Page 121 - Deep Learning

104 Creativity

better. In a completely familiar situation, the search process is so constrained
that there is only a single option to consider in each successive problem state.
We then say that the person knows what to do; he has mastered the task. The
successes of chess-playing computers against the best human players illustrate
how effective analytical problem solving can be in stable task environments,
37
particularly when backed by the memory and speed of electronic hardware.
Analytical thinking is nevertheless limited in ways that are unrelated to pro-
cessing capacity. As mathematical derivations from a set of axioms cannot lead
to a theorem that contradicts those axioms, so analytical thinking cannot find
any solution that is not included in the search space in which it moves. To
reach all possible solutions, the mind must be able to jump over its own axi-
oms and move sideways from one search space to another.

The Theory of Insight

If the theory of heuristic search through a solution space is at least approxi-
mately accurate as a theory of analytical thinking, how do the alterations in
mode and tempo in the insight sequence arise? Why do people encounter
impasses on problems which they are, objectively speaking, capable of solving?
Once they have entered an impasse, by what processes do they break out? 38

The Causes of Unwarranted Impasses

If a person is faced with an unfamiliar problem, he cannot know with cer-
tainty which interpretation of it will turn out to be most useful. In a turbu-
lent and imperfectly known world, there is no guarantee that the biases laid
down in the course of experience are predictive of which knowledge elements
are most useful for solving a current problem. The perceptual encoding rules
acquired in past experience will nevertheless execute and construct some
initial representation of the problem. An unfamiliar problem might give rise
to an initial representation that accidentally activates knowledge elements
that are not, in fact, useful for constructing the solution. Once activated, those
knowledge elements constrain further retrievals from the knowledge store via
inhibitory links as well as by hoarding the available activation. Consequently,
activation might not spread to those knowledge elements that are crucial
for the solution. The result is an unwarranted impasse; the problem solver
possesses the knowledge needed to solve the problem but fails to retrieve it.
This explanation was already formulated by Woodworth in 1938: “When, as
must often happen, the thinker makes a false start, he … falls into certain