Page 402 - Deep Learning

Elements of a Unified Theory 385

occurrence of multiple conditions go a long way toward explaining the low
probability of non­monotonic change.
These factors cannot be the whole story, because they affect everyone
equally. But some individuals create more than others, behave more flexibly
under changing circumstances or show higher willingness to revise their beliefs
in the face of new evidence. such individual differences must be explained by
factors that vary from person to person. one traditional approach attributes
novelties in thought or action to some unanalyzed power called “creativity,”
“flexibility,” “mindfulness,” “tolerance of ambiguity” and so on, or by its
opposite (e.g., “perseverance,” “rigidity,” etc.). Variations in the disposition
to restructure are explained by saying that an individual with a long résumé
of novel beliefs, ideas or strategies possesses more of this power – however
labeled – than individuals with fewer such accomplishments. This way of deal­
ing with individual differences in the disposition for non­monotonic change is
circular unless the key explanatory factor – “creativity,” “flexibility,” “mindful­
ness” – is analyzed in terms of mental processes and mechanisms.
it is possible that variations in cognitive flexibility – to choose one of the
popular terms – can be explained by intrinsic, possibly innate variations in
quantitative parameters of the cognitive architecture. For example, people’s
brains might differ with respect to the total amount of activation that is avail­
able to distribute among the relevant cognitive processes. Another possibil­
ity is that people differ in the threshold for feed­forward from a processing
unit. A high threshold means that only highly activated processing units get to
propagate their computational results. This presumably leads to highly effec­
tive projection of past experience. A low threshold means that even units with
little activation can propagate their results. The consequence might be a high
rate of conceptual fluency, ease of making remote connections or other types
of behaviors often associated with cognitive flexibility. The size and organi­
zation of the various processing systems, especially the density of feedback
links, is another factor that might influence the propensity for non­monotonic
change. set the values of such parameters to certain values and the probability
of non­monotonic change will turn out to be low; set them to other values and
the probability will be higher. There is little reason to expect every brain to be
born alike, so the idea that the cognitive architecture exhibits slight quanti­
tative variations from person to person is highly plausible. We do not know
how much of the individual differences in the disposition for non­monotonic
change might be explained by variations of this sort.
it is consistent with the notion of minor differences in the cognitive archi­
tecture that the evidence for the effectiveness of creativity training programs