Page 82 - Deep Learning

The Production of Novelty 65

a combinatorial theory of creative thinking must specify, at a minimum,
the cognitive elements that can enter into combinations and the process by
which they are combined. The generative power that makes combination
an attractive principle for explaining the production of novelty also poses a
conceptual puzzle. an educated adult knows some 50,000 words, and hence
24
approximately that many concepts. He is capable of producing many billions
of conceptual combinations, only a few of which would stir anyone’s interest. a
satisfactory theory must specify how the mind tames the combinatorial explo-
sion by giving direction to the creative process, lest the interesting combina-
tions be lost among the rest.
The French 19th-century mathematician and scientist Henri Poincaré pro-
posed a combinatorial theory of creativity in mathematics, based on his own
25
experience. He suggested that conscious work on a problem activates ideas
and concepts related to the problem. When the problem is set aside for lack of
progress, unconscious processes form random combinations of the activated
ideas. The combinations vary in their aesthetic qualities. Combinations with
higher mathematical beauty pass into consciousness and become objects of
deliberate reflection. Poincaré claimed near-infallibility for this process. The
most beautiful combination almost always turns out to contain the solution
to the problem at hand, although he did admit the occasional exception. The
key feature of Poincaré’s theory is that it breaks down the creative process into
two structural components: an unconscious combinatorial process that gener-
ates new idea configurations and an equally unconscious evaluation process
that selects those that are to be subject to conscious reflection. in computer
science terminology, this is called a generate-and-test process, with the uncon-
scious processing playing the role of generator while conscious reflection plays
the role of evaluator. The generation and testing are related to the transition
between the unconscious and conscious levels of mind, so this is a form of pro-
jection from one system level to the next; see Figure 3.1.
The structure of Poincaré’s theory recurs in later proposals, albeit with dif-
ferent articulations of the generative and selective components. in Psychology
of Science, Dean Keith Simonton also claimed that new ideas come to mind as
chance combinations of mental elements (sensations, concepts, schemas, etc.);
“… the fundamental generating mechanism in scientific creativity involves
26
the chance permutation” of mental elements. as in Poincaré’s theory, a small
number of the new combinations are retained, but configurations are selected
on the basis of their stability instead of their mathematical beauty. a stable
configuration functions as a single cognitive unit and therefore can enter into
yet more complex configurations; Figure 3.1 illustrates this concept. Simonton