Page 431 - Deep Learning

414 Notes to Pages 75–79

by the emergence of what we now call information sciences and information
technologies. Hoffman and Deffenbacher mention statistical information the-
ory, cybernetics (a.k.a. control theory), signal detection theory and computer
programming. They could have added game theory and the theory of decision
making. These developments loom large in the reminiscences of the pioneers
(Miller, 2003; Newell & Simon, 1972b). Thus, the war effort prompted the cogni-
tive revolution both by presenting new problems to think about and by inspiring
new ways of thinking about them.
43. See Conway and Siegelman (2005) for a biography of Norbert Wiener and a
history of the rise and fall of cybernetics. Wiener’s original work, Cybernetics
(Wiener, 1948), is too technical for any but mathematically competent readers.
44. Miller, Galanter and Pribram (1960), especially Chapter 2, “The Unit of Analysis.”
In the authors’ own words, they were pursuing “the ‘cybernetic hypothesis’,
namely, that the fundamental building block of the nervous system is the feed-
back loop” (pp. 26–27).
45. Powers (1973).
46. Newell and Simon (1972a).
47. See Newborn (2003). Newborn (2000) writes: “Deep Blue used a brute force
search that was very different than the narrow highly-directed heuristic searches
[prominent scientists of the 1950s and 1960s] imagined necessary to avoid the
exponential growth of the search space” (p. 28).
48. See Kaplan and Simon (1990). Others have applied the heuristic search theory
to explain creativity in more complex domains, e.g., musical composition (Alty,
1995).
49. If two diametrically opposite corner squares are cut off from a checkerboard, can
the remaining 62 squares be covered exactly by 31 dominos, if one domino is a
rectangle the size of two squares? The answer is negative because the two cut off
squares are necessarily of the same color, while a domino the size of two squares
necessarily covers one square of each color. Hence, 31 dominoes cannot cover 30
squares of one color and 32 of another.
50. Some authors deny that any special explanation is needed. Weisberg and Alba
(1981) write: Empirical studies “seem to indicate that the solution behavior for at
least two of [the classical insight] problems can be understood in a straightfor-
ward manner: People apply their knowledge to new problems, and if their knowl-
edge is not directly useful, they try to produce something new that will solve
the problem through a straightforward extension of what they know. No exotic
processes, such as sudden insight, are involved” (p. 189). The same perspective
is expressed in Weisberg (1986, 1993). See also Perkins (1981): “What is special
about the mental processes that explain sudden insight? The answer seems to
be: very little” (p. 71).
51. Newell and Simon (1972a) began modern research on problem-solving strat-
egies by describing strategies for a class of letter arithmetic puzzles (given
DONALD + GERALD = ROBERT; which digits should be substituted for the
letters to make this a correct addition?), for a class of formal logic problems and
for chess. During the following two decades, researchers identified strategies in
wide variety of task domains. The fact that strategies are domain specific, i.e., that