Page 418 - Deep Learning

Notes to Pages 27–30 401

10. J. J. Gibson (1966) advocated an ecological approach to perception which assumed
that there is information in the environment and that the task of our perceptual
systems is to identify it and to receive it; no complicated processing is required.
For example, ambient light contains information about relative distance in the
form of degree of clarity of the retinal image. Close objects have sharp edges
while distant objects look fuzzy. Register this information and relative dis-
tance judgments follow. This type of analysis applies to auditory perception as
well (Gaver, 1993). In a different take on this theme called rational analysis, J. R.
Anderson (1989, 1990) proposed that if the human mind evolved to be rational,
or nearly so, then an analysis of the environment and the tasks that the mind
performs is sufficient to predict regularities in behavior. If there is only one best
way to perform a task and if our minds evolved to perform it that way, then we
should be able to predict behavior by assuming that the mind performs the task
in that way. All we need to explain behavior is the right analysis of the task. Some
regularities regarding memory can indeed be derived from the assumption that
processes such as retrieval conform to statistical regularities in the environment
(Anderson & Schooler, 1991, 2000). Pirolli (2005) analyzed information foraging
on the Web in this manner. See Oaksford and Chater (1998) for yet other appli-
cations of rational analysis. The question is how much insight we gain into the
workings of the mind from predictions of behavior, even accurate predictions,
derived from environmental regularities. These lines of work are closely related
to Artificial Intelligence work on embodied cognition (M. L. Anderson, 2003).
11. See Cole, Engeström and Vasquez (1997) for a collection of papers that expli-
cate and exemplify the anthropological approach to cognition. D’Andrade (1995)
has written a history of this line of inquiry. For a statement of the related situ-
ated cognition approach to learning, see Clancey (1997) and Lave and Wenger
(1991). This approach is not without its critics; see Vera and Simon (1993) and the
response by Greeno and Moore (1993).
12. My favorite example is the study of navigation practices aboard U.S. Navy ships
by Hutchins (1995).
13. Pylyshyn (1986, pp. 6–7).
14. Varela, Thompson and Rosh (1993) and Clancey (1997) have tried to formulate
nonrepresentational theories of cognition. The skepticism regarding represen-
tational theories is not new. Furth (1968, 1969) scrutinized the role of represen-
tation in J. Piaget’s developmental theory and concluded that “… knowing in
Piaget’s theory is never a mere matter of representation” (p. 150).
15. Visual mnemonic techniques were invented in antiquity (Yates, 1969). The earli-
est extant description of the technique known as the system of loci is by Cicero
in De Oratore (On the ideal orator, II:350–360), completed in 55 b.c. The key
idea in this mnemonic technique is to first visualize a series of familiar places or
locations (loci), to form visual images of objects that correspond to the content
to be memorized, and then “place” those images in the locations. To retrieve the
content, imagine “visiting” the locations in order and encountering the visual
reminder “placed” in each. In the late 16th century, the Jesuit missionary Matteo
Ricci gained the confidence and respect of Chinese scholars and civil servants by