Page 205 - Deep Learning

188 Adaptation

Thorndike displayed the time it took individual animals to escape from a box
as a function of trial number. He proposed a two-mechanism theory known
as the Law of Effect, which says that organisms weaken their disposition to
perform an action if that action is followed by negative consequences, but
strengthen it if the action is followed by positive consequences. At the same
time, William L. Bryan and noble Harter’s analysis of training data for Morse
code operators stimulated interest in the acquisition of complex, real-world
30
skills. Although Hermann Ebbinghaus had already published curves for the
memorization and forgetting of lists of syllables, Thorndike and Bryan and
Hartner were the first to plot what we now call learning curves for skills. table
6.2 highlights select milestones since this beginning. 31
Learning became the major theme of the behaviorist movement, conven-
tionally dated as beginning with John B. Watson’s 1913 article, “psychology as
the Behaviorist Views it.” during the 1913–1956 period, “experimental psy-
32
chology” and “learning theory” became almost synonymous in the United
States, but the dominant experimental technique was to study the memori-
zation of lists of letters, syllables or words. R. S. Woodworth’s 1938 review of
experimental psychology included a chapter on practice and skill that sum-
marized a mere 27 studies that tracked learning in complex tasks like archery,
33
telegraphy and typing. Woodworth discussed the shape of the learning curve
and stated the idea that skill acquisition goes through phases: “. . . it appears
that the typical process of learning a complex motor act proceeds from whole
to part and back to whole again.” He did not propose a rigorous theory. 34
World War ii prompted psychologists in Britain and the United States
35
to focus on complex skills. The war posed novel psychological problems,
such as how to train anti-aircraft gunners. A second transforming influence
was that psychologists came to work alongside engineers, mathematicians
and scientists who were in the process of creating new information technolo-
gies. Code breaking, long-range direction finding, radar tracking and other
information-processing problems led researchers to realize that information
can be measured and processed in objective and systematic ways and that it
is possible both to build information-processing systems and to view humans
and animals as examples of such systems.
After the war, norbert Wiener at the Massachusetts institute of technology
envisioned an interdisciplinary science to be called cybernetics, which was to
study complex information-based systems, encompassing humans, machines
and animals, in terms of feedback circles. Like Thorndike’s Law of Effect, the
36
principle of feedback describes how behavior changes in response to positive
and negative action outcomes. The idea of replacing the stimulus-response