Page 347 - Deep Learning

330 Conversion

the interaction between truth and utility, between epistemic and instrumental
considerations, at the center of belief revision.

BELIEFS, BELIEF SYSTEMS AND COGNITIVE CONFLICTS
Precise specifications of change mechanisms require a description of the entity
that is changing, so an explanation of conversion needs to start with a char-
acterization of beliefs and belief systems. This task was begun in Chapter 9.
Routine, monotonic belief formation events are assumed to follow the two
principles of Ubiquitous Encoding (all received information is encoded and
some subset is stored in long-term memory) and Truth as Default (new infor-
mation that is consistent with prior beliefs is regarded as true unless there
are reasons for doubt). I begin by extending the account from Chapter 9 with
additional concepts and distinctions.

The Dimensions of Beliefs
A belief consists of a proposition and a handful of parameters. The proposition
is the content of the belief. A proposition, when expressed in language, corre-
sponds approximately to the meaning of a declarative sentence, like State Street
runs north and south. Propositions are symbolic structures stored in memory.
The parameters associated with a belief indicate a person’s stance toward
the relevant proposition. The first parameter of a belief is its truth value. To
assign a truth value to a proposition is to take a stance as to the veridicality
of that proposition. Although the values true and false have received most
attention from philosophers, people use many different linguistic markers to
specify the status of propositions: “certain,” “likely,” “plausible,” “probable” and
their opposites are commonly used, and “don’t know,” “undecided” and “not
sure” should be used more often than they are. For present purposes, the truth
value parameter need only take the two values truth and false. A second
parameter records the person’s level of confidence in the belief. If this variable
ranges from 0 (no confidence) to 1 (complete confidence), then [P, truth,
.90] represents the fact that a person is almost certain that P is true, while
[P, false, .25] means that he has tentatively decided that P is false.
Cognitive representations are sometimes referred to as beliefs and some-
times as units of knowledge, but cognitive scientists, laypersons, philosophers
and psychologists draw the distinction between belief and knowledge in dif-
ferent ways. In statements like “I don’t know for sure, but I believe it is so”
and “I don’t just believe it, I know it,” the distinction appears as one of degree