Chapter 1
argued that theories in cognitive science and cognitive anthropology suffer
from problems that are related through some common strands of intellectual
history yet take on unique forms in the respective disciplines. In this concluding chapter, an attempt is
made to restate these issues, showing how the theoretical reformulations and
data analysis of the last five chapters help illuminate these problems and make
steps toward rectifying them. First,
however, some caveats and definitions should be made or restated.
By "a theory of
embodied cognition" I mean a theory of human cognition which explains the
properties of individuals (intact with sensory, motor, and other capacities)
acting in specific contexts, as a consequence of their engaging a material (and
often, socially and historically constructed) situation, informed by some
history of such engagement. The term
"embodied cognition" subsumes (in my usage) the similar terms, "situated
cognition," and "situated activity," since these latter terms tend (in many
instances in the literature) to deny or ignore many of the structures and
processes which are literally part of the body and which are fundamental
contributors to knowing and acting (cf. Lave, 1988; Suchman, 1987). The term "embodied cognition" (in my usage)
is also different from the use of this term by some who characterize human
cognition (more and less exclusively) in terms of universal aspects of body
function, denying, ignoring, or failing to acknowledge the fundamental roles
played by history and social context in cognitive events (cf. Lakoff & Johnson,
1980; Johnson, 1988). Both "embodied"
and "situated" imply (in my usage) an emphasis on context-sensitive performance—the employment of material structure
in the socio-historically constructed world, by real people, for acts of
cognition. The terms "embodied" and
"situated" are largely used interchangeably throughout this dissertation,
although they are also employed individually to emphasize aspects of body and
situation, respectively, in specific instances.
At the same
time, my use of "embodied cognition" has not addressed (except superficially)
the affective components of human cognition.
This omission is neither desirable nor necessary, except as a means for
making the enterprise manageable.
Furthermore, I do not imagine that the inclusion of properties of affect
will render problematic or less valuable the contribution made here. While many scholars find comfort in their
belief that an account of human emotion will render obsolete any information-processing view of mind,
one aspect of my effort here has been to open up new avenues for the
reintegration of affect and cognition in theory by attempting a reformulation
of the information-processing account.
Cognitive
science, a product of intellectual history centered around cognitive psychology
and computer science, has no theory of humans as cognitive actors situated in
material, historical, and social contexts.
Instead, human cognition is seen in terms of underlying universal mechanisms
whose properties are investigated through laboratory experiments which attempt
to isolate the functioning of those mechanisms. This investigation, it is further assumed, will provide
constraints upon a model (theory) of the universal human cognitive
architecture. The blueprint for this
human cognitive architecture is explicitly laid out by Allen Newell and Herbert
Simon in their notion of a "physical symbol system."
A physical symbol system consists of a set of entities, called symbols, which are physical patterns that can occur as components of another type of entity called an expression (or symbol structure). Thus a symbol structure is composed of a number of instances (or tokens) of symbols related in some physical way (such as one token being next to another). At any instant of time, the system will contain a collection of these symbol structures. Besides these structures, the system also contains a collection of processes that operate on expressions to produce other expressions: processes of creation, modification, reproduction, and destruction. A physical symbol system is a machine that produces through time an evolving collection of symbol structures (Newell & Simon, 1990[1976]:117).
Having thus
defined (under conditions of necessity and sufficiency, ibid:118) a class of
machines which are capable of intelligence, Newell and Simon conclude: "A
physical symbol system is an instance of a universal machine. Thus the symbol system hypothesis implies
that intelligence will be realized by a universal computer" (ibid: 119).
Digital
computers—the material instantiations of one kind of universal machine (the
"Turing Machine")—were thus married to a model of human cognitive architecture
who's material instantiation could be investigated in the laboratories of
cognitive psychology. Articulating the
hypothesis underlying the dominant research agenda in cognitive science, Simon
says:
[I]ntelligence is the work of symbol systems....On the one hand, by constructing computer programs that are demonstrably capable of intelligent action, we provide evidence on the sufficiency side of the hypothesis. On the other hand, by collecting experimental data on human thinking that tend to show that the human brain operates as a symbol system, we add plausibility to the claims for necessity, for such data imply that all known intelligent systems (brains and computers) are symbol systems (Simon, 1981:28).
In this
fashion, the field of cognitive science has come to equate human cognition with
the formal manipulations afforded by symbol systems, as exemplified in the
operations of digital computers.
Neither the laboratory investigation of properties of individual minds or
brains, nor the physical symbol system hypothesis, nor the modeling of mind
through information processing techniques are, by themselves, unsound
components (in my view) of the science of cognition. However, the practical reality of their marriage has been a problem. This problem lies with the fact that, in
practice—as Hutchins (1995) points out—the physical symbol system hypothesis
has yielded a science which has mistaken the
properties of human cognition as constituted in cultural process for the
properties of individual brains which compute by manipulating internal
symbols. Regarding Turing's formulation
of the abstract principles of the digital computer (the "Turing Machine"),
Hutchins says:
The heart of Turing's discovery was that the embodied actions of the mathematician and the world in which the mathematician acted could be idealized and abstracted away in such a way that the mathematician could be eliminated. What remained was the essence of the application of rules to strings of symbols....The mathematician who was a person interacting with a material world is neither modeled by this system nor replaced in it by something else. The person is simply absent from the system that performs automatic symbol manipulation. What is modeled is the abstract computation achieved by the manipulation of symbols. All of that is fine, if your goal is to extend the boundaries of human computational accomplishments. But it is not necessarily a model of the processes engaged by a person doing that task. These programs produce the properties, not of the person, but of the socio-cultural system.....The physical symbol system architecture is not a model of individual cognition. It is a model of the operation of a socio-cultural system from which the human actor has been removed (Hutchins, 1995, p. 517 in manuscript, original emphasis.)
The primary
implication of mistaking the properties of socio-cultural systems for the
workings of individual minds is that symbols, those material entities of the lived-in world which signify for persons due
to their being in that world as mediators of embodied action, get
abstracted into logically related forms and, simultaneously, shoved into the
heads of individuals. Freed from the
problems of grounding the meanings of these symbols in the world—that is, free
to operate on the symbols with formal rules without recourse to interpreting
the symbols via the functional properties of the cognitive system acting in the
world—computer programs can do some amazing (intelligent?) things, and yet fail
miserably at other tasks which seem to be trivial human accomplishments. In
fact, one can read into Hutchins' argument a prediction about that class of
phenomena that computers will fail at (e.g., perception, recognition, semantic
interpretation of language) and those tasks at which computers will succeed
(e.g., language translation at the level of syntax, problem solving of formal
games and mathematics or logic, and "expert systems" of certain types).[1] These latter phenomena are successfully
tackled by artificial intelligence programs because these domains are already elaborated in terms of formal
symbol systems that the computer has then merely to emulate, something it was
built—in the image of Turing plus rules of predicate logic, plus pencil, plus
paper, etc.—to do. Of course, these
predictions do not need to be made because they have already come true in the
last 30 years of artificial intelligence research.
One area that
has been particularly well-represented in artificial intelligence research goes
by the name "problem solving." In the
context of what has just been said, Newell and Simon's proposal for problem
solving gives us cause to reconsider a piece of reasoning of Plato's, in spite
of N&S' skepticism.
If you think there is nothing problematic about a symbol system solving problems, you are a child of today, whose views have been formed since mid-century. Plato (and, by his account, Socrates) found difficulty understanding even how problems could be entertained, much less how they could be solved. Let me remind you of how he posed the conundrum of the Meno.
Meno: And how will you inquire, Socrates, into that which you know not? What will you put forth as the subject of inquiry? And if you find what you want, how will you ever know that this is what you do not know?
To deal with this puzzle, Plato invented his famous theory of recollection: when you think you are discovering or learning something, you are really just recalling what you already knew in a previous existence. If you find this explanation preposterous, there is a much simpler one available today, based on our understanding of symbol systems. An approximate statement of it is:
To state a problem is to designate (1) a test for a class of symbol structures (solutions to the problem), and (2) a generator of symbol structures (potential solutions). To solve a problem is to generate a structure, using (2), that satisfies the test of (1) (Newell & Simon, 1990[1976]:126-127, original emphasis.).
While the
proposal put forward by N&S is commendable for its attempt to
operationalize the concept of problem solving, when put into practice as a
model of human cognition it suffers from the fallacy of the "brain == physical
symbol system == digital computer" conflation.
The basic approach to implementing the proposal by N&S is to imagine
that cognitive agents construct internal symbol structures which can stand for
a goal state, as well as possible intermediary states linking the goal state to
the current state by way of contingencies or state transitions. Through trial and error—or more
sophisticated "tests" of the current state and the relation between current
state and goal state—the agent can proceed toward the goal state by effecting
changes to the environment and, upon reaching the goal state and recognizing
it, has solved the problem.
While the
symbol system hypothesis does not rule out (in principle) the possibility that
the cognitive system qua problem
solver is sensitive to structure in the world—for example, employing the
effects of alternative representations, processes, and situations which may
substantially alter the mechanisms entailed in solving a problem—in practice,
there is no general account (short of ad hoc incorporation as an external
memory) for accommodating this structure and no empirical investigation
recommended by this shortcoming. In
sum, there is no theory here of how such external structure integrates with,
and constitutes, the cognitive system.
According to Hutchins, there is no recognition given to these issues
because structure and processes external to the system have largely been
incorporated into the functioning of the architecture which is then (mis)taken
to be a model of the individual human mind.
If we believed
that the proposal by Newell and Simon yields an accurate account of the way
minds work, we would go interview problem solvers (or experts or natives), or
record their verbal protocols of what they do, or time them in their
performances of tasks, and write a production system (a computer program), or a
flow chart of decision-making, which would represent the relevant data, their
relationships or structure, and provide for operations on them. In fact, this is precisely what cognitive
scientists and cognitive anthropologists have spent much of their time doing in
the last thirty years (cf. Gladwin, 1989; Miller et al., 1960; Newell & Simon, 1972; Spradley, 1972; Randall,
1987). For example, the problem of
deciding where to set the trawl, in mid-water pair-trawling, would entail a set
of states (e.g. "ship in port," "ship at fishing grounds," "ship on top of good
fish," etc.) and contingencies for transitioning between states to reach the
goal state (e.g. "if in port and need to be at fishing grounds, leave port,"
"if ship on top of good fish, set trawl," etc.) along with, of course, the
"effectors" which turn representations internal to the system into behaviors taken
by the system.
The data and
analysis of Chapter 4 suggest that this would not be a good characterization of
the cognitive properties of any of
the individuals involved in the activity of deciding where to set the trawl in
the reported context. Although it might suffice as an abstract
representation of the activity as a whole, it would suffer from an inability to
articulate the meanings of its terms (for example, when is fish "good," and
when should attention be paid to other boats' activities) without incorporating
ever more components of the history and context of the activity—at which point,
if possible, it would be a system
emulating the entailed cultural process but without lending understanding to
how this process works by way of the cognitive properties of the individuals
involved. In contrast, the analysis
given in Chapter 4 attempts to articulate some of the properties of this
cognitive system in a manner which separates out the functional roles of internal
and external structures and processes and how these interact to produce the
activity. It is worth reviewing those
findings (labelled here with the same heading numbers as in Chapter 4) in light
of the current argument.
1.0 Interpretations (acts of reasoning) made by
captains are mediated by external and internal representations.
1.1 External representations (radar displays)
are different for the two captains of the team due to (a) the boats' positions
relative to the pack and other objects of joint interest, and (b) different
radar devices which entail different coordinate systems, range settings and
resolutions, and display features.
These differences generate a distribution of potentials for action. For instance, when the captain's boat is
both physically situated in the middle of the pack and conveniently represented
to be in the middle of this pack, just noticing a change in spatial arrangement
of echoes on a relative motion radar display affords complex computation.
1.2 These external representations must be
coordinated with internal structures if they are to be employed for
reasoning. These internal structures
are, in turn, built up through embodied performance in the practice. Sometimes this coordination is
problematic—for instance, one captain's tendency to confuse west and east, even
though his radar display represents these directions unambiguously. Other times the coordination facilitates
automatic inference—for instance, if one is among a pack of boats which
maintains a constant heading (on the way to the fishing grounds) and the pack
doesn't split up, then boats which earlier on reported to be on the same
heading as one's own boat are probably now in the vicinity.
1.3 These external representations, when shared
through a common technology and context for generating them, make transparent
the problem of communication with minimal information exchange. That is, intersubjectivity is accomplished
through sharing of external structures which are put in place by the time
course of materially instantiated activity.
Knowing the referents of ambiguous speech like "those banks" or "him"
are built out of shared history (that is, the micro-time of ongoing actions)
employing external structures such as sea charts displayed on the electronic
navigator and streams of speech in discourse.
1.4 The technology enables direct (as well as
automatic) manipulation of external representations as computations in support
of reasoning. For instance, disabling
the "clearing" which takes place between sweeps of the relative motion radar
results in echoes leaving traces which thus represent relative changes in
velocity between the host and other boats over time. Although this particular representation is difficult to reason
with (e.g., the untrained observer tends to read change in position into this
trace) it does afford certain kinds of computations for skilled users—that is,
users who have built coordination between their internal structures and this
particular kind of external structure.
Another kind of computation is made possible by simply placing the
relative motion radar's cursor on top of an echo. Over time, any change in spatial arrangement between cursor and
echo makes immediately salient changes in relative velocity of host and object
represented by that echo.
1.5 External representations must be coherently
organized to enable reasoning, and there are means (deriving from both
conscious efforts and built-in properties of the language) employed by
interlocutors to accomplish this coordination.
2.0 Elaborating further on the theme of (1.5),
we can say that intersubjectivity is accomplished through the negotiation of
actors' understandings about the state of the world. This negotiation entails processes involving the give and take of
information and ongoing attempts to make coherent all of the constraints and
resources which are salient to, and impinge upon, the cognitive actors
involved.
2.1 One of these processes is embedded in the
structure of the discourse itself.
Frequent radio exchange between captains—something done in the name of
efficiency as well as for relief from boredom and for social interaction—makes
possible the establishment of frameworks for building further shared or
collective understanding. These
frameworks—disambiguated or intersubjectively established aspects of the situation—are
employed to "hang" further constructions upon.
For example, pronouns index referential speech used in previous
exchanges. Although this is a
commonplace function of discourse, it's extension across time and space in this
setting is importantly supported by information provided by (and selectively
extracted from) the situation being confronted and collaboratively constructed.
2.2 Conventions of general discourse style carry
over into the on-line communications between captains to inform listeners of
speakers' intentions. "Commands" for
action, which are antithetical to broader cultural conventions, must be
(re)cast in the appropriate constructions.
Furthermore, the conventions which dictate between boat and between pair
cooperation play a role in the shape of the processes which support each
actor's reasonings about the meanings of situations.
2.3 The distribution of understandings (and of
information collection and processing) between captains makes possible more robust
(e.g., error tolerant and adaptive) "knowing" than either captain can
accomplish alone.
2.4 Another process which facilitates
coordination between interlocutors is taking the perspective of the other in
framing questions and providing answers.
This can also be built into the sharing of meanings of terms directly
(such as the practice-based sharing of terms used to describe sonar
representations of fish, see Chapter 5).
3.0 Communication generates cognitive effects
which are not predicated upon the content of messages exchanged between
interlocutors, nor what is directly known by interlocutors.(i.e., not
internally represented in any conscious manner, and sometimes not internally
represented at all)
3.1 Discourse—the act of attending to the speech
issued over the radio—causes one to attend to instruments and evaluate their
representations.
3.2 An interlocutor's need to reply causes him
to vocalize held beliefs which may not be in conscious attention. This process, in turn, creates the
juxtaposition of possibly only tenuously (or even, inconsistently) held
beliefs. That is, the vocalization of
unstructured memory traces forces structure upon them which generates
resolution-finding measures for sets of inconsistent beliefs.
3.3 In a similar vein, discourse generates the
publication of interpretations of the situation (themselves, perhaps, only
inchoately formulated) which can create the commitment to act according to that
publication. At the very least, discussing
the current state (thus making the choices to highlight certain understandings
and background or ignore others) has the effect of joint planning about what
should be done next.
In summary,
there were a large number of processes and structures observed to be at work in
the situated activity (i.e., the "problem") of deciding where to set the
trawl. These were not ad hoc
observations of random events, but were both grounded in and made possible by
the theory of distributed cognition outlined in Chapter 1. The theory anticipates, it will be recalled,
that cognitive actors employ structure in the environment opportunistically,
and yet by incorporation of those regularities into their daily routines become
conditioned to exploiting it. This, in
turn, generates the residua of action (structure internal and external to the
actor) which propagates the existence and form of the practice in an ongoing
cultural process.
Although
Plato's proposal for problem solving by the individual ("recollection from a
former existence") is not particularly revealing as it stands, one could read
it as entailing a sensitivity to issues which get swept away by (or rather,
swept up into) the mechanisms of Newell and Simons' cognitive
architecture. When the boundaries of
the cognitive system are properly redistributed, the difficulty of "knowing"
for the individual actor and the ease with which "solving" is accomplished by
the activity system in which the actor is embedded, make Plato's proposal (if
not literally acceptable) at least metaphorically accurate. That is, while it is doubtful that any
"former existence" of the individual
problem solver accounts for his or her "knowing," the system composed of actor
plus environment has indeed been built out of the recurrence of this problem (i.e., constructed through a history of
actors engaging the problem). This
recurrent engagement by other and former members of society, with the available
material means, generates resources for cognitive action—structure in the
environment and internal to actors which accommodate the kind of problem
solving only humans have proven capable of achieving.
Cognitive
anthropology—a product of intellectual history centered around cognitive psychology,
psychological anthropology and linguistics—also lacks a theory of humans as
cognitive actors situated in material, historical, and social contexts. Instead, cognitive anthropologists—being
concerned with articulating a relationship between historically bounded groups
of people and cognition—have focused exclusively on the equation of "culture"
with "shared knowledge" (cf. D'Andrade, 1981).
Without a theory of embodied cognition and cultural process, this
amounts to claiming for culture the status of historically derived "mental
contents" in the classical view of mind.
In this view, culture becomes the data structures upon which formal
operations are carried out inside the heads of individuals. As outlined in Chapter 1, this state of
affairs in cognitive anthropology is due, in large part, to the influence of
theory in neighboring fields of cognitive psychology and structural
linguistics, and the inherited epistemological and methodological traditions of
American cultural anthropology.
Lacking a
formulation of embodied cognition the central role of "culture" in this
classical theory of cognition is to provide the "shared knowledge," "common
sense assumptions," or more formulaic "cultural models," which organize and
provide content for more specific scenarios, schemas, scripts, and goals, which
members of a society employ for knowing.
As a result—I claim—the actor, action, and situation, disappear from the
scene of "cultural cognition" as completely as Turing does from his formulation
of universal computation. While it was
claimed in Chapter 1 that the central problem for cultural anthropology is to describe interpretive processes,
Naomi Quinn and Dorothy Holland (henceforth Q&H), in the introduction to
their recent (1987) edited volume Cultural
Models in Language and Thought, claim that the central question is "How are
meaning systems organized?" (Quinn & Holland, 1987:3). The entailments of any cognitive theory
addressing this question, they say,
must be able to explain the apparent systematicity of cultural knowledge...[It also] ought to explain how we come to master the enormous amount of cultural knowledge that the people of any culture have...Moreover, the large base of cultural knowledge we control is not static; somehow, we extend it to our comprehension of particular experiences as we encounter them...A theory of the organization of cultural knowledge must explain the generative capacity of culture (ibid:3-4).
Leaving aside,
for the moment, the conflation here of any one person's knowledge with some
collective store of shared cultural knowledge, as well as the conflation of
knowing with acting, it is clear that Q&H are defining the central problem
for cognitive anthropology in a vein highly reminiscent of Chomsky's (1957)
argument for the proper study of language.
That is, cultural knowledge/behavior is systematic and productive or
generative, therefore it's study is properly one of the underlying formal
relations which obtain between elements of cognitive deep structure (cultural
models) which are (somehow) the generators of behavior. Consecrating the affinity between
linguistics and the proposed enterprise for cognitive anthropology, Q&H
claim that the new methodology (adding to a stock of traditional elicitation
and analysis techniques) entails two new data sources: "systematic use of
native-speaker's intuitions, and analysis of natural discourse" (ibid:16).
The theoretical question is: How is cultural knowledge organized? The methodological strategy is to reconstruct the organization of this shared knowledge from what people say about their experience (ibid:18).
Quinn and
Holland make clear that cultural models—the analytic models inferred from talk
to reside as structures in peoples' minds—are problematically related to behavior. However, this issue is not seen by Q&H to warrant a refocus
of study on action more generally because, they argue, cultural models
nonetheless "frame experience, supplying interpretations of that experience and
inferences about it, and goals for action" (ibid:6). Here, Q&H make a strong theoretical commitment to "goals" in
an attempt to bridge a complete formulation of individual acts of cognition
with the abstractions of shared collective knowledge, "cultural models," which
are the products of analysis. Goals, as
part of the repertoire of shared understandings in a culture, are said to guide
behavior and make that behavior mutually intelligible for members of the
culture.
Sometimes these cultural models serve to set goals for action, sometimes to plan the attainment of said goals, sometimes to direct the actualization for these goals, sometimes to make sense of the actions and fathom the goals of others, and sometimes to produce verbalizations that may play various parts in all these projects as well as in the subsequent interpretation of what has happened. Complexity in the relationship between what people verbalize about what they do and the execution of other, nonverbal activities is inherent in part because speakers so frequently undertake complex tasks with many goals that may or may not include producing a veridical description of what they are about (ibid:6-7).
Apparently, the
ontological status of cultural models—that is, their existence in the minds of
individuals and their contribution to cognitive events—is characterized by the
ways these models specify actors' goals.
The behavior generated to accomplish cognitive tasks in service of these
goals may, in turn, not be veridically articulated in talk due to matters of
complexity, performance limitations, intentional disguise by informants, or the
fact that the talk is about something other than the cognitive task being
addressed by the behavior. This
formulation creates a number of epistemological dilemmas which are cause for
concern. One dilemma follows from the
discussion above, which questions the validity of the standard theory in
cognitive science regarding how goals relate to human action (cf. Suchman,
1987). A second dilemma—the extent to
which cultural models can be claimed to be resident in individual minds as
structures of "knowledge"—will be considered below in some detail. This dilemma is laid bare in Q&H's
claims for how cultural models compare to the analytic use of "scripts" in
cognitive science (cf. Schank & Abelson, 1977).
Quinn and Holland
claim that cultural models resolve the problems faced by artificial
intelligence researchers who are trying to characterize the knowledge required
to act appropriately in concrete physical and social settings. These researchers often conclude, Q&H
inform us, that "there is too much common sense knowledge of the world in even
the humblest normal human head for present computer systems to begin to cope
with" (Abelson, 1975:276, quoted in Quinn & Holland, 1987:19). The actions or sequences of events in question
are characterized by artificial intelligence researchers in structured but
simplified scenarios or "scripts" such as ORDERING FOOD IN A RESTAURANT or
BUYING GROCERIES. These scripts are
imputed to reside in cognitive actors' minds, are "derived from daily routine,
[and] are standardized sequences of events that fill in our [i.e., actors']
understanding of frequently recurring experiences" (Quinn & Holland,
1987:19). While cultural models are
like scripts in their ontological status as mental structure as well as their
employment of sequential or causal connections and in their prototypicality
entailments, cultural models are unlike scripts because "their enactment is not
tied to a concrete physical setting, as is that of the restaurant script" (ibid:19). Where cultural models come into play,
Q&H assert, is in providing the high-level themes which generate script goals, and even scripts
themselves.
Cultural knowledge is key to the higher-order structures that embody goals in Schank and Abelson's formulation...As Schank and Abelson are led to ask, how do these goals themselves arise? How are story understanders and other observers of the everyday world able to assess actors' goals and predict their future goals? Schank and Abelson's answer is that related goals are bundled together in "themes." These themes are said to generate actors' goals as well as other people's inferences about these likely goals. It is possible to make such inferences about the goals of other people, presumably, because knowledge of themes, no less than knowledge of scripts, is shared (ibid:20).
Schank's (1982) more recent reformulation of the theory of scripts is much more sophisticated about how knowledge must be hierarchically organized and continually modified in memory in order to account for such processes as reminding and the generalization of learning...Failing [however] to make a place in his account for knowledge that is culturally shared and transmitted, Schank is left with the awkward supposition that an individual's understanding of the world is accumulated through the painstaking generalization of knowledge from one firsthand encounter to another. It is difficult to understand how people could learn as much as they know, even by the time they reach adulthood, from personal experience alone (ibid:21).
Here, Quinn and
Holland are proposing (again, in very Chomskian fashion) that cultural
knowledge (the underlying "stuff" of knowing and acting) must be the mechanism by which open-ended, productive, and
generative cognitive performance is made possible. However, as with the argument for the role of innate grammar
underlying productive language abilities, the argument here rests heavily on an
inability to imagine it otherwise—a position many linguists have begun to reconsider. That cultural process entailing social
interaction makes available to actors resources which solve the problems of
"knowing" about their world, is not being questioned (cf., Hutchins &
Hazlehurst, 1991, 1995). What is being questioned is the extent to
which the mechanisms of knowing reside in the contents and structure of shared
mental models, as captured in the proposition-like formalisms of scripts and
cultural models. Quinn and Holland
respond to the (ill-formulated) "size problem" of artificial intelligence by
positing a "cultural grammar" which generates not only knowledge but, also,
shared knowledge. The mechanism in this
account is strictly a product of mental contents, operations which can operate
upon those contents because of their structure, and the universal basic
properties of brains. This commitment
by Quinn and Holland—no less than the one by Newell and Simon as to where the
boundary of the symbol system lies in problem solving—effectively removes the
actor, the situation of action, and the cultural process from consideration
regarding the nature of human cognition.
An alternative
explanation for the "size problem" (and the "intelligence through knowledge"
issue which lurks beneath) is to recognize it as an ill-formed problem and
shift the burden for knowing onto the structures and processes which exist in
the socially constructed world and with which actors come into coordination through
histories of embodied action. Notice
that "embodied action" here is not the same thing as "firsthand
encounters"—apparently, the learning of structure in the world by brute force
or without mediation by socially constructed tools—which Q&H attribute to
Schank's account. The concept of
"embodied action" includes—in fact, requires—this direct experience with
material structure in the world, but it also includes the learning of socially
constructed word meanings, forms of narrative, technology, and other structures
which mediate acting in (and, consequently, knowing about) the world.
Of course,
cultural models theorists do not deny the relevance of learning—after all,
cultural knowledge is not genetically inherited. It's just that learning has not been a focus of interest for
these researchers, often described by them in terms of processes of knowledge
"transmission" and "acquisition"—as if the relevant properties are somehow just
carbon-copied into the minds of learners.[2] In fact, a focus on learning would require
that actors be put back into the world and that the nature of internal
structures—their status as messy associations of embodied experiences, their
differences across individuals, the mechanisms for their coordination with
external structures, the ways they come into being, and their relations to
behavior—become true problematics rather than givens in a theory of
cognition. In this reformulation,
learning itself must take on a new status.
Learning in a cultural process is not the inheritance of abstracted
propositional structure but, rather, the generation of means for coordinating
with material structures in the world—including the speech and behaviors of
other actors—through interaction with them.
Through use of these
structures, by and between actors, internal structures are built which make
possible organized behavior. By sharing
the material means of cognitive performance in a socio-historically constructed
world, actors become competent at navigating in this world such that their performances
have predictable (yet non-deterministic) functional consequences.
Chapters 4 and
5 took the perspective shown in Figure 58 in mapping out the ways the cultural
process works via particular histories of particular cognitive performers. These two chapters entailed an attempt to
make more explicit how resources provided by the cultural process become
available for use by individual actors, and how this use reproduces the
respective practices in which the actors participate. This reproduction makes the means for knowing which are so hard
to account for in models of mind—often described in terms such as "background
assumptions," "implicit knowledge," and "common sense knowledge"—quite
transparent to the actors themselves.
As Lucy Suchman puts it, drawing upon a framework established by
ethnomethodologists:
[T]he "taken for granted" denotes not a mental state, but something outside our heads that, precisely because it is non-problematically there, we do not need to think about. By the same token, in whatever ways we do find action to be problematical, the world is there to be consulted should we choose to do so. Similarly, we can assume the intelligibility of our actions, and as long as the others with whom we interact present no evidence of failing to understand us, we do not need to explain ourselves....To characterize purposeful action as in accord with plans and goals is just to say again that it is purposeful and that somehow, in a way not addressed by the characterization itself, we constrain and direct our actions according to the significance that we assign to a particular context. How we do this is the outstanding problem (Suchman, 1987:47, original emphasis).
The thesis in
this study has been that answers to the question of "how purposeful action is
generated" are located in the means for organizing behavior which are
established in a cultural process.
These means do not exclude
internal structures and processes internal to individuals—means for
coordinating with structure in the world—which are, obviously, essential
components of cultural process. On the
other hand, it has been argued that the means for organizing behavior are not
adequately addressed by views which impute symbolic, propositional, and formal
properties to individual minds in characterizing human abilities to "problem
solve" and to "know".
The fishermen from Vindö do not "know" (or, at least, do not all know) about those properties of the DECCA navigation system which generate the partitioning of regions of water into series of lanes bounded by hyperbolas. They do know how to employ this system to determine their positions on the water because they have (more in the past than now) interacted with these coordinates as lines overlaid onto sea charts and as symbols registered on devices which (together) are used to compute fixes of position for the purpose of navigating and finding fish (see Chapter 3). In time, the paper charts have become obsolete—replaced by electronic versions—and the DECCA system has been replaced by a satellite-based system as the primary means for navigating. DECCA coordinates, however, are still the basis for "knowing" about underwater wrecks and (for some) how to catch fish because these were the coordinates with which (over many years) observations of wreck locations and productive fishing grounds were recorded. However, with regard to epistemic contributors to action, fishermen may not even "know" this—that is, they do not volunteer it in conversation, it's not necessary for accomplishing their objectives, the actions they take do not rely upon it being formulated and held their heads, and you cannot find it disguised in the logic of their talk (which isn't, by the way, to say that you cannot get them to explain it to you). Instead, fishermen "know" that when they drive their boats so as to maintain coordination with the (now computer-drawn) trajectories on their charts and so as to avoid near-encounters between DECCA receiver symbol (now automatically plotted on the computer screen) and symbols for wreck locations (displayed on this same screen), that they are doing the right thing.
Chapter 5 was an attempt to address
more specifically the means by which fishermen know about the prospects for
catching herring, the ultimate goal of their activity. Sonar devices, the results of technology
developed for making fishing more efficient, do not yield unambiguous
displays of underwater objects. These
displays, as representations of fish, must be constructed—that is, converted from images made on paper or color
monitors into evaluations about the fish they represent. This is accomplished by direct experience
with the displays, the results (caught fish) of relying on these displays for
taking action, and the language about these displays employed in practice to
improve the chances of catching fish.
The
experimental results of Chapter 5 demonstrate that these elements of practice
account for much of the means by which these fishermen interpret sonar
displays. The fishermen who fish
together and who spend the most time interpreting the displays and
communicating about them agreed with each other most about the meanings of
those displays and the meanings of words used to describe them. If the meanings of these words were unproblematically
shared among fishermen, we would not expect to see the distribution of
differences in assignments of meanings to words which the experiment, in fact,
revealed. These words—produced
artifactual structures employed to communicate about the displays—do not come
"attached with meanings" or the basis for knowing but, rather, the meanings
must be established in the course of using them to mediate performance in the
world. These meanings are—the findings
suggest—constructed through histories of cognitive action mediated by many
layers of structure and process which are engendered by the practice-based
activities of fishermen. The meanings
become shared because the functional properties entailed by acting under these
conditions are shared. These conditions
include the following four.
(1) The properties of the display device itself makes available the raw materials from which evaluations and linguistic representations become possible.
Sonar devices
yield analog-like representations (employing the conversion of returned sonar
signals into degrees of intensity) using density of ink on paper or a color
scale on a CRT monitor. These devices
continually generate display features which give information about those
objects which return sound waves and which are lying underneath the water,
below the boat or entering the trawl.
These display features provide the raw materials out of which linguistic
structures are created and by which linguistic structures take on meaning for a
skilled interpreter. This is clearly
reflected in constructions such as "groups," "marks," "it's completely white,"
"it's been half a paper now," "showing a bit of color," "not very red," "a
toothpick," and "a stake" used to describe and give meaning to what one sees.
(2) The patterns of interaction between actor, display device, and situations make possible (and salient) both the visual evaluation of sonar displays and practice in the linguistic formulation of what one is seeing.
Not all
fishermen spend their time in front of the sonar device. In fact, it is characteristically the case
that very few members of the crew are on the bridge at any one time, and this
group nearly always includes one captain.
Furthermore, modern technology places almost the entire responsibility
for driving the ship, navigating, communicating with other boats, and
commanding actions for fishing, in the hands of one individual. For a vast majority of this time, this
individual will be one of the 2 or 3 owners (captains) of the ship. Members of the crew who are not owners, and
are not captains, rarely spend much time focused upon what the sonar device is
representing. These crew members spend
even less time constructing the linguistic structures to communicate and make
explicit their understandings of what the sonar device is representing.
These
understandings, built through embodied action, migrate into the language
employed, integrating the linguistic constructs which denote properties of the
display with constructs whose meanings are attributed to the fish interpreted
to be represented by the display.
Examples of this integration, often yielding a multi-referential
framework which is quite transparent to skilled interpreters, include: "it's
not hard like a real fleck," "it's completely empty," and "almost the whole
fleck has taken hold of the bottom." In
each of these cases, properties of the display device and of the represented
fish are interwoven in a manner which is quite seamless for skilled
interpreters, yet always open for renegotiation based on the effectiveness of
resultant communication.
(3) Feedback from evaluating the results of actions (taken in accord with constructed meanings) is used to construct new meanings.
Although the
feedback obtained from the results of the catch are difficult to assign to many
display images (possibly 5 hours worth, in the case of trawling), there is
plenty of evidence that this is in fact done.
Captains regularly ground their interpretations of what they are
"seeing" on the device—and publish this as justification for their decisions,
expectations, and hopes—in terms of the judged similarity or dissimilarity of
the display to previous experiences (sightings of the day before, or even of
the last time they were in this region of water). Again, non-captain crew members, rarely engage this problem
directly since—although they are very interested in the catches actually
made—rarely spend time reflecting on the sonar displays which made it all
possible, engaging instead the problem of getting the catch sorted, packed, and
prepared for landing.
(4) The patterns of interaction between interlocutors—in which verbal representations are formulated and interpreted for the purposes of taking action—shape the distribution of meaning sharing.
The captains of
boats which fish together in complete cooperation (such as the team of
pair-trawlers, or the mother ship and light boats of the purse seining team)
regularly communicate with each other about what they are "seeing" in order to
make their enterprise more efficient.
This communication takes place over private radio channels whose
functions (as private channels) are maintained as closely guarded assets of the
business. These channels make possible
the coordination of activity in which the representations of fish are employed. This coordination is reflected in the
distribution of internal residua from actions entailing the integration of
visual and linguistic representations of fish.
In Chapters 2
and 3, the description took the perspective shown in Figure 1 in mapping out the
ways the cultural process works via both typical and actual instances of
individuals acting in the world. In
Chapter 2, the internal structures which were posited to participate in the
ways fishermen "know," were characterized as a product of subjective experience
in an objective world (i.e., a world which, although socio-historically
constructed, is materially instantiated).
This characterization was cast in terms of the phenomenon of "identity,"
or the means by which subjective experiences are, and become, organized. It was argued that this organization comes
about through the residua of embodied participation in practices which are
themselves reproduced through that participation.
The practice of
fishing on Vindö is a product of local and historical circumstances which
constrain and provide resources for the activities entailed in fishing. That is, the cultural process here has put
in place diverse, and complexly embedded, resources for conducting the modern practice
of fishing. To follow just one trajectory
of this embedding for a moment: the patterns of herring migration specify
constraints upon the matter of where fishermen spend their time, because
fishermen have large boats which require catching a lot of fish (to meet debt
payment schedules), because institutions such as story-telling in the harbor
and in church set up expectations about how one feels properly a part of the
community, because the community is economically self-dependent and
geographically isolated and has historically prospered due to patterns of
herring migration.
Each one of the
waypoints along this trajectory has, in turn, its own entailments. Again, to take just one example: large boats
are the locations where fishermen share experiences, these experiences form the
basis for reminiscing and (thereby) constructing meaning and value out of
current and past work routines, work routines in the operation of fishing
entail a distribution of authority which (although hidden behind institutional
mechanisms such as the equal sharing of income among all crew members
regardless of individual expertise) give rise to the value of owning ones own
boat or owning a larger boat.
This "web" of
links is a cultural process—or part
of one—and is instantiated in the activities of real people. No claims are being made that this web
constitutes a "superorganic" entity or structure of society which somehow
exists independent from the on-the-ground activities of the people. On the contrary, the links of this process
are instantiated in real constraints involving real materials (e.g., flesh,
blood, metal, money, speech, and fish) and the significance of these for real
people through the acts of cognition they perform.
[1]Of course, this prediction becomes vastly more complicated by the computer's ability to overcome human limitations (e.g., speed in symbol manipulation in the world) which make possible "successful" imitation of human performance through alternate mechanisms (e.g., exhaustive or heuristic searching of possible moves in chess programs).
[2]See, however, Quinn and Strauss (1993) for a different treatment.