Page 376 - Deep Learning

Belief Revision: The Resubsumption Theory 359

various types of events and the reinforcements, or lack thereof, that accompany
them. Logical and physiological descriptions are worlds apart, and yet must in
the final analysis be understood as referring to one and the same event. Belief
revision engages as perhaps no other issue in cognitive science both sides of the
mind-brain duality, and cognitive psychologists are stuck with the task of mint-
ing a coin that has intention and cause, meaning and process, as its two sides.
The Resubsumption Theory takes a modest step in this direction by
departing from the idea, central to most prior theories, that belief revision
is grounded in the evaluation of evidence, and by explicitly linking cognitive
utility, a material concept, to the assignment of truth value, an intentional con-
cept. According to the theory, the type of cognitive conflict that drives conver-
sion is incompatibility between already acquired theories or belief systems, not
between theory and evidence. Outside academia, the worth of a belief system
is a function of its cognitive utility – that is, how successful it allows a person to
be when performing whatever cognitive tasks he undertakes. The careful, con-
scious, explicit and public accounting of the logical relations between hypoth-
eses and observations that characterizes theory change in science is a highly
developed and culturally supported practice that does not mirror, process
for process, the basic cognitive mechanisms that make conversion possible.
32
Estimation of cognitive utility on the basis of environmental feedback gener-
ated by action, not the processing of observations through Modus Tollens, is
the key process by which a person adopts one belief rather than another.
Utility and truth are linked because we are disposed to believe that what
works is true. Over the long run, people will adopt the beliefs that enable
them to be successful, although they might not be aware of this fact, and the
adoption of a belief with high utility might occur gradually and outside con-
sciousness. People do not decide to believe in the same way they decide to buy
a house. Instead, they find themselves believing such-and-such a proposition
that has proven useful in multiple contexts. This pragmatic imperative is pre-
sumably an innate part of our cognitive architecture. After all, we evolved to
be hunters and gatherers, not philosophers. Pragmatism might not be a good
philosophical theory of truth, but it is a plausible psychological theory of truth
value assignment. Perhaps this theory will turn out to work for psychologists,
33
in the sense of generating powerful explanations for cognitive phenomena. If
so, psychologists will no doubt come to believe it.