Page 434 - Deep Learning

Notes to Pages 90–95 417

memory capacity is related to analytical problem solving but not to insight
problem solving.
11. The insight sequence is superficially similar to the four-stage theory of illumina-
tion proposed by Graham Wallas (1926). The main difference is that the insight
sequence describes problem solving without a pause, while Wallas’s stages include
an incubation period in which the problem solver rests. Incubation is discussed
further at the end of this chapter.
12. See, e.g., Durkin (1937), Fleck and Weisberg (2004), Maier (1931), Metcalfe (1986),
Metcalfe and Wiebe (1987) and Ohlsson (1990c).
13. Brain-imaging techniques enable researchers to capture moments of insight; see
Bowden and Jung-Beeman (2003); Bowden, Jung-Beeman, Fleck and Kounios
(2005); Jung-Beeman et al. (2004), Kounios et al. (2006); Luo and Knoblich
(2007); Luo, Niki and Knoblich (2006); and Sandkühler and Bhattacharyya
(2008).
14. The classical self-report is Poincaré’s autobiographical account of a mathemati-
cal discovery, already discussed in Chapter 3 (Poincaré, 1908/1952), but similar
events have been reported by many others; see Ghiselin (1952) for a collection of
testimonies from artists and writers and Hadamard (1949/1954) for an analysis
of self-reports from mathematicians. Examples are mentioned in Chapter 5; see
Table 5.1.
15. Using a strict criterion of cessation of all activity to identify impasses, Fleck and
Weisberg (2004) found only a single example of the full insight sequence (impasse,
restructuring and successful solution) among 34 subjects who attempted to solve
Duncker’s Candle Problem. However, 13 additional subjects experienced both
impasse and restructuring but did not solve the problem within the time limit set
by the experimenters. Depending on which number is used, these data estimate
the prevalence of the insight sequence as either 3% or 41%; at either frequency, the
phenomenon is of interest.
16. The idea that perception is an interpretive process and that consequently there
are no theory-free observations – instead, observations are theory-laden – was
discussed by Hanson (1965), who used the term “theory loaded” as well as “the-
ory laden”; only the latter seems to have survived in philosophical discourse. The
concept that observations are theory laden is widely accepted in the philosophy
of science. Brewer and Lampert (2001) analyzed this concept from a cognitive
point of view. A classical psychological work that emphasizes the complexity
of perceptual processing is Neisser (1967). The point is equally visible from an
Artificial Intelligence perspective (Marr, 1982).
17. Textbooks on sensation and perception make this abundantly clear (Levine,
2000; Wolfe et al., 2006). See also Biederman (1987) and Marr (1982) for con-
structive theories of particular aspects of vision.
18. “The whole visual system can be considered a long series of filters, with
each stage in the system responsible for extracting a particular aspect of the
visual world and passing this aspect on to the next stage” (Wolfe et al., 2006,
p. 40).
19. Wolfe et al. (2006, pp. 93–94).
20. Kintsch (1988).