Page 348 - Deep Learning

Belief Revision: The Resubsumption Theory 331

of certainty, which makes beliefs and knowledge elements two distinct sets of
entities. Philosophers, on the other hand, define knowledge as justified true
belief, which makes a person’s knowledge a subset of his or her beliefs (because
not all beliefs are both justified and objectively true). Neither of these usages is
the one adopted in this book.
The normative concept employed in philosophy needs to be distinguished
from the naturalistic concept of knowledge. In psychology, we routinely refer to
what a person thinks is true as that person’s knowledge, regardless of its objec-
tive truth. For example, developmental psychologists write about children’s
knowledge of, for example, living things; educational psychologists write about
students’ prior knowledge about some subject matter; and cognitive analyses of
science frequently refer to the knowledge of scientists in antiquity. Once we put
objective truth to the side and focus on what a person thinks is true, there is lit-
tle difference between saying that “he believes that X” and saying that “he knows
that X.” The terms “knowledge” and “belief” are in this usage interchangeable,
as are “informal theory” and “belief system.” These observations are not new;
I am merely codifying the standard psychological usage.
Cognitive scientists who build Artificial Intelligence systems deviate from
these other usages by referring to all mental representations as “knowledge
representations.” But not all representations encode beliefs. It is not obvious
what it would mean to assign truth values to, for example, analogies, goals,
images, rules or scripts. In this usage, “knowledge” is a superordinate category,
so a person’s beliefs constitute a subset of his or her total set of knowledge rep-
resentations, in direct opposition to the philosophical usage. As an illustration,
consider the fact that after close and repeated study of John R. R. Tolkien’s vol-
umes about Middle Earth, I know something about that place, its inhabitants
and the events surrounding the famous magical rings. My mental representa-
tions of these events do not differ in any essential way from my knowledge of
the set of events that we refer to as World War II. I can, for example, engage
in counterfactual reasoning in either domain (What would have happened if
Frodo had been captured with the ring inside Mordor? What would have hap-
pened if the German Luftwaffe had continued to bomb Royal Air Force airfields
in 1940 rather than switched to London and other cities?) Yet my views of the
Battle of Helms Deep and the Battle of Britain obviously differ; I believe the
latter took place but not the former. It is nevertheless as natural to speak about
my knowledge of Middle Earth as to speak about my knowledge of World War
II. In the cognitive science usage, my knowledge encompasses all my mental
representations, while my beliefs are those knowledge representations that I
designate as true, which makes my beliefs a subset of my knowledge.