by Anthony Chemero & William Cordeiro
Scientific and Philosophical
Studies of Mind
In a way that is rarely even attempted, and even more rarely actually pulled off, Susan Hurley, in her book Consciousness in Action, brings scientific ideas into contact with mainstream philosophy. It is not at all unusual for empirical results from cognitive science, psychology, and neuroscience to be raised in discussion of issues in philosophy of science and philosophy of mind--Dennett and the Churchlands, for example, have been doing so for years. But Hurley attempts to draw empirical results even closer to the center of philosophy, using them to make points about metaphysics and epistemology more broadly, especially Putnam’s Twin Earth cases. We are very fond of Hurley's book, and we agree with nearly all of her conclusions. We do think, though, that there are two important cases where Hurley has misunderstood scientific work. First, we think she misunderstand dynamical systems theory; second, we think her criticism of ecological psychology is misplaced. In neither case do these misunderstandings derail Hurley’s overall project--indeed, the former of them makes her conclusions all the more plausible. We consider them in order.
A. On Dynamical Systems
A significant portion of Part II of Hurley's book is devoted to arguments against what she calls the Input-Output Picture. According to the Input-Output Picture, personal-level mental states match up with sub-personal physical vehicles which mediate causal interaction with the world. The idea is that sub-personal sensory inputs are the (causal) stuff of perceptions, and sub-personal motor outputs are the (causal) stuff of intentions. Imagine looking inside your refrigerator. Light reflected from items therein impacts your eyeball and produces an upside-down image on your retina. This is the causal input from the environment. Add a small amount of peripheral, sub-personal processing and you perceive the items in your refrigerator. (Notice input and perception lining up nicely.) You then form beliefs about the things in your refrigerator: "There is some pizza; there is a chocolate cake; there are some carrots; there is an apple; there is a jar of mustard . . ." These beliefs are combined with other beliefs (about nutritional value and calorie content, about your personal likes and dislikes, etc.) and desires ("I want to become less hungry; I would prefer not to gain weight; I prefer chocolate to carrots . . ."). Then, a plan of action is formulated, and an intention to follow it is formed ("I will eat an apple.") With the intention comes a (sub-personal) motor plan, which in turn causes your (sub-personal) motor behavior, which implements your personal-level action. (Notice how motor plan and motor output line up with intention and action, respectively). All very neat.
This Input-Output Picture is closely allied with what Hurley calls the Classical Sandwich, the idea that cognition or thought is the filling located both spatially and conceptually between perception and action, which serve as the bread. In the case just described, the belief formation and decision making is the cognition, residing happily between perception/input and action/output.
Hurely's aim, and it's an aim that we applaud, is not just to cast doubt upon the Input-Output Picture and the Classical Sandwich, but to demolish them with an onslaught of empirical evidence that, philosopher’s fantasies aside, things aren't so neat in the real world. To this end, Hurley marshals dozens of arguments (we lost count) that perceptual and intentional content are constitutively linked <1> ; that is, she describes case after case in which sensory input and motor output (or the physical state of the external environment) are held constant, but a change in intentional content leads to a change in perceptual content (or vice versa). Put more simply: she presents a series of instances from psychological literature in which what I'm trying to do changes what I experience and/or what I experience changes what I try to do, even when my bodily motion and sensory irritations remain the same. An example (about which we will have more to say later) is mild paralysis of the muscles that move the eye. As Hurley, following Gallistel 1980, describes the situation, a subject with mildly paralyzed eye muscles tries to glance to one side, say the right. When she does so, and when the eye fails to move because of the paralysis, she experiences the world as jumping to the left. Compare this to the case in which the subject does not try to move her eye. In both cases, the light impinging on the eye is the same, but the experience is different: when the subject tries to move her eye to the right she experiences the world jumping to the left; when she does not try to move her eye at all, she experiences the world as standing still. So her intention to move her eye changes her perceptual experience: perception and intention are constitutively linked.
How does showing constitutive linkage between perception and action damage the Input-Output picture and the Classical Sandwich? Well, if intention to act changes what we perceive without also changing the input (as happens when paralyzed eye subjects try to move their eyes), then sub-personal causal input (light impinging on the retina, etc.) cannot be sole determinant of perceptual content <2> : sub-personal input and personal perceptions don’t line up. So the Input-Output Picture can’t be correct. And if the Input-Output Picture fails in this way, the Classical Sandwich goes down with it. If the perceptual and intentional content are constitutively linked, then they are not properly speaking separated from one another. Then it makes no sense to think of them as peripherals that cognition sits between--there’s just no ‘between’ there in which cognition can sit.
As noted, Hurley provides dozens of arguments like the one just outlined against the Input-Output Picture and the Classical Sandwich. Each of them arises from a similar locus, really: Hurley uses scientific evidence to argue that the sub-personal is a tangled causal web, with motor commands affecting sensory states and sensory states affecting motor commands. As she puts it, the sub-personal displays lots of dynamical feedback loops, making it difficult to draw the line between a sub-person and its environment. This, many will have noticed, smacks of the dynamical approach to cognition, and not accidentally: Hurley derives her understanding of the sub-personal from work by Thelen and Smith (1994), Kelso (1995), and the papers collected in Port and van Gelder (1995), among many others. Drawing on this dynamical cognitive science, Hurley says that a sub-person is a 'dynamic singularity', which is
a continuous and complex dynamic system centered on an active organism, with feedback loops that may have external as well as internal orbits. On this view, to explain input-output relations, we need to consider not just what functions outputs are of inputs, but also what functions inputs are of outputs, where the latter may be determined in part by external states… (p.333)
Two quick points about this view of sub-persons. First, this view of the sub-personal implies a version of externalism, but not of the familiar "meanings ain't in the head" type. Here, we have not just externalism about content as in Putnam and Burge, but an externalism about vehicles as well: not just what determines content is potentially outside the skin of the organism, but also what bears content and carries the causal oomph when the organism perceives and acts. Second, and this point is necessitated by the first, not just any part of an organism's environment is part of a dynamic singularity, and hence part of vehicles of content. The dynamic singularity is just those parts of the environment required to causally explain an organism's input-output relations, and this may include both internal feedback loops (as with efference copy) and interactions with the environment.
And here's the rub. We think this is a misunderstanding of dynamical cognitive science. The problem is that the usual notions of 'causal', 'input', and 'output' are incompatible with dynamical systems theory and dynamical explanation (and so with Hurley's notion of 'dynamical singularities'). To make this clear, we will review a few points about a rather familiar dynamical system: the Watt governor, as described in Tim van Gelder’s landmark 1995 paper "What might cognition be, if not computation?" In that essay, van Gelder recommends that the mathematical model of the operation of the Watt governor, a mechanism for controlling the speed of a steam engine, be taken as a prototype model for cognitive science. The Watt governor controls the speed of a steam engine as follows.
It consisted of a vertical spindle geared into the main flywheel so that it rotated at a speed directly dependent on that of the flywheel itself. Attached to the spindle by hinges were two arms, and on the end of each arm was a metal ball. As the spindle turned, centrifugal force drove the balls outward and hence upwards. By a clever arrangement, this arm motion was linked directly to the throttle valve. The result was that as the speed of the main wheel increased, the arms raised, closing the valve and restricting the flow of steam; as the speed decreased, the arms fell, opening the valve and allowing more steam to flow. The engine adopted a constant speed, maintained with extraordinary swiftness and smoothness in the presence of large fluctuations in pressure and load. (van Gelder, 1995, p. 349)
In van Gelder’s proposed dynamical explanation of the governor, its operation is described mathematically. Just as Newton did in his descriptions of the physical world, the behavior of the system of interest is observed, and mathematical equations that describe that behavior are found. In the case of the Watt governor, the instantaneous change of the arm angle when the steam engine is disconnected from the throttle valve is described by the following equation:
where q is the angle of the arms, n is a gearing constant, w is the speed of the engine, g is the gravitational constant, l is the length of the arm, and r is a friction constant (see van Gelder 1995; this paragraph and the next follow van Gelder’s discussion of the Watt governor closely). This equation describes the instantaneous acceleration of the arm angle, given the instantaneous arm angle. Only q , the arm angle, is a variable in this equation; n, w , g, l and r are parameters, hence remain constant and fix the dynamics of the system. This equation is completely general in that it gives the acceleration for any arm angle. Solutions to this equation specify a state space, and trajectories through this space can be used to predict future instantaneous accelerations and arm angles, given the current values of these variables.
The governor’s behavior when connected to the throttle valve can be described by the following, more complicated equation:
where t is the setting of the throttle valve. This equation, also perfectly general, describes the instantaneous change of the speed of the engine w as a function of the throttle setting, which is itself a function of the arm angle q . Just as w is a parameter in the former equation, q is a parameter in this equation, so we should think of these two dynamical systems as coupled. Any change in the arm angle q changes the total dynamics of the system that describes the speed of the engine w , in which it is a parameter; and any change in the engine speed w changes the total dynamics of the system that describes the change of the arm angle q , in which it is a parameter. Indeed, given the closeness of the relationship between the two systems, for many purposes, it is best to just think of them as just one system (governor plus engine).
Note that there is no reference to representation, computation, or teleology in this explanation of the Watt governor’s behavior. Furthermore, and this is crucial to what we will say here, there is no reference at all to causation, inputs, or outputs. And with good reason: these concepts (at least as we usually understand them) are not involved in dynamical systems theory; some would even argue that they are anathema to dynamical systems theory. Yet Hurley refers to these concepts again and again in her characterization of the sub-personal.
As van Gelder (1998; see also Bechtel 1998, Chemero to appear) points out, dynamical cognitive science is an attempt to fulfill Hume’s goal of a scientific psychology similar to Netwon’s mechanics--a psychology in which cognition would be explained by mathematical laws, just as the material world is explained. A key feature of such a psychology, as anyone familiar with Hume’s philosophy is aware, would be that it avoids talk of causation. And dynamical cognitive science does just this. One of the most striking and important features of Newtonian physics is that the sort of covering law explanations that were provided by Newton’s mechanics obviated any need for teleology in physical explanation. A covering law explanation is one in which a fact is explained if and only if a statement of that fact is deduced from statements, at least one of which is a general scientific law. Note also that according to the covering law model of explanation (and dynamical models of cognition), prediction is the dual of explanation: a fact is explained if and only it could have been predicted. (See Hempel 1965). In providing a fully general set of mathematical laws for physics, Newton sidestepped speculation about Aristotelian final (=teleological) causes, taking his laws of motion as axioms not in need of further explanation. Indeed, Netwonian mechanics does without causes of any kind. So too does explanation in dynamical cognitive science. But Hurley’s conception of dynamical singularities is causal, and is therefore incompatible with the dynamical cognitive science upon which she draws in its description.
Hurley’s conception of the sub-personal as causal is, as far as we can tell, detachable. That is, she could simply admit that dynamical explanation is non-causal, and in so doing, repair this inconsistency in the notion of a dynamic singularity. For example, one of the most interesting features of her understanding of sub-persons is that they are "leaky" (see Clark 1997) in that causal processes that are parts of the dynamic singularity do not respect the boundary of the skin. But rather than describing this fact in terms of causal connections across the skin, one might say that the one dynamical system at issue has parameters both inside and outside the organism. That is, we might explain the behavior of the agent in its environment over time as coupled dynamical systems, using something like the following equations, from Beer 1995 (see also Beer 2000):
where A and E are continuous-time dynamical systems, modeling the organism and its environment, respectively, and S(xE) and M(xA) are coupling functions from environmental variables to organismic parameters and from organismic variables to environmental parameters, respectively. As Beer notes, it is only for convenience that we think of the organism and environment as separate; in fact, they are tightly coupled (variables in each serve as parameters in the other), and so best thought of as comprising just one system, U. Such a dynamical model would allow us to describe the organism’s (that is, the sub-person’s) behavior in its environment, and would capture Hurley’s laudable vehicle externalism: the vehicle must include all of the aspects of the system at hand, including those that are outside the organism’s skin. But rather than describing the way external (and internal) factors cause changes in the organism’s behavior, such a model would explain the way U, the system as a whole, unfolds over time. This, we take it, is just the sort of thing that Hurley has in mind in her discussions of dynamic singularities. So we think she can take the non-causality of dynamical systems explanation in stride.
But conceiving of sub-persons this way fundamentally alters our notions of input and output, and this does seem to affect Hurley’s arguments against the Input-Output Picture and the Classical Sandwich. How does this thoroughly dynamical conception of sub-persons alter the notions of input and output? Well, if the dynamic singularity is modeled as U the organism-world system, everything of relevance will already be part of said singularity. And if that is the case, there is nothing outside the system for input to be from, nor is there anything for output to be to. When we understand sub-persons as organism-world dynamical systems, "external" environmental features that were originally thought of as causing sensory input to the organism are instead simply part of the overall system, and are modeled as system parameters. That is, instead of causally setting the organism’s internal state at time t via sensory input, aspects of the environment are seen as influencing the overall shape of ongoing behavior. To view sub-persons as dynamic singularities, and we think Hurley see this (consider her criticisms of temporal atomism in Chapter 1), one must step away from the input-cognition-output cycle, toward a view of cognition gradually unfolding over time. This requires a rather different conceptualization of inputs and outputs. From a dynamicist point of view, inputs are neither states that result from instantaneous causal contact with the environment, nor that instantaneous contact itself. Instead, an input it is a factor within the system that remains constant for some time, and influences the system’s overall behavior. An input, that is, is something that is modeled as a parameter, and it influences the way factors modeled as system variables change over time. The point, then, is that taking a dynamical approach to the sub-personal, as Hurley wishes to, fundamentally alters inputs and outputs--so much so, in fact, that they Input-Output Picture’s understanding of them no longer makes sense.
To see this, consider Thelen, Schöner, Scheier, and Smith’s (to appear) model of the infant reaching. This model is designed specifically to replicate the well-known "A not-B Error", originally discussed by Piaget. The A not-B Error is a typical error made by 7-12 month-old infants who are shown a desirable toy, which is then placed in one of two boxes, A. After seeing the toy placed in box A several times, and being allowed to retrieve it from Box A, the infant is shown the toy being placed in box B instead. This is when the infant makes the A not-B Error: he will look for the toy in box A, despite the fact that he has seen it put in box B. Piaget thought that the problem is in the infant’s mind: the infant’s concept of "objects" is insufficiently developed. Thelen et al, on the other hand, focus on the infant’s activity, developing a dynamical model of what they call the movement planning field, which is an explanation of the way the infant’s inclination to reach in particular directions changes over time. This model captures not just the standard cases of A not-B Errors, but also the apparently exceptional cases that have plagued earlier attempted explanations. (For more on this, see Thelen et al to appear). The mathematical model is rather complicated and the mathematical details do not matter for our purposes. What does matter is that the changes of the movement planning field (which models the infant’s inclination to reach in particular directions) are a function of three things:
We are interested in inputs specifically here. Thelen et al specify three types of input:
The key to understanding the way these inputs differ from Input-Output Picture inputs is that these inputs are not merely instantaneous causal pushes from things outside the system. Instead, they are parts of the dynamic system, parts of the infant-environment system. They are modeled as parameters in the system: factors that shape the overall change of the system’s behavior, but remain relatively constant themselves. In this case, all these factors but the direction-of-reach inclination will remain constant for the duration of an experimental trial (from when a toy is put in a box, until the infant reaches for it).
In the Thelen et al model, as in most dynamical models, there is simply nothing like the Inputs and Outputs of the Input-Output Picture. This is the case for two reasons, both of which Hurley is aware: first, dynamical understandings of sub-persons reject temporal atomism and the input-cognition-output cycle, which entails a rejection of instantaneous inputs to or outputs from the organism; second, since dynamical systems and dynamical singularities are "leaky" in that they extend to include parts of the environment, the factors that would be considered sources of inputs of the Input-Output Picture variety are actually parts of the system, so there is no need for input to come from them to the system.
How does this affect Hurley’s arguments against the Input-Output Picture and the Classical Sandwich? Well, mostly, it makes them superfluous. The Input-Output Picture and Classical Sandwich are bad understandings of the mind, but not for the reasons Hurley gives so meticulously over several hundred pages. These ideas are faulty because if sub-persons are well understood in terms of dynamical systems theory, then the ideas of input and output (as understood by the Input-Output Picture) don't apply to them. Dynamic singularities include more than just the organism. There are often loops that go out into the external environment (as well as internal loops). These loops are part of the (sub-personal) dynamic singularity. But if the singularity is the whole dynamic system, the traditional ideas of inputs and outputs don’t make sense. Inputs from what? Outputs to what?
B. Ecological Psychology
We think Hurley misunderstands Gibson. In her Chapter 10, "Alternative Views of Perception and Action", Hurley argues that Gibsonian ecological psychology (1979), like her preferred "Two Level Interdependence Theory", views cognition as "loopy", with perception affecting action, and action leading to alterations in what we perceive. But, she claims, ecological psychology views the connection between perception and action as being merely instrumental. That is, action affects perception by changing a perceiver’s location in the environment, thereby changing what the perceiver sees. Her view, and other correct views, agree that perception and action are linked, but suggests that the link between them is constitutive and not just instrumental: changing intentions affects perceptual content, even when sensory input is the same. Hurley represents this difference between her account and other "alternative" views of perception and action <3> , with a box diagram like Table 1.
To place ecological views of the relations between perception and action into a box designating those relations as merely instrumental belies a fundamental misconstrual of Gibsonian psychology. While it is true, as Hurley points out, that Gibson rejects the idea of efference copy and thus has difficulty accounting for the illusion that the world "jumps" when someone with paralyzed eyes intends for their eyes to move, this rejection is partly due to the fact that the illusion would be confounded as soon as any part of the individual was within its optic field. Proprioceptive information about the relative position of the face and body would quickly correct the illusion since the success of the illusion would now require the additional illusion that the individual’s face and body moved without the eyes; this further illusion would be well-nigh impossible to pull off. Gibson held that such information from proprioceptive feedback was sufficiently ubiquitous in an organism’s normalized relations to its environment to allow it to make the self/world distinction. In other words, the illusion would be a short-term effect that only lasts while the organism is in the process of re-normalizing its relationship with its environment.
Furthermore, Gibson (see Gibson 1982) rejects the idea of "basic" intentions which Hurley relies upon here, recognizing a range of actions between the voluntary, fully volitional sort and brute, machine-like reflexes. The attempt to apply this dichotomy to ecological psychology just does not make sense because part of the point of ecological psychology is that there is no exact line--but rather a continuity--between the voluntary, intentional "personal" level and the "sub-personal" processes which occur involuntarily according to causal machinations. Ecological psychologists wish to avoid the unperspicuous distinction between the voluntary and involuntary because all actions are controlled by the organism at some level. The farther one gets from philosophic table-talk, the less such artificially tidy dichotomies make sense: the relevant question about an action is a question about what aspects of the environment is the organism using in order to control its behavior, not anything about the level of decision making¾ let alone linguistic decision making¾ taking place "within" the organism. Rather than attempt to parcel out which activities (or even what parts of which activities) conscious decision making is responsible for and which activities are controlled by lower-level causal reactions, ecological psychologists seek to study an organism’s total repertoire of behavior as it is controlled by the entire organism. Such control may come at many different, interacting and overlapping, levels; by focusing on the organism’s behavior as it relates to its environment, Gibsonians bypass the fruitless gerrymandering of actions engendered by abstracted preconceptions. They look at everything an organism does in its normal environment and try to understand what accounts for such a delicate control of an organism’s actions in response to changes in its natural environment.
For example, if the muscles which control the paradigmatically involuntary movements of the eyes (saccades, focusing, etc.) were also paralyzed, then a different sort of illusion would occur (see Owens and Reed, 1994). In this case, according to Hurley’s formulation, neither input and output nor "basic" level intentions would change, yet there would still occur a difference in perception. Thus, it is not intentions per se which affect perceptions but the holistic relations involving how the organism is situated in relation to its environment and the inter-relationship of the organism’s abilities in that environment.
Most importantly, however, it is clear that ecological psychologists do not regard the relation between perception and action as merely instrumental: the very things they propose animals perceive are affordances, or opportunities for action. Affordances are neither features of the external world nor internal representations; they are neither objects nor mental phenomena. Instead, they are the relational properties between the available resources in an environment and an organism’s abilities to utilize those resources. Thus, they are specific to an organism’s abilities without being insulated inside an organism’s head. They are real, objective features that depend on both environment and organism: in short, they are what the environment allows the organism to do. Therefore, if it is affordances that an organism perceives, as Gibsonians claim, then perception inherently involves action because they very kinds of things perceived are relative to the actions an organism can perform. We stress that this is, moreover, a constitutive (i.e., non-instrumental) link between perception and action: an organism’s perceptions are in part constituted by its behavioral repertoire. The claim, to make it more explicit, is that an animal doesn’t so much as see a ‘tree’ (as we call it in English) as it perceives an opportunity to hide from a predator or a nice mid-day snack. Because the ontological make-up of what is perceived (that is, affordances) is structured in terms of an organism’s normal behavior in this way, it would be wrong to characterize the ecological psychologists’ approach as positing a non-instrumental relationship between perception and action.
According to ecological psychologists, animals perceive their environment in terms of their abilities to act in it; in a very direct and noninstrumental way, perceptions are constituted by actions rather than trivially affected through the mere means of action. Seeing an object we linguistically designate as a ‘hammer’ is dependent on perceiving the affordance of something like "pounding-nails-in-ability". The structure of the perception is such that it is defined by the available normal actions of an animal. Indeed, ecological psychologists have been arguing for years that perception and action cannot be studied separately but are instead interdependently--and noninstrumentally--woven together such that what we perceive is a function of what we can do.
The overall point is that Hurley is right to wish for an understanding of the mind that both rejects the perception-action cycle and has constitutive connections between perception and action. It is just that ecological psychology, along with motor theories, dynamical systems theory, control systems theories and Hurley’s own two-level interdependence view, is such an understanding of the mind.
These two complaints (Hurley’s misunderstandings of dynamics and of ecological psychology) are in some sense nit-picking. This is especially the case given our overall agreement with Hurley’s main points: we think she is right that temporal atomism, the Input-Output Picture and the Classical Sandwich ought to be rejected. And we think that, generally, she wants to reject them for the right reasons.
<1> Hurley speaks of instrumental versus non-instrumental dependence of perception on action, or of action on perception. Where she says there is ‘non-instrumental’ dependence of action (perception) on perception (action), we will sometimes talk of ‘constitutive links’ between perception and action.
<3> Note that Hurley is using the word ‘alternative’ not to indicate views that are different from her own, but rather in the way those in the music business now use that word, when they lump punk rock, garage rock, folk, and electronic dance music into one category. That is, by ‘alternative’ here Hurley means ‘different from the mainstream’. Her own view is also ‘alternative’ in this sense.
Bechtel, William. (1998). "Representations and cognitive explanations." Cognitive Science 22: 295-318.
Beer, Randy. (1995) "Computational and Dynamical Languages for Autonomous Agents." In Port and van Gelder 1995.
Beer, Randy. (2000). "Dynamical Approaches to Cognitive Science." Trends in Cognitive Science, 91-99.
Chemero, Anthony. (to appear). "Anti_representationalism and the Dynamical Stance." Philosophy of Science.
Clark, Andy. (1997). Being There. Cambridge: MIT Press.
Gallistel, C.R. (1980). The Organization of Action: A new synthesis. Hillsdale: Erlbaum.
Gibson, James J. (1979). The Ecological Approach to Visual Perception. NewJersey: Houghton Mifflin.
Gibson, James J. (1982). Reasons for realism. E. Reed and M. Jones (eds.) Hillsdale: Elrbaum.
Hempel, Carl. (1966). The Philosophy of Natural Science. Englewood: Prentice-Hall
Hurley, Susan. (1998). Consciousness in Action. Cambridge: Harvard University Press.
Kelso, J. Scott. (1995) Dynamic Patterns. Cambridge: MIT Press.
Owens, D. Alfred and Reed, Edward (1994). "Seeing Where We Look." Behavioral and Brain Sciences 17, 271-2.
Port, Robert. and van Gelder, Timothy (eds.). (1995). Mind as Motion. Cambridge: MIT Press.
Thelen, Esther and Smith, Linda B. (1994). A Dynamic Systems Approach to the Development of Cognition and Action. Cambridge: MIT Press.
Thelen, Esther, Schoner, Gregor, Scheier, Christian, and Smith, Linda (to appear). "The Dynamics of Embodiment." Behavioral and Brain Sciences.
van Gelder, Timothy. (1995). "What might Cognition be if not Computation?" Journal of Philosophy, 91, 345-381.
van Gelder, Timothy. (1998). "The Dynamical Systems Hypothesis." Behavioral and Brain Sciences, 21, 1-14.