Page 424 - Deep Learning

Notes to Pages 41–45 407

explanatory even if the units are not causal laws. I do not claim that explanation
in terms of chains of component transitions is the only type of explanation, or
that this concept resolves all issues in the study of scientific explanation, only that
it accurately represents how change is typically explained in science. Empirical
data show that it also describes commonsense explanations of biological change
(Ohlsson, 2002).
42. Einstein and Infeld (1938) summarized what they called “the mechanical view”
in these terms: “In mechanics the future path of a moving body can be predicted
and its past disclosed if its present condition and the forces acting upon it are
known. … The great results of classical mechanics suggest that the mechanical
view can be consistently applied to all branches of physics, that all phenomena
can be explained by the action of forces representing either attraction or repul-
sion, depending only upon distance and acting between unchangeable particles”
(p. 67). In chemistry, the main type of change is the formation and dissolution of
bonds between atoms: “Dalton’s recognition that a chemical reaction is a process
in which atoms merely change partners, with reactants one combination of atoms
and products another combination of the same atoms, is still the all-pervasive
foundation of explanation in chemistry” (Atkins, 2004, p. 112). This is not to deny
that at more specific levels of description, the concept of an atomic bond turns
out to be quite complex (Silvi & Savin, 1994).
43. I have not been able to locate a single, conceptual review of all types of erosion
mechanisms, but Montgomery (2002), Stallard (1995) and Trenhaile (2002) pro-
vide windows onto the technical literature.
44. Suggestions for such single, all-encompassing laws of learning are not lacking;
see, e.g., Taatgen (2002, 2005).
45. Compare the summary of the Gestalt work in Ohlsson (1984a) with Kaplan and
Simon (1990).
46. Compare Abelson’s (1959) and Kelman and Baron’s (1968) lists of modes of conflict
resolution with those of Chinn and Brewer (1993) and Darden (1992). Although
these lists do not map onto each other one-to-one, they capture approximately
the same concepts. Janoff-Bulman (1992) applied these concepts to the resistance
of trauma patients to their altered circumstances.
47. See Anderson and Bower (1973, Chap. 2), for a review of associationism as a psy-
chological theory.
48. Gagné (1965) identified eight types of learning and hence “eight sets of con-
ditions under which changes in capabilities of the human learner are brought
about. The implication is that there are eight corresponding kinds of changes
in the nervous system which need to be identified and ultimately accounted
for” (p. 57). The view in Ohlsson (2008a) is similar. What has changed over
time is the conception of the types of learning and (hence) their exact triggering
conditions.
49. For two examples of close analyses of individual learning events, see Schoenfeld,
Smith and Arcavi (1993) and VanLehn (1991, 1999). D. Kuhn (1995) and Siegler
and Crowly (1991) have reviewed the so-called microgenetic method in the con-
text of cognitive development.