The Act of Creation (83 page)

novelty
is of course also achieved when
the cat learns the open-sesame trick in the puzzle-box. To argue that
the cat's novel response was acquired by Trial and Error, the chimp's
by Insight, is to argue in a circle, since novelty itself is supposed
to serve as a differential criterion.

 

 

Thus Sultan's Eureka processes,
once we have got rid of thinking in
S.-R. schemata
, are interpretable in terms of the same theory which
covers all lower forms of learning. They make a spectacular impression,
because (since the separate skills which had to be integrated into
the new skill were well-exercised items in his repertory of habits),
the problem to be solved was just one step beyond the limits of that
repertory, and all was set for a single spark to trigger off the fusion.

 

 

At the opposite end of the learning scale, the dog in Pavlov's
laboratory is not equipped with pre-existent rules of the game which
could be combined with each other; it must
construct a new code of
rules
, starting more or less from scratch. Therein lies the main
difference. The dog must start with an agonized reappraisal of which
environmental events are relevant and which are not; then extract and
codify the recurrent invariant features from the stimuli promoted to
significance; then discriminate between finer features within those
features. The rat must piece together, bit by bit, his cognitive map
of the maze; the cat must gradually extract, by empirical induction,
the rules of Thorndike's game from a surrealistic universe.

 

 

The 'missing link' in between the cat and Sultan is provided by Koko.
He does not have to start from scratch; he has already played with the
box; he has sat on it and pushed it about. Was his first successful
solution of the problem a random try? Certainly not. It had all the
'dramatic suddenness, smoothness, directness and definiteness' that
one can wish for. It was more than a 'provisional try', rather like a
hypothesis which carried implicit conviction; yet on the other hand, it
had been preceded by hesitant action along the correct lines which was
abandoned; and it was succeeded by forgetting all but a fragment of the
successful solution -- the fragment 'there is something about that box'.

 

 

The reason for this paradoxical behaviour is evident. Koko had to combine
two skills; the reaching-jumping-climbing skill M1, and displacing
the box to serve as a platform, M2. But M2 is not part of the chimp's
habit-repertory; in all Köhler's experiments with manipulating boxes
and putting them on top of each other, his chimps proved surprisingly
stupid.* Thus the skill of manipulating boxes is not a well-exercised,
'ready-made' item in Koko's repertory; and Koko is not really ripe for
the task set for him, because M2 is still too tentative and unstable to
become firmly attached to M1; linking did occur in a lucky moment, but the
link broke again. This description presupposes that the box-manipulating
skill was developing independently from banana-collecting, as a purely
playful occupation -- as it did in fact when Koko was made to play with
the box before the experiment started. And it further presupposes that
if Koko had been given sufficient time to become proficient in that
playful skill, then he would have become ripe for a true bisociative
act. Instead of this, however, he was led to form a 'premature linkage'
between boxes and bananas (cf. Book One,
IX
).

 

 

This interpretation differs from Köhler's 'appearance of a complete
solution with reference to the whole layout of the field'; or rather,
it specifies the condition under which such a 'complete solution' has a
chance to occur. But our interpretation is borne out by later experiments
by others, in which chimpanzees were given the same raking task as set by
Köhler. One chimpanzee out of six had previously used sticks in play; this
animal was the only one which had the 'insight' to rake in the food placed
outside the cage; the other five failed, although the stick was lying in
plain view. For the next three days all six chimps were given sticks to
play with. They used the sticks, as usual, to push and poke, but never
as a rake. Then the experiment was repeated -- and all six sticks turned
into rakes instantly. [15] Experiments with dogs reared in isolation
and then set 'insight situation' tasks, gave similar results. [16]

 

 

It may seem pedantic to lay so much stress on the independent primary
development of skills which are later combined' in the 'moment of
truth'. But the point does become relevant on higher levels. The
experimental sciences of electricity and magnetism developed
independently, and the discovery of electro-magnetic induction was a
truly bisociative act; in comparison to it the subsequent improvements of
electromagnets were a pedestrian procedure. The previous independence of
the cognitive structures which are made to fuse in the creative act is,
as we saw before, one of the criteria of originality; I shall return to
the subject in the final chapter.

 

 

 

The Ambiguities of Gestalt

 

 

The Gestalt theory was developed in Germany, before Wertheimer,
Köhler, and Koffka settled in America; and the original German
term for insight was 'Einsicht'. In a footnote on page 291 of the
English translation of
The Mentality of Apes
we read (my italics):
'The German word
Einsicht
is rendered
by both "intelligence"
and "insight"
throughout this book. The lack of an adjective derived
from the noun "insight", apart from other considerations, makes this
procedure necessary.

 

 

In this casual footnote we find the clue to the sad confusion which has
bedevilled the controversy from its beginning: Gestalt theorists used
the word 'insight' indiscriminately to mean either (A) intelligence,
understanding, judgement, knowledge in general, or (B) specifically
the acquisition of new understanding and knowledge under the sudden and
dramatic circumstances specified in the previous section. By equating
(A) with (B), the Gestaltists were led to regard (B) as the only type of
'intelligent' learning, everything else as 'blind' learning; and their
explanations of why and how learning of type (B) occurs become unavoidably
tautological. Thus Köhler writes:

 

We can, in our own experience, distinguish sharply between the kind of
behaviour which from the very beginning arises out of a consideration
of the structure of a situation, and one that does not. Only in the
former case do we speak of insight, and only that behaviour of animals
definitely appears to us intelligent which takes account from the
beginning of the lay of the land, and proceeds to deal with it in a
single, continuous, and definite course. Hence follows this 'criterion
of insight: the appearance of a complete solution with refrence to
the whole layout of the field. [17] (Köhler's italics).

 

But a few pages further we read: '
trying around
consists in
attempts at solution in the
half-understood
situation; and the real
solution may easily arise by some chance outcome of it, i.e. it will not
arise from chance impulses, but from actions, which, because they are
au fond
sensible, are great aids to chance'. [18] Köhler
further speaks of the 'dawning' of the solution and of 'good errors' --
that is, tries in the right direction vaguely sensed. Thus Köhler
admits that insight in the sense (B) is a matter of approximation, and
may be achieved in several steps, consisting of more or less 'sensible'
hypotheses and tries; yet in other passages he asserted the exact
opposite, namely that the criterion of insight was 'the appearance of
a complete solution . . . in a single continuous and definite course'.

 

 

To get out of this confusion, let me distinguish between two problems;
firstly, what constitutes understanding or Insight A; and in the second
place, how new understanding or Insight B is acquired. Regarding the
second problem, we have seen that the dramatic Eureka process is not
the rule, but rather an exceptional limit case; and that understanding
or Insight A enters to varying degrees into all forms of learning.

 

 

Let us examine for a moment the Gestalt school's answer to the first
problem: the nature of understanding or Insight A. On page 219 of
The Mentality of Apes
we read that it is based 'on the grasp of a
material, inner relation of two things to each other . . .'; 'relation'
being further defined as an 'interconnection based on the properties
of these things themselves, not a
frequent following each other
or
occurring together
'. The meaning of these formulations becomes
clearer in Köhler's later book (1949). There this 'material' 'inner'
relation between two things is expressed as 'our feeling of something
naturally depending on something else'. [19] 'Between the attitude
and its sensory basis we experience what in German is called ihr
sachlicher Zusammenhang . . .'; and
sachlicher Zusammenhang
is translated, with some hesitation, as 'intrinsic connection'. This
again is used synonymously with 'experienced determination' such as
that prevailing 'between a disease and its germ'. [20] Again: 'Here
not only the result is experienced, but also very much of its
why
and
how
is felt in just the actual context. Wherever this is the
case we apply the term
insight
.' But 'very much' is a relative
term, and its use as an all-or-none criterion -- 'wherever this is the
case' -- again confuses the issue. Turning to Koffka, we find that he
explains the difficulties confronting the cat in the problem-box by
the fact that the actions which it must perform to gain release are to
the cat 'objectively meaningless', that they have 'no sort of internal
connection with release', that they have no 'material relation' or
'intrinsic relation' with the opening of doors, and so on.

 

 

It should be clear by now that all these somewhat obscure terms are
shamefaced references to
physical causality
, and that the position
of the Gestalt school boils down to the tautology that the animal's
behaviour is the more intelligent the more insight it has into causal
relations. Nobody will quarrel with this statement; but it entirely
begs the central question of learning theory: namely, by what processes
and methods that insight into causal relations is
acquired
. The
loop in the puzzle-box, at the beginning of its training, means nothing
to the cat; at the end of its training it means escape. How is this
meaning acquired? Through trial-and-error learning, hypotheses, etc. The
problem can now be re-formulated as follows: can learning by trial and
error result in 'genuine' solutions, can it provide a correct, or true,
or meaningful representation of the causal connection between loop and
door? We can even go one step further and ask the same question with
regard to classical conditioning. If the dog could express itself in
Köhler's terminology, it would no doubt answer that the sound
of the gong 'signals' or 'means' food, that an 'objectively meaningful
connection', an 'intrinsic connection' or
sachlicher Zusammenhang
,
exists between the two. And this statement would entirely correspond to
fact, because in the laboratory universe this sequence
is
natural
law. Of course some connecting links are missing in the dog's inner model
of that law: the intentions of God Pavlov who has decreed it are unknown
to the dog. But such gaps occur on all levels of cognitive processes. When
the average citizen turns on his radio he has about as much insight into
the 'intrinsic connections' between the knob and the sound, the 'whys'
and 'hows', the 'interconnections based on the properties of the things
themselves', the 'total layout of the field' as Pavlov's dogs have.

 

 

Understanding is a matter of approximation. If we hold, with Craik, that
the basic achievement of the nervous system is derived from its power to
parallel or model external events [21], then the cat which has learned to
open the door by tugging at the loop can be said to have made a correct,
if crude, model of a causal sequence. The crudeness of the model is mainly
due to the fact that the rope connecting the loop and the door is hidden
from the cat's sight -- whereas under normal conditions the cat can 'see
what it is doing'. But this difference is one of degree not of kind; and
it is not justifiable to argue -- as Gestalt theorists have occasionally
done -- that because the cat cannot see the connection, it has no
'insight'. Because sight is our main sense organ, we have come to use
the words 'I see' as synonymous with 'I understand'. The Gestalt school
with its strong emphasis on visual perception has carried this tendency
to extremes, and thus came to believe that to have all relevant facts
of a situation or problem laid out before one's eyes is both necessary
and sufficient for its understanding. [22] In fact, of course, it is
neither. Rats learn to know a maze, and to form a mental map of it --
which amounts to as complete an understanding of the situation as one
could wish for -- without having been offered a bird's eye view of it.
Nor is seeing a
sufficient
condition for knowing in chimpanzees or
humans, even if the 'whole field' containing all the necessary clues
is laid out in full view. Thus Sultan establishes a visual connection
with the banana outside the cage by pushing one stick towards it with
a second stick, but that procedure does not testify to much insight. In
other experiments, where a string is attached to the banana and one or
more strings are laid out in the vicinity, Sultan will pull at random at
any of the strings, although he can clearly see which string is connected
to the banana, which is not.* The young child behaves in similar ways;
Piaget has called this phenomenon 'optical realism'. It 'consists in
considering things as being what they appear to be in immediate perception
and not what they will become once they have been inserted in a system of
rational relations transcending the visual field. Thus the child imagines
that a stick can draw an object because it is beside it or touches it,
as though optical contact were equivalent to a causal link. [22]
Adults are also quite often unable to understand how a simple mechanism
works although it is laid bare before their eyes. The visual concurrence
of all elements which belong to the problem facilitates understanding
(in various degrees according to species, see previous note), but does
not guarantee or imply it; it does not provide a 'direct and complete'
insight into the causal connections of the situation, nor, except in
certain cases, does it occasion a 'complete answer according to the total
layout of the field'. Even in relatively simple situations there are
always gaps in understanding the 'intrinsic material relations' between
things. Why can I pull but not push an object with a string? What are the
whys and hows of rigidity, flexibility, cohesion, etc.? Why is a hemp-cord
stronger than a paper-cord, and less elastic than a rubber-cord? Our
insight into the 'inner material relation of phenomena' is full of
missing links, so to speak; the model of the outside world which we form
in the matrix of our minds represents, by the very nature of that matrix,
not a point-to-point correspondence but a point-to-blur correspondence,
a more or less rough approximation. Not only this blurred microstructure,
but even the macrostructure of the universe, and the laws which govern
it, put us into much the same perplexed condition as the cat which
has to infer, by empirical induction, the laws which Thorndike made
in his wisdom. And the cat's behaviour in the box is, in fact, a first
approximation to the methodology of exact science as formulated by Craik:
'induction supported by experiments to test hypotheses'. [24]
It is necessary to remind ourselves of these truisms, because the
controversies in learning theory have almost made us lose them from
sight. The extreme wing of the Behaviourists has tried to banish the
concepts of understanding, memory, purpose, consciousness, hypothesis,
from the groves of Academe, to interpret trial and error as a random
process, and human induction as an equally mechanical affair. Classical
Gestalt theory sinned in the opposite direction; its attitude is
epitomized, e.g. in the passages from Köhler already quoted:
according to which true understanding can be derived only from 'an
interconnection based on the properties of the things themselves,
not
a
frequent following each other
or
occurring
together
.' To pay attention to the 'frequent following each other' or
'occurring together' of two events, which is so contemptuously dismissed
here is the very essence of inductive inference. To quote Russell again:
'Let there be two kinds of events, (a) and (b) (e.g. lightning and
thunder), and let many instances be known in which an event of the kind
(a) has been quickly followed by one of the kind (b), and no instances
of the contrary. Then either a sufficient number of instances of this
sequence, or instances of suitable kinds, will make it increasingly
probable that (a) is always followed by (b), and in times the probability
can be made to approach certainty without limit. . . .' [25] Instances
of a very suitable, clear-cut kind -- that is to say, situations for
which the animal is 'ripe', may then lead to inductive certainty derived
from a single occurrence -- the chick inferring that all caterpillars
of appearance (a) are accompanied by a disgusting taste (b).
The mistake of the classical Gestalt school was
to identify
trial-and-error learning with chain-reflex theory
-- a historically
understandable mistake since the two were lumped together in the
S.-R. schemata of Thorndike, Pavlov, Watson, Guthrie, etc. Yet the
history of science abounds with examples of brilliant discoveries
which were preceded by long periods of more or less fumbling tries in
half-understood situations. To try is to adventure; and to quote Russell
once more: 'If an induction is worth making, it may be wrong.' [26]
The same kind of bias led to the Gestalt school's uncompromising rejection
of any associationist theory of learning. Again this radical attitude
is historically understandable as a reaction against mechanistic
interpretations of associations as rigid neural connections, fixed
pathways, conditioned reflex arcs and the rest. But the Gestalt school did
not succeed in offering any valid alternative of its own for associative
memory. Köhler's trace theory, elaborated by Koffka, does not give
even a remote clue as to how a visual percept -- say a caricature of
Nelson, which is 'pictured' by an electrolytic field-current in the
optical cortex -- will activate, 'by similarity' the auditory trace of
the word 'Trafalgar'. The influence of past experience on the present is
minimized by Köhler wherever possible, and learning is virtually reduced
to spontaneous 'Insight B', where new knowledge emerges all in a piece
like Minerva in full armour from Jupiter's head.
According to Köhler, association has to be 'given up as a special and
independent theoretical concept. It is not more than a name for the fact
that organized processes leave a trace picturing their organization and
that in consequence of it reproductions are possible. . . . Our conclusion
is, that association depends upon organization because association is just
an after-effect of an organized process.' [27] 'Organization' in this
context means perceptual organization -- that is to say, the animal's
innate faculty to order its perceptions into Gestalt-configurations
which arise spontaneously as the 'experienced direct determination' of
'objectively meaningful intrinsic connections' based on the 'whys and hows
of the interaction of the properties of the things themselves' -- and
so on. Gestalt has become the ultimate panacea. As one critic remarked:
'The Gestalten thus become primary realities, existential ultimates,
in terms of which all events should be comprehended.' [28]
As so frequently in the history of science, a school of outstanding
achievements has succumbed to the magic power of a unifying formula.
That formula, in its turn, is based on a metaphysical assumption: the
existence of an a priori correspondence, or coordination, between the
physico-chemical processes in the nervous system and the events in
the outside world. Owing to Köhler's theory of 'psycho-physical
isomorphism', the sight of a square gives rise to a field current
in the cortex, and 'this cortical process must have the structural
characteristics of the square.' [29] The 'intrinsic', 'material', or
causal relations in the outer world are automatically mirrored by isomorph
electro-chemical Gestalt processes in the brain, and
'insight' turns
out to mean the spontaneous operation of the isomorph faculty
. Thus
the organism does not have to build a symbolic model of reality in
the nervous system by the processes of learning -- such as scanning,
coding, abstraction, gradual approximation through inductive hypotheses;
the isomorph model in the head is prefigured, potentially given in its
native perceptual organization, and need only be
activated
by
'spontaneous insight'. To quote Polànyi:

From this principle (isomorphism) it would follow that the whole of
mathematics -- whether known or yet to be discovered -- is latent
in the neural traces arising in a man's brain when he looks at the
axioms of Principia Mathematica, and that the physico-chemical
equilibration of these traces should be capable of producing a
cerebral counterpart (a coded script) comprising the entire body of
mathematics. [30]

Thus the metaphysical assumptions of Gestalt psychology lead us
(as Koffka somewhat reluctantly admits [31]) back to the kind of
'faculty psychology' abandoned about a century ago; and ultimately to
Plato; whereas Behaviourism leads us ultimately back to the atomism of
Democritus combined with the scepticism of Ecclesiastes. Köhler's
chimps look at the world through the 'eyes of the soul'; Hull's rats
are wired marionettes jerked about by a non-existent puppet player.
Putting Two and Two Together
Contiguity is the
Deus ex machina
of Behaviourists such as Watson
and Guthrie, and the bête noire of Gestaltists, who reject association
by contiguity as 'blind' and 'meaningless'.
'There is not a single example', writes Köhler, 'of an effect
produced by the interaction of two things or processes quite independently
of their properties. Nevertheless this is the character of the classical
law of association as we find it stated in most textbooks.' [32]
I cannot remember having seen anywhere the law stated in quite that
form, and Köhler's description is obviously a caricature; how can two
things interact 'independently of their properties'? Does not the term
'interaction' imply that properties of the interacting processes are
involved? Contiguity -- the overlapping of two events in spacetime --
of a regular, recurrent kind is the base on which inductive inference
can build. This applies, impartially, to the dog learning that the gong
is a signal for food, and to Kepler observing that the tides follow the
moon. Kepler concluded that there must be a causal connection between
the moon and the tides, although his ideas of gravity were of the most
erratic kind; and his was a correct approximation, followed by the
closer approximations of Newton and Einstein -- though we are perhaps as
far as ever from 'grasping the intrinsic, material relations' between
bodies acting at a distance. More often than not, the starting point
of the scientist's inquiry is 'post hoc, ergo -- let's hope -- propter
hoc'. Even in the rare limit case of the Eureka process, contiguity --
the simultaneous activity in the mind of the two matrices which are to
be integrated -- provides the link; and the provider may even be 'blind
chance' -- a fungus sailing through the window into Fleming's laboratory.
Learning, then, in the most general sense of the word, consists in putting
two and two or A and B together. It may be done gradually, by plodding
through hypotheses and eliminations; or all of a sudden, following upon a
single implicit try. A and B may stand for recurrent features abstracted
from a series of perceived events; or A may be a signal-pattern requiring
the correct choice of reaction B among other possible choices; A and B
may be codes of behaviour or universes of discourse, each complete in
itself and adequate for routine tasks; but to solve a certain new task
they must be put together. If the subject is ripe for the solution, the
putting together can occur in a single flash. But let me repeat that such
lightning inductions based on a single case are possible only if both M1
and M2 are well-established, flexible matrices which the subject knows
'inside out'. Only those chimpanzees discovered spontaneously the use
of a stick as a rake who had previously played with sticks; those who
had not, though of equal intelligence, failed to see the light.