Page 365 - Deep Learning

348 Conversion

spontaneously created for that domain. If a person carries with him some 2,500
latent conflicts, and the contender is superior to the resident theory in only 1%
of cases, the person has the potential for 25 deep conceptual discoveries, far
more than most people experience in their lifetime.
Competition on the basis of task performance only settles the issue of rela-
tive utility. Utility is, in turn, one of the factors that determine a person’s confi-
dence in a theory or belief system. The more successful the person is, the more
confidence he will have in his theory. Little would change in the resubsump-
tion theory if the two variables of utility and confidence were collapsed into a
single variable. Although the theory would gain in parsimony, common sense
suggests that confidence in a belief is strongly influenced by other factors over
and above its estimated utility. The nature of the source of information affects
our confidence; the possibility of formulating a theory mathematically might
increase confidence for some; others are more prone to believe in theoretical
mechanisms that they can visualize than those they cannot; and so on. For
these reasons, to collapse utility and confidence carries too high a theoretical
cost, so the two variables are kept separate throughout.
It is plausible that the probability of a switch in truth value is a function of
the ratio of the person’s confidence in the contender theory to his confidence in
the resident theory. Conversion occurs when the confidence levels cross over
so that the person’s confidence in the contender theory becomes stronger than
his confidence in the resident theory. At that point the relation between the two
theories has been reversed, so that Th(A+B) has become the standard or default
choice in any situation in which both theories apply. But there might be a certain
inertia that moves the switching point off-center, so to speak, and there is no
guarantee that the relation between the confidence ratio and the disposition to
switch truth value is linear. For present purposes, it is enough to hypothesize that
the ratio between contender and resident confidence is the main determinant of
a switch in truth value and that the former is, in turn, a function of the person’s
level of success when using either theory to solve everyday cognitive tasks.
According to the resubsumption theory, change is driven by successes
rather than by failures. It is the fact that the contender theory turns out to
apply to the target domain and that the applications are better or more suc-
cessful than those of the resident theory that sets the conditions for change.
According to this view, a belief might be revised even though it never encoun-
ters any anomalous, contradictory or falsifying evidence. Conflict is not pro-
duced by flaws and failures but by the creation of an alternative that turns out
to be superior. The disposition to convert depends on how well the contender
works, not how poorly the resident fares.