Page 335 - Deep Learning

318 Conversion

“the elimination of incorrect schemata”; “the generalization of schemata”; and
finally, “there is explanatory extension” in which an explanatory schema is
embedded within a larger schema. These phrases are suggestive of plausible
62
processes, but Kitcher does not explain how any of these processes work. For
example, how is a schema eliminated? Presumably, this is triggered by evi-
dence that the schema is inaccurate, but Kitcher provides no explanation of
how resistance to such information is overcome. The emphasis on explanatory
power is an important contribution, but Kitcher does not propose any new
idea about how theory change comes about.
A similar problem adheres to nancy nersessians’ theory of model-based
reasoning. The cognitive basis for model-based reasoning in science is a
63
capacity to build mental models, dynamic knowledge structures that can be
executed in the mind’s eye to provide an internal simulation of the event, phe-
nomenon or system that those structures refer to. As a simple illustration of
the power of running mental simulations, consider the following scenario: A
tired traveler returning home leaves his key ring in his travel bag, and in a
moment of mindlessness puts the bag in the basement; the next morning, the
question arises, where are the keys? The inference that the keys are in the bag
and the bag is in the basement, therefore the keys are in the basement as well is
difficult to carry out in deductive logic, but the conclusion is obvious to anyone
who visualizes the situation and simulates the sequence of events in the mind’s
eye. nersessian identifies five distinct components of model-based reason-
ing: visualization, abstraction, modeling, analogy and thought experimentation.
once again, this is a plausible and interesting theory of scientific reasoning.
The problem for present purposes is that there is no explicit hypothesis about
when and how scientists employ model-based reasoning (as opposed to any
other type of reasoning), nor for how model-based reasoning enables a scien-
tist to overcome his resistance to change.
Accounts of theory change tend to implicitly assume that the new, con-
tender theory is better than the resident theory. But new is not necessarily bet-
ter, so how does the scientist know which of the competing theories to believe?
Toulmin asserts that conceptual variants are evaluated by asking, “Given the
current repertory of concepts and available variants, would this particular
conceptual variant improve our explanatory power more than its rivals?” But
64
he provides no details as to how scientists carry out this comparison. A spe-
cific explanation is provided by the theory of explanatory coherence developed
65
by Paul Thagard and others to explain the outcomes of cognitive conflicts.
That theory applies when there are two competing hypotheses and a set of
facts, each of which might be consistent or inconsistent with either hypothesis.