Page 145 - Deep Learning

128 Creativity

solver more prone to react to an accidental hint, should it happen to appear in
the environment. They called this the memory-sensitization hypothesis.
69
Patrick Langley and Randolph Jones have specified a computational
mechanism for fortuitous reminding that adds yet another twist. In their
70
version of this idea, an impasse is particularly likely when the components of
the problem representation have many connections in memory that go off in
many different directions; this is called a high fan or a large branching factor.
A high fan rapidly dissipates activation as it spreads and thereby lowers the
probability that activation reaches the crucial knowledge elements. Fortuitous
reminding might remedy this situation if the fan is smaller in the opposite
direction, from the crucial knowledge elements toward the problem represen-
tation. Activation then might spread all the way, with the consequence that the
concepts crucial for the solution and the problem appear simultaneously in
consciousness. This differential fan hypothesis has not been tested empirically
but was found to work well in computer simulations.
Paradoxically, fortuitous reminding explains the beneficial effect of a
pause in a way that makes the pause itself irrelevant. The fortuitous event
could happen in the very first second of the pause, and it would be as helpful
as if it happened the next morning. Nothing happens during the pause except
that attending to other matters and engaging in other activities provide oppor-
tunities for events that were not part of the initial representation of the prob-
lem to influence the distribution of activation over long-term memory. There
are many sources of potentially relevant events: changes in the environment,
remarks by co-workers, and so on.
There is no reason to doubt that fortuitous reminding can affect prob-
lem solving in the manner described. However, this principle is not a general
theory of incubation effects, because it cannot explain those cases of incuba-
tion in which no such fortuitous event occurred. The prevalence of this type of
incubation effect in real life is not known.

Resolving impasses: Laterality in spread of activation?
M. Jung-Beeman and co-workers have advanced a hypothesis about impasse
resolution that relies on differences in how the two halves of the brain encode
information, and consequent differences in the spreading of activation. The
71
basic idea is that if nodes are densely connected, there is a large fan, and acti-
vation does not travel very far from its source node. In a conceptual network,
this means that a source node will very quickly activate many other nodes, but
only closely related ones. If nodes are sparsely connected, activation will travel
farther and activate more remotely associated nodes, albeit with a slight delay.