Page 219 - Deep Learning

202 Adaptation

steps that produced positive outcomes. (g) Learning from error, or revising
rules that produced negative outcomes so as not to generate such outcomes
in the future. (h) Shortcut discovery, or identifying redundancies and repeti-
tions in execution histories and deleting unnecessary steps. (i) Optimization,
or anticipating processing demands on the basis of statistical regularities
extracted from the environment.
The point cannot be stated strongly enough: The theory of stable behavior
and the list of proposed learning mechanisms, sorted by the nine types of infor-
mation, constitutes a scientific theory of how skills are learned during practice.
Although this is stated as a theory for the first time here, it is not my the-
ory. it is a collective accomplishment of the community of skill acquisition
researchers. The steady growth of this theory over a century of research is one
of the great success stories of cognitive psychology. it remains for cognitive
psychologists to claim it.
The theory does not only answer the question of mechanism, but it also
answers the questions about sufficiency, necessity and gradual change. practice
is sufficient for learning in the sense that practice automatically engenders
improvement, at least in the beginning of practice. The reason is that the
learning mechanisms, like most cognitive processes, are not under voluntary
control but run outside consciousness. to practice is a conscious and deliber-
ate activity that one can choose to engage in, but a mutation in a production
rule is caused by a mental process that occurs at a particular moment in time
because its triggering conditions are satisfied, not because the person has an
intention or volition to improve. improvement in skill is a side effect of inten-
tional activity. practice is sufficient to cause cognitive change in the same sense
that ingestion is sufficient to cause digestion.
practice is necessary because improvements cannot arise out of noth-
ing. Learning incorporates new information into the strategy-under-con-
struction, and that information has to come from somewhere. The paradox
of learning by practicing is that the learner generates that information him-
self by attempting to perform the target task. Action triggers the construc-
tion of initial rules, and the execution of those rules produces positive and
negative feedback. By performing the target task repeatedly, the learner
produces an execution history that might reveal shortcuts, accumulates
information about statistical regularities in the environment and stores
solutions and answers in memory. in short, practice is necessary because
the response of the task environment to the learner’s actions is the ultimate
source of information about the task; no action means no information and
no improvement.