Page 34 - Deep Learning

The Need to Override Experience 17

Finally, to reap the benefits of our abilities to remember, process and proj-
ect past experience, a person has to be able to use the resulting expectations
to plan and act accordingly. People maintain complex systems of goals and
multiple alternative plans for reaching them, and although we often have occa-
sion to react on the spot to unexpected events, most of our actions throughout
a normal day are parts of some plan or another. Like memory, generalization
and transfer, the process of planning has been extensively studied by cognitive
psychologists and the reality of this cognitive process is not in doubt. 46
In short, the received view of our cognitive system in general and our
capability for learning in particular is tightly interlaced with the clockwork
view of reality; see Figure 1.3. Because past experience provides clues to the
regularities that underpin the appearances, we can handle future situations
by storing episodes in memory, and then process them to extract the underly-
ing regularities, project those regularities onto the future to generate specific
expectations, and plan and act accordingly. I suggest the term empirical induc-
tivism for this view of the how mind and world interact, to distinguish this
psychological theory from work on induction in artificial intelligence, mathe-
matics and philosophy.
Empirical inductivism runs like a subterranean river under the landscape
of cognitive research, watering not only studies of learning but almost all other
subfields as well. It is the perfect partner for the clockwork view of nature. As
seemingly perfect partners sometimes do, these two bring out the worst in
each other. An inductive learning theory justifies the self-understanding of
the clockwork scientist, and the findings of the latter in return confirm the

Figure 1.3. The clockwork interpretation of the epistemological situation.