Page 91 - Deep Learning

74 Creativity

Theorizing about this dependency began with the work of edward L.
Thorndike, another of Poincaré’s contemporaries. His advisor was William
James, co-founder of the pragmatist school of philosophy. The collaboration
between one of the last scions of the philosophical era in psychology and one
of the founders of the experimentalist tradition produced one of the most
influential doctoral dissertations in psychology. 39
Thorndike studied how cats, dogs and chickens solved problems, with
the purpose of laying a foundation for understanding the higher cognitive
functions in people. His method was to shut the animals into problem boxes.
The boxes were not comfortable or inviting and the animals were typically
observed when they were hungry, so they tried to escape. This required some
action the animal was physically able to perform – pull a string, press a bar,
and so on – but which was unfamiliar. Thorndike observed that the animals
went through their repertoire of actions until they happened to hit upon the
one that opened the box. Over multiple trials, the animals came to execute
the correct action sooner and with more precision, until they performed it
effectively and immediately upon being shut in the box. although the term
trial and error had already been used by others and occurs only once in
Thorndike’s dissertation, his study has become the iconic illustration of this
concept. 40, 41
Thorndike’s work was an important stimulus for psychological research
on learning; the behaviorist school was the response. although the behav-
iorists’ theoretical concepts proved insufficient to explain human cognition,
their focus on observable behavior provided a useful corrective to prior theo-
rizing. Poincaré and other theorists assumed that the relevant cognitive unit
for explaining creativity is an idea. This view is plausible in mathematics.
elsewhere, the production of novelty is often better thought of as a matter of
action; the issue is what to do about a problem or how to proceed in an unfa-
miliar situation. The novelty produced by Thorndike’s animal subjects was not
a concept but a behavior, the action that got them out of the box. trial and
error is generate-and-test applied to action instead of ideas.
Thorndike’s main contribution was to formulate an enduring principle
about the links between successive actions. He claimed, and later formalized in
his Law of effect, that the choice of action is molded by what he called afteref-
fects. if an action is followed by a satisfier – a reward – the animal’s inclination
to perform the action grows stronger. if an action is followed by an annoyer –
punishment – the inclination diminishes. By blocking the action that caused
them, annoyers provide an opportunity for other actions to be tried, as long
as the individual persists in his efforts. a sequence of actions is a sequence of