Page 333 - Deep Learning

316 Conversion

multiple attacks by several different research teams using different approaches.
in addition, a single anomaly is not sufficient reason to abandon a theory.
Anomalies have to accumulate before they trigger a search for a better theory.
Far from a single Modus Tollens inference, the rejection of a resident theory is
a temporally extended, cumulative process.
Kuhn provided a much needed naturalistic turn in the study of scientific
theory change, but his account is as incomplete as Popper’s. Both accounts
focus on successful conversions so they do not explain how falsifying observa-
tions or anomalies avoid being swept under the conceptual rug by peripheral
change mechanisms. How do anomalies accumulate if each anomaly is dis-
missed as it arrives? A second problem is that neither Popper nor Kuhn pro-
vided any insight into the origin of the alternative theory. Popper stated that
this question would remain unanswerable; the title to his book* notwithstand-
55
ing, there cannot be, he claimed, a logic of scientific discovery. Kuhn took a
slightly softer stand: “… how an individual invents (or finds he has invented) a
new way of giving order to data now all assembled – must here remain inscru-
56
table and may be permanently so.” He did not try to provide a theory of the
origin of the alternative theory. But a theory change is typically a change from
one theory to another, and it would be surprising if the strengths of the alter-
native theory were not among the reasons for rejecting the resident theory. By
ignoring the second half of the problem of theory change, Popper and Kuhn
failed to solve its first half.

The Many roads since Structure

Kuhn’s grand narrative of scientific progress swept through the academic
and popular cultures, making “paradigm shift” as much of a household word
57
as “cognitive dissonance.” Although his account was met with skepticism
among professional philosophers and historians of science, it served as a
source of inspiration. The idea of basing theories of cognitive change on pat-
terns extracted from the history of science was, and remains, appealing to cog-
nitive scientists, historians and philosophers.
The story of the post-Structure period is one of proliferation, but even
when the purpose is to depart from Kuhn, accounts of theory change tend to
postulate the same triggering condition as the one Kuhn proposed. For exam-
ple, in Human Understanding, published in 1972, stephen e. Toulmin set out
58
to create an evolutionary alternative to Kuhn’s description. He describes the
*
The Logic of Scientific Discovery