Page 176 - Deep Learning

Creative Insight Writ Large 159

lead it out of an impasse. In the words of Max Planck: “A new scientific truth
does not triumph by convincing its opponents and making them see the light,
but rather because its opponents eventually die, and a new generation grows
up that is familiar with it.” 52
Impasse-breaking through replacement shares none of the specific prop-
erties of insight at the individual level. It is more gradual than punctuated,
and it does not require feedback. Instead, the collective gradually outgrows
its unhelpful constraints as it attracts individuals who do not share them. This
idea deserves a label; I will refer to this as the Principle of Replacement. It is
not a consequence of the redistribution of activation in individual insights but
is rooted in the essence of a collective – the fact that it consists of multiple
individuals – and the mortality of membership. A complete theory of creative
collectives should encompass the mechanisms of individual insight as well as
the change mechanisms that emerge at the collective level.

Summary
Collectives like individuals suffer ups and downs, periods of stagnation alter-
nating with periods of rapid progress. There are features of problem solving
that reappear at the group level for principled reasons, including heuristic
search, goal hierarchies and alterations in mode and tempo due to unwarranted
impasses and insights. The same mechanisms are operating: An impasse, no
matter how small, can block the progress of a collective project, no matter how
large. Inappropriate assumptions are potential causes. An insight of a single
individual can punch through like a butterfly effect and radically impact col-
lective problem solving. These scaling mechanisms preserve the massively
contingent and hence unpredictable nature of progress.
The theory of insight makes two specific predictions at the group
level: Groups whose members have diverse prior knowledge and groups
whose members accept mutual criticisms have a greater probability of avoid-
ing and resolving unwarranted impasses than homogenous groups and groups
in which criticism is considered bad form. The validity of the first prediction
is obscured by the possibility of process loss, but the second is supported by
empirical observations. As with scaling along the time dimension, scaling
along the dimension of collectivity only depends on a few gross features of the
relevant cognitive processes. It matters little how inappropriate presupposi-
tions act to constrain the search space, but it matters greatly that presupposi-
tions can have that effect. It matters little exactly how negative feedback acts to
relax constraints and thus allow a person to draw back to leap, but it matters