Page 361 - Deep Learning

344 Conversion

with the beliefs that are active at the time the information is received, while
the beliefs that contradict it are dormant. The resident and contender theo-
ries grow separately and in parallel through monotonic belief formation, each
within its own context.
Parallel, monotonic growth can result in a latent conflict because surface
features are poor predictors of essences. Seemingly distinct and separate phe-
nomena and domains sometimes exhibit deep similarities. Hence, the range of
applicability of a belief system cannot be determined by inspecting the beliefs
themselves. Building an informal theory for domain A, the person might
thereby unwittingly also build a theory of other domains B , B , B , … that do,
1
2
3
in fact, share the same underlying structure, but the act of acquiring a theory
for A does not in and of itself reveal those other applications. A person might
form a theory that in fact applies to some domain B without being aware that
it applies to that domain.
How often might this seemingly unlikely event happen? Recall that a per-
son might possess 500,000 individual beliefs. Belief systems vary in size, but
suppose they contain an average of 100 or so beliefs; then a typical person will
possess 5,000 distinct belief systems or local theories that guide thinking in
particular domains (health, money, sports, etc.). A latent conflict involves two
beliefs or belief systems, so the number of possible latent conflicts is approxi-
mately 5,000 , or 25,000,000. How often does it happen that two domains of
2
experience that seem to be distinct nevertheless are related in such a way that
a single theory can apply to both? Intuition suggests that such cases are rare.
How rare? If only 1% of all pairs of domains share the same structure, there are
250,000 such pairs in a person’s head. In how many of those cases are the two
belief systems in conflict, according to some background theory? If only 1% of
all pairs of belief systems that apply to the same domains are in conflict, there
are 2,500 latent cognitive conflicts waiting to be noticed in the average head.
This back-of-the-envelope calculation is not a serious mathematical model,
but it illustrates that even though people do not (intentionally) construct mul-
tiple theories for the same domain of experience, and even though two belief
systems that apply to the same domain are not necessarily in conflict, it is nev-
ertheless reasonable to believe that people carry around with them multiple
latent conflicts between their beliefs.
In short, the process that sets the stage for belief revision is that a per-
son responds to some domain of experience A by forming an intuitive theory,
Th(A); that Th(A) happens to apply to some other domain B as well; and that
Th(A) is incompatible with the person’s resident theory Th(B) for that domain
according to some background theory Th(0). Due to the impossibility of