Page 336 - Deep Learning

The Formation of Belief 319

Given a set of quantitative strength values for the links between these knowl-
edge units, algorithms adapted from neural network models enable the result-
ing network to compute which of the competing hypotheses provides the most
coherent account of the given body of facts. This idea has been embodied in a
computer model called eCHo that provides intuitively reasonable outcomes
for a variety of cases. The emphasis on theory-theory conflicts instead of the-
ory-data conflicts is useful, but the theory only explains why, given two com-
peting explanations for a set of facts, one explanation might be preferred over
the other. it does not explain how scientists arrive at a state in which they have
two fully developed theoretical proposals to choose between.

Kuhn in the Woodwork

Kuhn’s grand narrative of scientific progress, controversial though it was,
established a consensus practice for the cognitive study of theory change in
science: An account of theory change should be consistent with the practices
of scientists as captured in historical case studies and contemporary field stud-
ies. Theory change is triggered by cognitive conflicts, interpreted as failures to
solve scientific problems. A shift from one theory to another is a temporally
extended process, although it is brief compared to the preceding and following
periods of relative stability and so deserves the label “scientific revolutions.”
The source of the novel theories is to be found in the (deliberately unanalyzed)
individual and social creativity of scientists. The many descriptions of theory
change since the publication of Kuhn’s Structure do not represent roads away
from these ideas but are puzzle-solving efforts within the Kuhnian paradigm
of theory change as a response to the accumulation of anomalies.
But any conflict-based narrative of scientific progress requires an account
of how resistance is overcome, as well as an explanation of why anomalies
sometimes lead to resistance and sometimes to conversion. neither has been
forthcoming. in addition, it is highly plausible that the creation of a contender
theory and its acceptance as better than the resident theory are closely related.
After all, the contender is constructed with awareness of the weaknesses of
the resident theory and with the intent of overcoming those weaknesses.
The rejection of the resident theory and the construction of the new theory
are not separate processes. Pressing the argument further, it is plausible that
the strengths of the contender theory are among the reasons for dissatisfac-
tion with the resident theory. if so, the temporal order implied by conflict-
driven theories – anomalies lead to dissatisfaction, which in turn leads to the
search for an alternative – is up-ended. information about the strengths of the