Page 262 - Deep Learning

Error Correction: The Specialization Theory 245

This explanation for cognitive flexibility implies that the question of how
learners acquire new skills and the question of how they adapt and transfer
already established skills from one task or set of circumstances to another are
one and the same. Contrary to long-standing concepts in cognitive psychology,
there is no separate problem of transfer; more precisely, there is no separate
problem of acquisition. People learn new skills by transferring their prior skills
to new situations. A successful transfer theory is therefore also a successful
theory of acquisition, but the reverse is not true. Theories that explain transfer
by adding a transfer process to an acquisition theory are unlikely to be correct.
Adaptation, not blank slate acquisition, is the primary case.

THE PROBLEM OF TUTORING

The natural sciences have taught us to expect accurate scientific theories to
generate successful practical applications. For theories of learning, the obvious
domain of application is instruction. The fact that instruction can work is so
familiar as to seem self-explanatory, but it poses the question of how utter-
ances from one person can enter the mind of another person and, once there,
facilitate learning. How does instruction achieve its beneficial effect? A related
question is how unassisted learning is related to learning from instruction.
All mammals learn, and, as a group, mammals learn better than other
51
animals. Primates, in turn, learn even more effectively than other mammals.
If the genetic basis for neural plasticity dates from the point in time at which
hominids branched off from other primates, then the hominid evolutionary
lineage emerged with a genetic basis for superior learning already in place
some 5 million years ago. If our capacity to learn is so ancient, then it is plau-
sible that unassisted learning is the primary mode and that learning from
instruction emerged later.
How much later? It is plausible that instruction did not begin to make a dif-
ference for human survival until the emergence of language. One can instruct
by example, gestures and nonverbal feedback, but the full power of instruc-
tion is not realized unless instructor and pupil can talk. We do not know with
certainty at which point hominids began to talk, but the archeological and
paleontological records provide one suggestive clue: There are few if any traces
of representations and symbols – carvings, graphic signs scratched on bone,
paintings, statues – by members of any hominid species before our own. It
52
is possible that the use of articulated symbol systems, including language,
became possible as a result of a small number of evolutionary changes in the
brain of archaic Homo sapiens or its immediate predecessor. If so, language,