Page 58 - Deep Learning
P. 58
The Nature of the Enterprise 41
triggering conditions pertains to timing: A person is not contagious until some
time after being bitten, and the germs cannot live indefinitely in the mosquito
before being transferred to the next victim. The triggering conditions for the
dissolution and formation of atomic bonds are even more precise, turning on
energy levels, the number of electrons in an atom’s outer shell and the so-called
ideal gas configuration. The specification of triggering conditions differenti-
ates a disciplined, generative explanation from an ad hoc one. Without trig-
gering conditions, the theorist has too much freedom to postulate whatever
sequence of basic processes will yield the observed outcome.
Sometimes a scientific debate is more about the triggering conditions
than about the processes themselves. Evolutionary biologists all agree that
natural selection is the mechanism of speciation – the appearance of a new
species – but they have never stopped debating the triggering conditions. A
key proposal, allopatric speciation, says that speciation happens in small, geo-
graphically – and hence reproductively – isolated populations that are subject
to different selective pressures in their different areas. After having evolved
40
in different directions, members of the two populations can no longer inter-
breed if the geographic barrier eventually comes down. Allopatric speciation
is not an alternative to natural selection but a specification of the conditions
under which natural selection might produce a new species. Other biologists
also embrace natural selection as the key mechanism but argue for differ-
ent triggering conditions for speciation. The small size of so-called founding
populations, mutations that directly affect reproductive behavior and extreme
selective pressures at the edge of the species’ normal range have all been pro-
posed as factors that contribute to the reproductive isolation that natural selec-
tion needs to be able to push two or more populations in different directions.
The important lesson is that a change mechanism and its triggering condition
are two distinct parts of an explanation. A hypothesized change mechanism
predicts different consequences, depending on how its triggering condition is
formulated, so the accuracy of an explanation can be improved by revising the
triggering condition for a given change mechanism.
In short, componential explanations explain a change of large scope as the
cumulative result of multiple changes of small scope, produced by basic processes
controlled by particular triggering conditions. I suggest that this is how we under-
41
stand change generally: We break it down into a sequence of smaller changes.
What is the nature of the unit changes in the case of cognitive change?
We infer that cognitive change has occurred on the basis of changes in what a
person says or does. If he is asked to name the capital of New Zealand at some
point in time t , and answers “Auckland,” and if he answers that same question
1