Against Cognitivism

From: Lucas, Melody (
Date: Tue Dec 05 1995 - 12:52:09 GMT

Chapter 1: Cognitivism as an approach to cognition.

> "The human is an animate organism with a biological basis and an
> evolutionary and cultural history. Moreover, the human is a
> social animal, interacting with others, with the environment, and
> with itself. The core disciplines of cognitive science have tended
> to ignore these aspects of behaviour. The results have been a
> considerable progress on some fronts, but sterility overall, for
> the organism we are analysing is conceived as pure intellect,
> communicating with one another in logical dialogue, perceiving,
> remembering, thinking when appropriate, reasoning its way through
> well formed problems that are encountered during the day. Alas that
> description does not fit actual behaviour."
> (Norman, 1982, p266)

C&S complain that psychology has become cognitive psychology,
ignoring creative thinking and conciousness. The book seeks to
address the problem of the domination of psychology by cognitive
psychology, and of cognitive psychology by cognitivism.

Cognitivism is the belief in "internal cognitive processes" in which
rules and representations are defined in symbolic or prepositional
form. There are four problems inherent in cognitivism:
-Problem of solopsism: how can the knower ever reach beyond internal
representations to the reality they are supposed to represent?

-Problem of development: how can a system of formal rules ever be
flexible enough to capture the mutuality between a growing organism
and its richly structured and changing environment?

-Problem of relevance: how does anyone following a rule know when to
apply that rule? (The so-called frame problem).

-Problem of meaning: how do symbolic representations attain their
semantic status?

These problems do not seem to bother cognitivism because there is "no
alternative" to this paradigm.

Chapter 2: Mechanism and Romanticism- A. Still

Still gives an overview of Descartes' method of aquiring scientific
knowledge, describing his insistence that one should only rely upon
the 'clear and distinct' ideas of the kind found in mathematics. In
this view, the proper function of language is to describe the world
by attaching labels to ideas with the ultimate achievement of
representing reality uncovered by science.

As a result, the physical sciences are held in high esteem at the
expense of the social sciences, leading to partiions of academic
disciplines. Cognitive psychology has identified with the hard
sciences by drawing methods from the laboratory and vocabulary from
computer engineering, neglecting areas of human experience whih
cannot be assimilated in this mechanistic way.

William James purported that the stream of thought was not a set of
discrete, static ideas, but a flow of vague, fleeting feelings of
relationships. He criticised the two main dualisms in cognitivism:
mind/body and organism/environment. Nowadays the dualism is
*N.B. This is the dualism I complained about last week in the seminar;
Stevan, I thought we were supposed to be avoiding dualism, yet you
seemed surprised I even queried it.

Romanticists claim that the function of language is referential,
providing an unambiguous and precise description of reality, whether
at the level of theory or observation. Personally, I think that
there will alwys be a certain amount of abstractness (O.K. I know
that isn't a proper word.. sort of proves my point, really) in
Still asks whether the vagueness in experience, which cognitivists
are so keen to dispell, is an obstacle or has a function i its own
right. Then the chapter became a bit more of a boast of literary
quotes etc. However, he did put a rather apt quote from Coleridge
(a romanticist) here:
> " In the infancy of the Human Mind all our ideas are instincts; and
>language is happily contrived to lead us from the vague to the
>distinct.....this distinction between the instinctive approach
>towards an Idea, and the Idea itself, is of high importance in
>Methodizing Art and Science" (Coleridge, 1934,p6)

Coleridge believed that vagueness is not a kind of unfocussed state
to be corrected, but it is an essential part of experience. James and agreed
 with this (see James, Principles of Psychology, 1890). As a dualist
he provided the basis for verbal construction of the dual poles of
mind and matter, and of subjectivity and objectivity.

CONCLUSION: the pragmatist, social constructionist and
perceptual realist accounts provide the most important alternatives
to cognitivism.
>" If there is a single event that should be celebrated by the
> opposition to cognitivism, it is William James's reversal of
> Cartesian priorities by his reinstatement of the vague." (Still,
>1991, p21.)

O.K. I'll write about chapter 3 later on.

Chapter 3: On Not Doing and On Trying and Failing (Don Mixon).

Explanations in terms of underlying rules that 'generate' behaviour
have been an imortant feature of modern cognitive theory. Critics
stipulate that a distinction must be made between activities that are
actually governed by explicit rules and those which are merely
describable in terms of rules.

John Dewey: tried to carry out simple everyday activities in novel
ways (e.g.trying to sit down with his body in a different alignment
to usual) and realised that skilled actions are more than a matter of
following rules. He knew exactly what the rules were yet was unable
to execute them.

Mixon: believes that if cognitions are to be acceptable explanations
for doings, they should also account for not-doings. Cognitive
explanations appear plausible because of the seemingly direct
connectio between a cognition (e.g. command, intention) and the
execution (when common tasks are used as examples). But cognitive
explanations lose even surface plausibility when one looks at
behaviour in its full sense. The connection between command and
execution IS NOT THERE.

Dewey suggested the notion of COGNITIVE HABITS (where the term
'habit' refers to an acquired pre-disposition to ways or modes of
response, not to particular acts or acts which are necessarily
repeated). Dewey purported that habits (unlike attitudes which are
purely cognitive) involve the entire psychophysical being. He
stipulates that ways and styles of thinking are just as skill-
dependent as ways of movingor speaking (you cannot think like a Zen
master just by deciding to do it).

MIXON: said that "the alternative to cognitivism would be anything that
puts cognitive studies in their place".
" Mind should be studied, but it should be studied where it lives"
(p.34) The claim that behaviour can be explained only by appeal to
internal cognitive processes cannot be taken seriously. You need
external contingencies too.

F.M. Alexander (quoted by Dewey) presented several problems:
1. Faulty sensory appreciation: habit influences, even seriously
distorts perception e.g. your senses tell you that your neck muscles
are relaxed even when they are, in fact, tense.
2. Futility of "end- gaining" (going directly to the end without
mastering the means to gain it).

Humans are not, even by analogy, machines which are impervious to the
environment and controlled by cognitive processes. The only way you
can do anything ina certain way is to gain this skill, and this
usually takes time and practice.
i.e. COMMAND / INTENTION-------------------------> END

The alternative to cognitivism is to give up the notion that any part
can describe the whole.

Chapter 4: Frederic Bartlett and the Rise of Prehistoric Psychology
- A. Costall

This will be very brief because most of it was irrelevant (and a bit
tedious, to tell the truth)
Bartlett was the pioneer of two conflicting approaches. His early
writings presented a social anthropology of cognition and
emphasised the person within the group. Later on, through influence
of new technology in WW11, he took a mechanistic point of view.

I'll concentrate on the first, which includes:
NATURALISTIC APPROACH: Experimenters should not lose sight of the
world beyond the lab. (Neisser, 1976, discussses this in detail)
e.g. the research on memory for digit sequences was relevant to the
design of telephone codes.
is driven by the criteria of functional adaptiveness to social
conditions and context of everyday existence (cf. cognitivism:
mentality as individual, with component stages. levls and processes).

Basically; Bartlett in his early days studied observable human
behaviour. This was a form of behaviourism but "differed from
behaviourism in its dogmatic form". In conclusion, he stated that
tendencies are correlative o the situation and do not imply the
existence of any underlying, permanent mental structures.

Chapter 9: Graceful Degradation: Cognitivism and the metaphors of the
computer - A. Costall.

Cogntive Psychology in Question (Still and Costall, 1987) Identified
modern cogntivism with the representationalist theory of mind, i.e.
"the assumption that, for scientific purposes, human cognitive
activity must be described in terms of symbols, schema, images, ideas
and other forms of mental representations" (Gardner, 1987, p39).

The new wave of cognitive psychology seems to take the "parallel
distributed processing" approach (PDP). This is based on artificial
networks of simplified neural connections (moving away from symbol-
based theories and " embracing the radical insights of the
opposition" such as gestalt and Gibson).

Costall said that representationalism seems lke an alternative to
mechanism, but really it is just a supplement of S-R theory. PDP is
an alternative to this.

Cognitive psychology represents itself as a revolution to
neo-behaviourism but there are significant continuities between them.
e.g. both share the same concept of ultimate data (observable
behaviour) and share the same general notion of theoretical
explanation ( info----> processing---->output )
                   ( S ------------------------> R )

Cognitive psychology relies heavily on metaphors in order to be
explicit e.g. the use of the vocabulary of the computer metaphor

>ml: surely isn't this a contradiction in terms? Surely a metaphor by
definition isn't explicit, or am I splitting hairs?

Information Processing: "Programs have much in common with theories
of cognition since both are descriptions of the vicissitudes of input
information" (Neisser, Cognitive Psychology, 1967,p8) However,
Neisser stated later his rejection of simulation and concluded that the
analogy of computers should stay as a metaphor (cf. Pylyshyn:
"Cognition is literally a species of computing",1989, p52).

Rules and representations: The distinction between modern cognitive
and neo-behaviourist accounts of mediational processes is their
appeal to representational accounts expressed in terms of
propositions. A problem arises when representation and rule-
following become internalized and privatised in psychological theory
so that they seem to explain everything.

>S&C: One might say that to make computers intelligent, computer scientists
adopted a psychological metaphor, but psychologists suppose thst the
properties that they attribute to the computer reside within the
machine. So far, all that seems to be at stake is the literal or
metaphoric status of the rules and representations in the human case.
Researchers haven't questioned whether computers actually follow
rules and interpret representations. Pylyshyn's claim; "the notion of not a metaphor but part of a literal description of
cognitive activity" (1979, p435) can only be justified if the
computer itself literally performs rule-governed transformations on
intentionally symbolic expressions, interpreted by the computer

Dennett supposed that any system of representation could be broken
down into smaller and smalle sub-components to the point where they
become so trivial, they can be handled by "stupid homunculi". This
didn't eliminate the problem of intentionality and intelligence, it
just re-located the problem through the appeal to an analyst (who
breaks the task down).
This is similar to Turing's attempt to formalize and automate
thought: he was aware that the representational status of the
computer is DERIVATIVE (i.e. symbols within a machine are
representations for the user, not the computer).
>C&S: If computers appear to prove that meaning ca exist within the
self-enclosed realm of representations devoid of any interpreter, it
is because we forget ourselves (who invest the symbols with our

Representationalism is just one aspect of modern cognitivism. An
important related assumption is that both representations and the
processes that operate upon them are, for the most part, unconcious.
The fact that a system can be modelled by means of a program does not
imply that ths system itself is running a program. Costall gives the
example of planetary motion: We model planetary motion by means of
calculus, but this doesn't mean the planets are solving calculus
problems in order to stay on course!

The computer metaphor is a new reaction to dualism as it binds 'mind'
and brain together as software and hardware, yet allows us to
separate them.
>S&C: How can the human body be compared with computer hardware
(whose functioning is solely constrained by instructions)?
>ML: Surely this depends on what one refers to as 'instructions'.
They may take the form of electrical impulses within the brain or
different concentrations of biochemicals over a semi- permeable
membrane......or not? I suppose these do not necessitate
intentionality, though.
Regarding the Turing test, Russell (1984) argued that the test
reflects the capabilities of the user i.e. it's not the machine that
passes it, but the people who fail it.
>ML: Maybe so, but isn't this grossly irrelevant?
>ML: For computers to model humans, they would have to be able to
learn independently (i.e. without being programmed by a user). Is
there an example of a computer doing this?
>ML: In the seminar we discussed how hardware is irrelevant.
However, the only way of interpreting a program is through an item of
hardware, (in most cases a VDU) . How can you separate them

PDP now threatens the long-standing alliance between computers and
cognitive psychology. Schneider (1987) reckons there is going to be
a BIG paradigm shift.
PDP rejects programs where explicit symbols are processed in a
sequential order. Instead, there is simply a network of
interconnected nodes where the strength of the connections changes in
accordance with various learning rules. This is a continuation of an
early emphasis in computer science on systems modelled upon neural
Important Discoveries: (Hopfield,1982): Neural networks with
symmetrical connections between nodes will settle down into stable
configurations and hence be able to store knowledge.
(Rumelhart, Hinton and Williams, 1986) the technique of the " back
propagation of errors" provided the solution to the problem of how to
train the more powerful networks that include an additional layer of
"hidden units" between both the input and output layers.

PDP moves away from representationalism in favour of a cognitive
model comprising a rich network ofconnections. Knowledge is implicit
and distributed across the entire pattern of interconnections.
Knowledge is IN the connections rather than in a change of state of
any isolated unit of the system.

Modern cognitivism has obscured the distinction between following a
rule and merely acting in accordance with a rule. In contrast to
explicit rule-formation, PDP states that learning is the aquisition
of connection strengths which allow a network of simple units to act
as though it knew the rules. McClelland, Rumelhart and Hinton (1986)
believe that PDP is not an alternative to the rule-based explanation;
the rules are, at best, an approximate description of what is really
happening at the level of the microstructure of the connections.

There is no distinction between hardware and software in PDP, just
connections which are in some sense hardware AND software. For
example, PDP talks about networks 'settling' or 'relaxing into'
solutions, rather than inferring or even computing them. There is a
heavy emphasis on the analogy to the brain.

The most prominent objection to PDP (by the way, this is parallel
disributed processing; it seems such a long time ago that I
identified it!) isits faailureto providea distinctive cognitive or
psychological level of theory to stand between stimulus and response.
PDP and traditional cognitive theory still subscribe to
organism / environment dualism as they both disregard the fact that
the organism acts upon and transforms the environment.

CONCLUSION: Traditional cognitive theory attributes a property of the
entire system to a single location, i.e. cognition in peoples' heads.
PDP states that cognition is a transformative and social process and
is radically distributed.
>ML: Lucky I've got my right elbow to work this one out, then.


Metaphors have a distancing effect i.e. we treat things that people
have made as if they were completely independent of us.
>ML: This is what I was trying to put over the other week.But I
expect I'm wrong again, anyway.

N.B. Costall mentioned a valid point in his conclusion which I think
is of practical importance but has rarely been mentioned as a factor
promoting the mechanistic point of view: AI was originally funded by
American Defence in order to obtain a lead in technology. Reearchers
are largely restricted to the same sources of funding. I'm not
saying that this is an argument against cognitivism, I just think
it's partially responsible for the 'no alternative' argument.
End of chapter.....see ya!

This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:58 GMT