate which instantiates a content-laden property. But in another sense, to be a vehicle of content is to be a process, or structure, or item, which plays a role in enabling, or facilitating, or transforming a content-laden property. Let us tentatively call these two senses V1 and V2, respectively.

It is important for me to stress that this is not just a fanciful embellishment on my part but rather that these two senses are actually employed in the literature—albeit without sufficient awareness for their differences. Thus, on the one hand, it is said that the vehicles of content are the bearers of mental states (Rowlands 2006, 33), an interpretation which corresponds to V1. On the other hand, it is sometimes maintained that the vehicles of content are “certain enduring material aspects” which “play a special role in enabling the system to possess… a given mental state” (Clark 2005, 1), an interpretation which corresponds to, or at least accommodates, V2. Still at other times, it is maintained that vehicles of content are processes which explain “particular thoughts on particular occasions” (Hurley 1998, 2), an interpretation which evidently accommodates both senses.

Yet, clearly, these two senses are distinct. For just as it is possible for the sun to play a special role in enabling a person to possess sunburns without it being the case that the sun itself is a bearer of sunburns, it is also possible for a variable V (a process, an object, a state of affairs) to play a special role in enabling a system to possess a given mental state M without it being the case that V is a bearer of M. I therefore propose that, instead of referring indiscriminately to vehicles of content, we distinguish between instantiative and transformative vehicles of content, corresponding to V1 and V2, respectively. .

An instantiative vehicle of content is any causal structure which instantiates intrinsic content at the most direct level of instantiation.20 Neurally realized mental states are, of course, our paradigmatic examples of instantiative vehicles of content. In contrast, a transformative vehicle of content is any enduring material aspect of a cognitive process (to borrow Clark’s jargon) which plays a special role in enabling or facilitating the occurrence of a given instantiative vehicle. In other words, transformative vehicles of content act as operators which transform the flow and the quality of the instantiative vehicles of content enjoyed by a cognitive agent.

Few clarifications are in order here before we proceed. First, let me stress that the two senses just described are not mutually exclusive. It is quite common for mental state tokens (percepts, concepts, etc.) to not only be instantiative vehicles of content but also to act as transformers of other mental states further downstream. Moreover, a mental state had by one person may act as a transformer of the mental states of another person (see “Extended cognition without externalized mental states” for further discussion). Nevertheless, the two senses are mutually independent which is to say that neither of them implies the other.

This conceptual separateness is of crucial importance for my argument since it shows that from the fact that something, a certain X, plays the role of a transformative vehicle of content one cannot validly deduce that X is also an instantiative vehicle of content. This has significant implications for the present discussion. Mental state tokens, I argue, are instantiative vehicles of content, from which it follows that in order to substantiate MSE supporters of this thesis must show that their transcranial candidates for the title “mental state” qualify as instantiative vehicles of content. But instead, what one finds in the literature on extended vehicles of content gives credence only to the conclusion that such vehicles act merely as transformative vehicles.

For my part, this is anything but surprising. Earlier, I advocated an account of cognitive processes which highlights the functional complementarity of the inner and the outer components of cognition. I would like to illustrate next how this division robustly parallels a differentiation between instantiative vehicles of content, on the one hand, and (merely) transformative vehicles of content, on the other hand. Going back to our example of visual tracking, I maintain that the neural substrates of the agent’s visual experiences play the role of instantiative vehicles of content while the agent’s motor adjustments, the affordances of the agent’s environment, and the structured optic array which specifies the affordances, all serve as transformative vehicles of content which regulate the flow and modulate the quality of the agent’s content-bearing states without, however, being themselves tokens of such content-bearing states. Ditto, when a child uses gestures to solve problems in spatial reasoning the gestures act as transformative vehicles of content, which modulate the flow of the child’s instantiative vehicles of content—her inner mental states—in optimizing ways, without, however, act as instantiative vehicles of content in their own right. In other words, I subscribe to a variant of what Rowlands (2006, 41) describes as the dual-component interpretation of vehicle externalism.

These examples illustrate just how important it is to distinguish between the two senses of being a “vehicle of content” and how mistaken it is to conclude that mental states are extended merely because some vehicles of content (namely, those that are purely transformative) are located outside the head. Admittedly, none in what has been said so far constitutes a conclusive proof that no transcranial structures are able to play the role of instantiative vehicles of content, yet I have yet to see any defense of MSE which manages to establish more than what we now know to be irrelevant, namely, that the proposed candidates are transformative vehicles of content.

At last, we are able to formulate the argument from intrinsic content in the language of vehicles: it is the claim that genuine mental state tokens are instantiative vehicles of intrinsic content and that, in contrast, externalized mental states are, at best, purely transformative vehicles and, as such, do not qualify as genuine mental states.

Significantly, it is not merely that in the absence of this distinction reference to the vehicles of content is chronically vague but also that this vagueness serves advocates of MSE in their polemics. For it becomes all too easy for them to use the fact certain external items act as transformative vehicles of content as a platform from which to argue that mental state internalism must be false, whereas, as a matter of fact, the latter doctrine is contingent on the claim that it is the instantiative vehicles of content which are brain bounded.

External symbolic storage devices as transformative vehicles of content

Yet, in order for the argument from intrinsic content to be complete, it must be shown that the dual-component interpretation applies adequately not merely with respect to items which are clearly non-semantic such as gestures and motor adjustments but also with respects to semantically interpretable external symbolic storage systems such as the scribbles in Otto’s notebook. Are the entries in the notebook (merely) transformative vehicles of content? Do they really play a role akin to that of gestures and not, as we have been told, a role analogous to that of inner mental states? It is time to comment on this topic.

To be sure, there is a rough sense in which Otto is using his notebook in a way which parallels the way Inga is utilizing her memory. After all, if this was entirely incorrect, Otto would not have been able to use the notebook to compensate for his memory loss. But the notebook does not instantiate intrinsic content (though see next section for further discussion). What does it do then? It serves as an external storage of information which, although not endowed with intrinsic content itself, is such that, when appropriately consulted, it generates in Otto mental states whose intrinsic contents make him well poised to accomplish the relevant cognitive tasks with which he is engaged—nothing more, nothing less. Insofar as coupling to the notebook generates intrinsically meaningful inner mental states in an orderly fashion conducive to the accomplishment of a cognitive task, the notebook plays a role analogous to that assumed by gestures in spatial reasoning or by motor adjustments in visual perception. The difference lies in the fact that the notebook constitutes a special case in which (a) the external operator is a locus of semantically interpretable symbol structures and (b) the internal transformation being effected is the result of interaction with those symbol structures. But the crucial fact remains that the contribution of the notebook to Otto’s intrinsic representational economy is transformative rather than instantiative.

Reflections and objections

I have now completed the outline of the position I am advocating, which may be described as moderate active externalism (MAE) since it affirms process externalism while rejecting the more radical thesis of mental states externalism. I believe that MAE offers a coherent alternative picture, significantly different from both the internalist orthodoxy and the rapidly expanding externalist avant-garde, and that it enjoys certain attractions which merit serious consideration. Nevertheless, being a middle-of-the-road position, I suspect that MAE will be scoffed by both camps—too radical for some, too conservative for others.

To illustrate, Rowlands considers a position similar to the one I advocate here only to reject it as being a “defensive, even reactionary response” (2006, 41), and as a “conservative view of what representation is” (Ibid, 48). Since I take no joy in being the village’s conservative, and since this paper is written, in the main, as a dialogue with radical active externalism, I would like to end on a critical note, examining several major arguments made by radical externalists in defense of MSE and pointing (somewhat briefly) to their shortcomings. While such critique will not do much good to my revolutionary credentials, it may help persuade some readers that perhaps this is not quite the revolution they were dreaming about.

In an attempt to avoid the detrimental consequences of the argument from intrinsic content, and to save MSE, radical externalists explored a variety of possible reactions, these include:
  1. 1.

    To deny that there is such a thing as intrinsic content in the first place.

  2. 2.

    To deny that there is a clear cut, or even meaningful distinction between intrinsic and derived content.

  3. 3.

    To argue that, appearances to the contrary, externally located “mental states” are, at times, endowed with intrinsic content.

  4. 4.

    To argue that even internal mental states are, at times, devoid of intrinsic content.

  5. 5.

    To argue that certain types of mental states—whether internal or external—need not be endowed with intrinsic content provided that they are appropriately connected to other mental states which are so endowed.

  6. 6.

    To argue that content cannot be assigned to a localized mental state since the vehicles of content are holistically distributed across brain, body, and world.21


I shall say next to nothing about the first two options. Although Clark announces his skepticism with respect to the distinction between intrinsic and derived content (response no. 2) on several occasions (e.g., 2005 and 2010a), he acknowledges the plausibility of an anticipated response to his skepticism and quickly moves to consider other arguments. Thus understood, Clark’s skepticism seems to serve more as a general safety net, something to fall upon when all else is lost, than a formidable ram with which to gore the opponent.22 Dennett’s (1987 and 1990) more radical claim that all content is extrinsic (response no. 1) has not, to my knowledge, been translated into a serious argument in favor of radical active externalism and has been criticized elsewhere (e.g., Aizawa and Adams 2005; Tecumseh Fitch 2007). This being the case, and since I find neither of these skeptical avenues convincing in the first place, I am content, for the time being, to leave things as they are.

Consider next response no. 6, which consists of the claim that since the vehicles of content are distributed holistically across brain, body, and world, content cannot be assigned locally to inner mental states (see, e.g., Clark 2005; Wilson 2004). Following the discussion in “Parity and mental states externalism,” I maintain that the suggestiveness of this claim is based on an equivocation, or a failure to discriminate, between instantiative vehicles of content and transformative vehicles of content. What does the work here for the radical active externalist is the fact that the combination of instantiative and transformative vehicles of content is, often, distributed across brain, body, and world (with the former being internally located and the latter externally located), but to defeat my assault on MSE, it has to be shown that it is the instantiative vehicles of content, in particular, which spread across boundaries. This, I submit, has not been established (however, for an intriguing proposal see Theiner 2011, chap. 5). In short, this objection to mental states internalism nourishes on the vagueness of the proverbial expression “vehicle of content” and its force is inversely proportional to the degree of exactness achieved in its analysis.

Some radical externalists, however, prefer to respond to the challenge by insisting that external variables can, at times, be loci of intrinsic content (response no. 3). For example, Rowlands (2006, 2009) argues that actions can instantiate content in much the same manner that thoughts do.23 Rowlands approaches the problem from a teleosemantic perspective. He assumes that mental states acquire their content in virtue of having been selected to perform certain functions (see Dretske 1988; Millikan 1989), and he reasons that just as inner mental states can acquire a normative representational status in virtue of possessing such selective history so can actions.

I deny the premise. Its popularity notwithstanding, I argued at length elsewhere (Shani 2007) that the idea that selective history (in either Dretske’s or Millikan’s sense) can be the source of representational normativity is seriously flawed (see also Christensen and Bickhard 2002). To give but a taste of the problem, consider one of Millikan’s stock examples, the tail splashing alarm signal of beavers. The teleosemanticist would argue that this gesture means danger because it has been selected for successfully triggering adequate danger response in beaver populations. But animals are no mechanical dummies. For this past success to have occurred in the first place, it must have been possible for some beavers to associate the signal with the imminence of unwelcome consequences (e.g., pain, injury, etc.), an association which was strong enough to motivate them to agitate, flee, and so on. Now, the point is that the association of the signal with such possible future consequences is already a primitive form of representation since it implicitly predicates the present situation, or environment, with such features as being dangerous, being flee-able etc. (see Barham 1996; Bickhard 1999; Shani 2006). Moreover, such primitive pre-selected representation is (already) normative because it internally indicates that something hazardous, something counterproductive to the animal’s natural urge for self-maintenance is imminent and that now is the time to do something about it (Bickhard 2000; Christensen and Hooker 2000).24 Thus, I deny that representational normativity can be adequately explained using the machinery of teleosemantics. Finally, notice also that, in orthogonal spirit to Rowlands’ proposal, the alternative sketched here explains the representational significance of the tail splashing action by reference to inner mental states, and does so precisely because these, alone, are instantiative vehicles of intrinsic content.25

The opposite strategy is to resist the argument from intrinsic content by arguing that even inner mental states are, at times, devoid of such content (response no. 4). This line is pursued by Clark (2005). Clark asks us to imagine Martians “whose biological memory allowed them to store, on demand, bit-mapped images of important chunks of text” (Ibid, 5) which they can later access upon recall. He then goes on to argue that “in that case, we would have no hesitation in counting these goings-on as part and parcel of Martian cognition,” and that “it is only skin-and-skull based prejudice that stops us extending the same courtesy to Otto” (Ibid.). But I think that, on this matter, Adams and Aizawa’s (2008, 49) response (modulated to the terms of the present discussion) makes perfect sense. Why should we accept the premise that the Martians’ passive inner images constitute mental states? In accordance with the discussion above, it is quite plausible to deny this premise precisely because the images are devoid of intrinsic content. It is perfectly coherent to turn the argument upside down and to say that, yes, such devices would be on par with Otto’s notebook but what this really means is that neither of them is a repository of mental states. The fact that the Martians’ images happened to be located inside the skull is inconsequential since it is not the topology as such that matters but, rather, the capacity to instantiate intrinsic content.26

Yet another line of defense of MSE pursued by Clark (2005) consists in arguing that although conventional encodings, whether they be located outside the head or inside it, are devoid of intrinsic content, this need not be detrimental to their status as mental states so long as they are appropriately connected to representations whose content is intrinsic (response no. 5). Combined with Clark and Chalmers’ (1998) contention that the entries in Otto’s notebook qualify as dispositional beliefs, this line of defense amounts to the claim that, provided that they are suitably connected to some intrinsically meaningful mentations, non-occurrent mental states need not exemplify intrinsic content.

But, as noticed by some critics (O’Brien 1998; Weiskopf 2008), genuine (including dispositional) beliefs and memories are rather different than Otto-style “beliefs.” The latter are inert and make no contribution whatsoever to the cognitive economy of an agent until the time comes when they are being recalled. By contrast, neurally constituted beliefs are dynamically active: they influence, and are being influenced by, other beliefs in a continuous process, forming an informationally integrated epistemic network in which beliefs are automatically updated in concert with each other (Weiskopf 2008). Consequently, genuine dispositional beliefs need not await intentional recall to become relevant: they may be activated by an associative train of thought, updated automatically, or impose constraints on the formation of other beliefs. Moreover, these behaviors are not merely material contingencies which can be downplayed; rather, they constitute an essential part of the cognitive role of beliefs (cf. Weiskopf 2008, 4).27

Some radical externalists are tempted to respond by arguing that the inertness in Clark and Chalmers’ example is accidental and can be avoided. Sutton, for example, argues that the entries in Otto’s notebook might be replaced by some “more dynamic new-media system” which is continually updating or appropriately reconfiguring itself and might even “come to highlight information or options aligned with Otto’s moods or emotions” (2010, 315).28 The most straightforward reply to this suggestion is to point out that, at the present, we simply do not have any trans-cranial system which can couple to our brains in such a manner that its states are adequately integrated with our inner mental states so as to be considered our very own beliefs. Thus, if this is the alleged rescue, then, as of now and until further notice, we must judge MSE to be no more than a wishful thinking. After all, we already have devices somewhat similar to what Sutton has in mind. Computers in marketing departments are busy making “educated guesses” regarding our reading habits, or our preferences with respect to potential mates, which they then address to us online, but we would never imagine mistaking such highlighted propositions for our own mental states.

But I think that the proposal suffers from a more fundamental problem. For such an external device to be epistemically aligned with our inner mental states (so that its states could be considered our mental states!) it must be able to anticipate, to interpret, and to respond properly to changes in our inner mental economy. But what principles could guarantee such proper synchronization? As Dreyfus (e.g., 1992) has argued relentlessly, it is ominously difficult to prespecify all the context dependencies which shape human cognition. The problem of creating an external medium which can keep up with the creativity of our inner cognitive economy seems to me to be tantamount to the problem of cracking the mystery of mind making—the Holy Grail of AI. This is by no means an easy feat, and it would be naïve to assume that we are anywhere near it.

But perhaps, we already have such intelligent external information storages, for aren’t we often systematically dependent on other people’s memory as a natural extension of our own memory (Wegner 1986; Smith and Conrey 2009; Sutton 2009)? While I do not deny the cognitive significance of this phenomenon, I do not think it justifies the conclusion that we literally store our mental states in other people heads. I think it is more natural to conclude that if I rely on my personal secretary (imagine!) to remember certain things for me, then the memories are mental states of the secretary while I am using those mental states as transformative vehicles with which to regulate my own mental states for my own benefit. It is, I think, the burden of the radical externalist to show why we should consider the secretary’s mental states as my own, how those states could belong simultaneously to the both of us, and why this proposal is superior to the less extravagant alternative I defend.


Moderate active externalism (MAE) is conveniently interposed between traditional individualism and the radicalism espoused by the majority of supporters of the extended mind. With the latter, it takes cognition to be an active engagement whose unfolding often ranges across brain, body, and world. Yet, with the former, it insists that mental states, those constitutive parts of cognition which serve as instantiative vehicles of content, are situated comfortably inside our heads. By affirming PE, it is able to accommodate many valuable insights and to preserve much of the attraction of the extended cognition movement, while, by denying MSE, it avoids its greater extravaganza. This, in itself, should be enough to make MAE an attractive option worthy of serious consideration. I hope I have done enough to show that it is, indeed, a coherent attractive alternative.


The term is due to Rupert (2004).


To my surprise, audiences and readers of this paper were, at times, inclined to doubt that one can make a meaningful distinction between cognitive processes and mental states and hence, derivatively, between PE and MSE.


This is hardly a coincidence given that PE and MSE are, in a sense, stronger than system externalism: For, one may argue, it is possible for a cognitive system to extend beyond the boundaries of the brain without it being the case that its cognitive states and processes are so extended. But if the latter are extended, then it follows that the very acts of cognizing are not brain-bounded (cf. Adams and Aizawa 2008, chap. 7).


This terminology is due to Sutton (2010).


While I am aware that valuable work has been done on such meta-theoretical issues (see for example Menary 2007; Noë 2004; Rowlands 1999), I am not utterly comfortable with the particular formulations on offer and would like to submit my own take on the problem.


It may be remarked that the point in identifying such a background conception is not to necessitate process externalism or, what comes down to the same thing, to rule out internalism with respect to cognitive processes. Rather, the goal is more modest, namely, to counteract an entrenched bias against PE and, by doing so, making the thesis more intuitive. I thank an anonymous referee for this journal for making me aware of the need to clarify this issue.


By speaking of “information” here, I do not, of course, refer to the standard technical sense in which this term is understood in information theory but, rather, to semantic information, information which is endowed with representational content. While I have my own opinions regarding how best to approach the intriguing issue of understanding, and explaining, semantic information (Shani 2006 and 2011), none of this is essential in the present context—the reader may fill in the blank with her own preferred account.


An especially relevant historical reference is John Dewey (1938), whose characterization of cognition as a form of inquiry captures much of what I understand in the concept of engagement as well as anticipates a great deal of the underlying rationale behind contemporary theories of embodied, situated, and action-oriented cognition (see Clancey 1997; Gallagher 2009). For explicit contemporary reference to the notion of cognitive engagement see Brighton and Todd (2009), and Smith (1999).


I give it a more detailed treatment elsewhere (Shani 2006; Shani under review); see also Bickhard and Terveen 1995; Hooker 2009; Moreno and Lasa 2003; Reed 1996.


Interestingly, although Uexküll seemed to have arrived at the notion of the functional cycle independently from Dewey, he was similarly motivated: like Dewey, he availed himself to this concept as a critical alternative to the reflex arc concept, which, like the American philosopher, he considered to be an overly mechanistic concept unsuitable for the description of biological and psychological phenomena.


Among protagonists of the present debate on extended cognition, it is probably Alva Noë (2004) who should be seen as following the footsteps of this tradition, although he endorses a version of extended cognition which, in my view, is stronger than what can be sustained merely on the basis of the argument from synergy.


I would like to stress in the clearest possible terms that although most of the sources I mention as influencing my own articulation of synergic coordination are anti-representationalist, the argument from synergy does not commit one to such anti-representationalism. As a matter of fact, precisely because the argument is predicated on a functional division of labor between the various components of an extended cognitive process, it leaves logical space for the idea that inner representations may make their own unique contribution to the process. Uexküll’s appeal to the notion of innenwelt in his articulation of the concept of the functional cycle is an early existential proof that representation and synergic coordination are perfectly compatible (see also Bickhard and Richie 1983; Rowlands 2006, for an illustration of how Gibsonian ideas can be assimilated into a representationalist-friendly framework).


Interestingly, Andy Clark, whose campaign for the parity principle (see next section) is an exemplar of first-wave extended cognition is also a progenitor of second-wave extended cognition (see Clark 1998; Sutton 2010).


Menary comes closer to acknowledging a real tension when he comments that the parity principle “has become something of an albatross around Clark’s neck”, adding: “I don’t think that C&C’s version of the parity principle is helpful as currently framed, nor as caricatured by the internalists” (2006, 333).


A small caveat: the realization which occupies us here occurs at the cognitive level—a level which is not directly concerned with such micro events as, say, the passage of a particular sodium ion through a sodium channel in a neuron in the primary visual cortex (cf. Adams and Aizawa 2008, 127). It is, however, concerned with macro components such as sequences of conscious experiences, large-scale neural patterns, behavioral patterns, sensory–motor contingencies, coordinated interpersonal behavior, environmental affordances, and so on.


Although I discuss, as well as endorse, the argument from intrinsic content, the reader must not assume that in doing so I closely follow Adams and Aizawa’s views on mental content (viz., what it is and how it is fixed). The issue is general enough and important enough to allow scholars of very different philosophical persuasions to agree on its significance in broad outlines.


By saying this, I do not mean to suggest that the meaning is fixed in one’s mind once and for all prior to the interaction with the text; this is clearly false. Rather, what I do mean is that only the inner mental states evoked during the course of the interaction are bearers of intrinsic content and that the text is imbued with meaning (for the interacting agent) courtesy of its association with such inner mental states.


Notice that the distinction, as I understand it, is between that which is capable of being meaningful in itself and that to which meaning must be assigned from without. Notice further that even if the contents of my thoughts involve representations whose meaning is conventional (e.g., Venn diagrams) the thoughts are still meaningful in themselves—they need not be interpreted in order to become meaningful. This issue seems to be confused by critics of intrinsic intentionality such as Clark (2005).


This echoes Adams and Aizawa’s (2008) complaint that the failure to deal satisfactorily with the problem of intrinsic content is a failure to respect a distinguished mark of the mental. However, unlike A&A, I do not direct this charge against extended cognition as such but, rather, specifically against MSE. Correspondingly, nor do I hold, as they do, that every cognitive part of a cognitive process must be endowed with intrinsic content.


I speak of a direct level of instantiation to emphasize that we are concerned with the closest theoretically relevant level of instantiation—the psychological or neurological—rather than with all the levels below it.


Yet another theoretical possibility is to profess an outright eliminativism with respect to mental states and mental content. But although this option, too, is explored by some advocates of extended cognition (e.g., Chemero 2009; Noë 2004; van Gelder 1995), it is not, in general, at the center of the debate between supporters and detractors of HEC, most of which—on both sides—accept some form of intentional realism. I shall therefore ignore the eliminativist doomsday solution thereafter.


In a recent paper, Clark concedes that “there is something quite compelling… about the idea that the notebook encodings are all conventional and derivative” (2010b, 87).


It may be born in mind, however, that Rowlands’ is but one possible way of making the case for the idea that external tokens are bearers of intrinsic content and that other defenses of this idea may not manifest the same shortcomings to which I point here. I thank a referee for this journal for stressing this point.


Note also that such representations allow for the possibility of error and of error detection (see e.g., Bickhard 1999) since the environment may fail to “cooperate”, that is, may fail to sustain the properties which the animal ascribes to it, and this functional failure can, in principle, be discovered by the animal (e.g., an alarm signal may be, and may be discovered to be, a false alarm).


To this, I would add that if you have a theory of content which implies that an action such as, say, stretching is a bearer of representational content in much the same sense that your thoughts are then this is a good indication that something went wrong with the theory.


Unlike Adams and Aizawa, I have no problem to admit that the bit-mapped images may well be “part and parcel of Martian cognition” (Clark 2005, 5) so long as it is understood that all that is mean by this is that these passive images play a constitutive part in some cognitive processes, to which they contribute as transformative vehicles of content.


Clark responds to this challenge by asking us to imagine a system in which inner integration and update occur not upon receiving new information but rather (in analogy with Otto’s case) “at the moment the outdated or otherwise affected information would have been called upon by some process of recall or action selection” (2005, 6). Such a system, he argues, would be behaviorally isomorphic to us and it is a mistake to deny its inner states the status of beliefs. But even if we grant Clark this outlandish scenario, the two systems are far from being cognitively isomorphic since, for example, in Clark’s imaginary system, there seems to be no room for important cognitive phenomena such as belief activation due to free association, or the contribution of non-occurrent beliefs to an ongoing constraining of cognitive processing.


To be fair to Sutton, I must stress that I use his example in a somewhat different context than he does and therefore that I do not direct my argument against him but, rather, against the general spirit of this kind of response.



This work was supported by a grant from Kyung Hee University in the year 2011. An early draft of the manuscript was presented at the Interactivist Summer Institute in Syros, Greece (July 30, 2011) as well as at the Korean Society of Analytic Philosophy (April 28, 2012). I thank the participants for their helpful comments and for their encouragement. I would also like to thank Alex Levine and Hyundeuk Cheon for reading an early version of the manuscript, Andy Clark and Dan Weiskopf for helpful personal communication, and two anonymous referees for this journal for their tremendously helpful comments.

Copyright information

© Springer Science+Business Media B.V. 2012