<<

. 29
( 62 .)



>>

narrow content becomes inexpressible. Two responses might be made to
this second objection.34
First, one might concede that, strictly speaking, narrow content is in-
expressible, but then point out ways of sneaking up on it. One might do
so by talking of how one can “anchor” narrow content to wide content, or
of how to specify the realization conditions for a proposition. But these
suggestions, despite their currency, seem to me little more than whistling
in the dark, and the concession on which they rest, fatal. All of the ways
of “sneaking up on” narrow content involve using wide contents in some
way. Yet if wide content is such a problematic notion (because it is not
individualistic), then surely the problem spreads to any notion, such as
snuck-up-on narrow content, for whose intelligibility it is crucial.35
If narrow content really is inexpressible, then the idea that it is this
notion that is central to psychological explanation as it is actually prac-
ticed, and this notion that does or will feature in the natural kinds and
laws of the cognitive sciences, cannot reasonably be sustained. Except
in Douglas Adamesque spoofs of science, there are no sciences whose
central explanatory constructs are inexpressible. Moreover, this view
would make the claim that one arrives at the notion of narrow content
via an examination of actual explanatory practice in the cognitive sci-
ences extremely implausible, since if narrow content is inexpressible,
Individualism and Externalism in Cognitive Sciences
174

then one won™t be able to ¬nd it expressed in any existing psycholog-
ical theory. In short, the idea that snuck-up-on narrow content is what
cognitive science needs or uses constitutes a lapse back to the cognitive
science gesture.
Second, it could be claimed that although it is true that it is dif¬cult for
common-sense folk to come up with labels for intentional contents, those
in the relevant cognitive sciences can and do all the time, and we should
defer to them. For example, one might claim that many if not all of the
representational primitives in Marr™s theory, such as blob, edge, and line,
have narrow contents. These concepts, like many scienti¬c terms, are
technical and, as such, may bear no obvious relationship to the concepts
and terms of common sense, but they still allow us to see how narrow con-
tent can be expressed. One might think that this response has the same
question-begging feel to it as does the claim that our folk psychological
states are themselves narrow. However, the underdetermination of philo-
sophical views by the data of the scienti¬c theories, such as Marr™s, that
they interpret remains a problem for both individualists and externalists
alike here. As my discussion of exploitative representation and wide com-
putation perhaps suggests, my own view is that we need to reinvigorate the
ways in which the computational and representational theories of mind
have usually been construed within cognitive science. If this can be done
in more than a gestural manner, then the issue of the (in)expressibility
of narrow content will be largely moot.


8 locational versus taxonomic externalism
Our extended treatment of Marr™s theory of vision over the previous four
sections has distinguished two different ways of thinking about exter-
nalism that make contact with computational approaches to cognition:
taxonomic and locational externalism. A cluster of related paradigms
within contemporary cognitive science that go under various names “
situated cognition, embedded cognition, distributed cognition, improvi-
sational cognition “ can be understood in terms of either or both of these
forms of externalism. In this section, I shall discuss speci¬c accounts of
cognition within these paradigms. In particular, I shall argue that these
accounts of cognition posit wide computational systems, and thus are
committed to the idea that cognition is locationally externalist.
Coincident with the rise of connectionist models of cognition in which
cognitive representations can be distributed across a number of nodes,
rather than localized in few of them, has been the idea that cognition
Representation, Computation, and Cognitive Science 175

can be distributed either across agents or across agents and their envi-
ronments. In the former case, cognition is construed interpersonally, in-
volving social and cultural mediation between people, while in the latter
case it is seen as extending beyond the individual into her environment.
Since the relevant part of the environment is often a cultural product,
such as a cognitive artifact, in both cases there are cultural and social
dimensions that are integral to cognition.
Edwin Hutchins™s work on seafaring navigation exempli¬es a dis-
tributed approach to cognition of both types. While Hutchins is critical
of some of the turns that the computational approach to cognition has
taken, his own views have been developed within an overarching compu-
tational framework, one with interpersonal and artifactual dimensions.
Hutchins argues that technology should not be thought of simply as a way
of augmenting individual cognitive capacities, however, but as a means
of changing the nature of the representational spaces or media in which
computations are performed. Computations are not simply performed
in the head of the individual, but occur in interpersonal activities that
make use of recent (and not so recent) cognitive artifacts. The locus of
computation is in the beyond-the-individual world.
Hutchins brings out the interpersonal dimension to the cognitive pro-
cesses involved in navigation when he says, in his introduction to Cognition
in the Wild, that he hopes

to show that human cognition is not just in¬‚uenced by culture and society, but
that it is in a very fundamental sense a cultural and social process. To do this I will
move the boundaries of the cognitive unit of analysis out beyond the skin of the
individual person and treat the navigation team as a cognitive and computational
system.36

Here the unit of cognition is not the individual but the navigation team of
which the individual is a part. Both the social organization of the team and
the relationships between its members and various cognitive artifacts (for
example, the alidade, the bearing log, phone circuits, the hoey, the chart,
the fathometer) serve as the cognitive architecture of this larger cogni-
tive unit. For example, consider the ¬x cycle, which plots the position
of the ship (output) given two lines of position as inputs. This cycle in-
volves the generation, transformation, and utilization of representations
in many individuals and many cognitive artifacts. There is no one place,
in particular, no one individual, in which this process is implemented.
Hutchins™s study of navigation does tell us a lot about the social distri-
bution of cognitive tasks in navigation, showing how modern navigation
Individualism and Externalism in Cognitive Sciences
176

involves exploiting social structures and relationships. And perhaps this
even provides an example of group-level cognition, a topic we will discuss
in detail in Part Four. But Hutchins also highlights how individual-level
cognition often extends beyond the boundary of the individual. For ex-
ample, he does so in emphasizing how integral the use of representational
artifacts “ from calendars, to maps, to alidades, to hoeys “ are not only to
team-level cognitive tasks, but also to component, individual-level cogni-
tive tasks “ such as reading the alidade, writing an entry on the bearing
log, or adjusting the hoey. Such cognitive processes are realized in wide
computational systems, where these systems constitute ways of extending
an individual™s cognitive capacities. Navigation involves a wide range of
locationally externalist cognitive processes.
A second example that exempli¬es this perspective is Dana Ballard™s
work on animate vision in general and on the role of deictic coding
in cognition in particular. Here the central idea is that cognition often
involves rapid updates of information from one™s environment in an un-
planned or improvised manner, where this improvisational cognition in-
volves adjusting one™s body, particularly through eye movements, to take
advantage of information that is stored in the environment and thus does
not need to be computed or stored by the individual. For example, vision
involves intense, repeated causal interactions with an environment via
rapid saccadic ¬xations, which shifts the burden of the computational
load from inside the head to the head-world. Inside-the-head representa-
tions are computationally expensive to construct and to maintain, hence
the idea that they are constructed only when necessary, with the repre-
sentational slack being taken up by interaction with the world.
Here we have an instance of Andy Clark™s previously discussed 007
principle “ know only as much as you need to know to get the job done,
where to “know” something is to internally represent it in some way. Ex-
ploitative representation reduces the informational load that individuals
bear through exploiting world-mind regularities, and in the case of an-
imate vision this exploitation makes use not simply of such regularities
but the ease with which our bodies can be adjusted to utilize them.
The central novelty of Ballard™s approach to vision is to combine the
concept that looking is a form of doing with the claim that vision is com-
putation. As we saw in section 6, although the ecological psychologist
James J. Gibson recognized the ¬rst of these points long ago, his views
were developed in direct opposition to the idea that vision was inferential
or computational. Ballard integrates these two points by introducing the
idea that eye movements constitute a form of deictic coding in that they
Representation, Computation, and Cognitive Science 177

are behaviors that orient visual attention and ¬xation in ways that allow
perceivers to exploit the world as an external storage device. Eye move-
ments are a type of pointing device, a type of “doing-it- where-I™m-looking”
strategy, that means that perceivers need not copy all of the information
in a scene in order to use it in guiding further action.37
Ballard applies this model of cognition not only to vision but also
to attention, memory, and action. Despite the name “deictic coding,” it
represents an approach to cognition that involves exploiting rather than
encoding beyond-the-head environments. Here too, I would suggest that
the idea of a wide computational system, and of cognitive processes being
locationally wide, aptly describe the program of animate vision.
To take a third and ¬nal example, consider the “intelligence without
reason” approach to robotics championed by Rodney Brooks that concen-
trates on developing behavior-based systems. Brooks makes no attempt
to model how people behave, and is explicit that he is not engaged in
any sort of “cognitive modeling” that captures (even in part) the cogni-
tive processes that people instantiate. His goal is to build robots, which
he calls “Creatures,” that behave intelligently in natural environments,
rather than to mimic human performance on some speci¬c task (for ex-
ample, decision making, block moving, question answering) in an arti¬-
cially restricted domain (for example, chess, the blocks worlds, restaurant
scripts, respectively). This is achieved by a subsumption architecture that
builds more complex layers on top of less complex layers, with each layer
achieving some action-centered goal, such as avoiding objects, wander-
ing, or exploring its environment. Layers themselves are composed of
simple ¬nite-state machines that achieve speci¬c behaviors.
One of Brooks™ recurring themes is that the world is its own best model,
and so there is no attempt to have his Creatures encode features of the
world in order to negotiate successfully in it; rather, they exploit those
features. But his Creatures are not relying on symbolic aspects of their
environments or external representations. Indeed, Brooks has claimed
that the notion of representation itself is not readily applicable to his
Creatures at all. Like the claim of some early connectionists that they
had bypassed the notion of representation altogether, this claim seems
to me hyperbole, especially given that these are computationally driven
processes. Given that we admit no computation without representation,
then either his creatures are a combination of reaction (world-creature)
and computation (in the creature), or they are computational all the way
out. Either way, we have a locationally externalist view of processing, and
in the latter case, wide computationalism.38
Individualism and Externalism in Cognitive Sciences
178

We can now, ¬nally, return to the question of whether wide com-
putational systems themselves “ as opposed simply to their in-the-head
realizations “ are properly thought of as cognitive. Suppose that one
grants locational externalism about computational systems. Might one
still maintain individualism about cognition by conceptualizing the cog-
nitive system proper as ending at the boundary of the individual? Such a
view rests on the plausibility of identifying what is internally realized as
always being the appropriate cognitive system. Consider the navigation
systems that Hutchins discusses. For individualism to be defensible as a
view of cognition in this case, there must be an in-the-head account of all
of the relevant cognitive processes; likewise for the case of animate vision
and behavior-based robotic systems. But recall that all three cases share
the strategy of pointing to individual-world interactions as a way to avoid
positing more costly internal representations and computations. Thus,
this individualistic move is unmotivated and runs counter to the recon-
ceptualization of representation within these paradigms. Representation
is not something implanted in individuals but something that individuals
do by exploiting the rich structures of their environments in cycles of
perception and action.


9 having it both ways?
Two strands to the discussion in this chapter might be thought to be in ten-
sion with one another. The ¬rst stems from my claim that once one moves
from the cognitive science gesture to a more full-¬‚edged examination of
theory and explanation in the cognitive sciences themselves, the debate
between individualists and externalists becomes tougher to resolve. This
debate turns not only on how one interprets particular theories but also
on broader issues in the philosophy of mind and science. Thus, I have
been arguing not only for the conclusion that the cognitive science ges-
ture itself far from resolves the debate in favor of individualism, but that
going beyond that gesture leaves much in the debate open. But, second,
in offering a wide computational interpretation of Marr™s theory of vision
in section 6 and in linking this interpretation to a range of contemporary
work on situated and embedded cognition, I have argued for a resolution
of the debate in favor of externalism. Surely, one might think, one can™t
have it both ways.
The way to resolve this tension, I think, is to lean on the distinction be-
tween taxonomic and locational externalism. The debate over Marr™s the-
ory of vision has been conducted almost exclusively in terms of taxonomic
Representation, Computation, and Cognitive Science 179

externalism and individualism, as manifest in the focus on narrow and
wide content. But whether any theory employs a notion of narrow or wide
content will turn largely on what it says about wildly counterfactual cases,
cases in which there are molecular (or even narrowly functional) dupli-
cates in minimally different environments. And Marr™s own discussion of
the computational level supports both a narrow and a wide interpreta-
tion of it. Thus, the nonconclusiveness of individualism construed as a
taxonomic thesis about the cognitive sciences.
Once we shift to the locational conception of externalism, matters be-
come somewhat clearer. The issue here is whether the cognitive sciences
sometimes investigate wide computational systems, and at least on the
surface this seems easier to resolve, since it is a matter primarily of iden-
tifying whether the computational system of interest literally extends be-
yond the boundary of the individual. Locational externalism is a stronger
view than taxonomic externalism and entails it unless there is a plausi-
ble, nonquestion-begging way to individuate mental states independent
of their total realizations. We have seen several failed attempts to do this,
in this chapter and the last, and the emphasis on locational externalism
clearly places an onus on individualists.
This is a matter of whether the cognitive sciences do or should employ
a strategy of integrative synthesis, as well as constitutive decomposition,
whereby individuals and their cognitive processes are located in systems
larger than those individuals. My argument in this chapter has been that
by refashioning the notions of representation and computation, there is
no reason why not, and much reason to be bipartisan or pluralistic about
these notions, as we should be about realization.


10 beyond computation
In introducing and defending wide computationalism as a perspective
on research in cognitive science, I have appealed to existing work that I
think can plausibly be seen as exemplifying a wide computational view
of cognitive processing. Such a view shifts the attentional focus of the

<<

. 29
( 62 .)



>>