Page 228 - Deep Learning

Error Correction: The Specialization Theory 211

white and the sun will rise tomorrow, and it becomes hopelessly inadequate
when confronted with counterfactual assertions (if we had more money, we
could afford a car), probabilistic statements (there is only a 50–50 chance that
this stock will increase in value) and abstract principles about theoretical enti-
ties (two electrons cannot occupy the same orbital shell).
There is no good evidence that the propositional view is psychologically
accurate. Nothing is more commonplace than the casual cocktail party com-
ment that people don’t always reason logically, do they? – a comment that wraps
the speaker and the hearer in a shared mantle of superior rationality. Cognitive
snobbism apart, this attitude is supported by systematic inquiry. Laboratory
experiments have demonstrated several types of illogical behavior on the part
of experimental subjects: atmosphere and framing effects (the wording of
an argument influences its acceptability), confirmation bias (people tend to
accept as valid arguments that end with conclusions that coincide with what
they believe), difficulties in processing negated statements and a disposition
to accept certain fallacious argument patterns as valid. The hypothesis that
13
people reason by applying truth-preserving inference rules to propositions is
contradicted by a large body of empirical evidence. If the propositional model
is not correct, it is possible that the function of declarative knowledge is not
primarily to support description, explanation or reasoning.
The question arises as to what the function of our vast database of declara-
tive knowledge might be instead. A novel departure is to cast declarative knowl-
edge as prescriptive rather than descriptive. The cognitive function supported
by declarative knowledge might not be deduction, description, explanation
or prediction but judgment. Declarative knowledge enables a person to assess
an object, event or situation. In particular, it enables him to guess whether his
current situation lies on the path toward his goal. While practical knowledge
enables the generation of action, declarative knowledge enables the evaluation
of action outcomes.
This view invites the idea that action and judgment are dissociated because
they rely on distinct knowledge bases. Consequently, a person might possess
the (declarative) knowledge required to judge a performance as incorrect but
lack the (practical) knowledge required to perform better. In the words of
D. N. Perkins in The Mind’s Best Work:
A fundamental fact – maybe even the fundamental fact – about making is
that critical abilities are more advanced than productive abilities. We can
discriminate differences we can do little to produce. … For many reasons,
the test for a certain property requires much less of us than producing
something with that property. 14