Page 399 - Deep Learning

382 Conclusion

range of different cognitive phenomena and they are grounded in prior empir­
ical research. The explanations offered by the micro­theories may or may not
be true, but they are complete in the sense of showing how certain subtypes
of non­monotonic change can be understood as the emergent result of the
interactions among processes that are not themselves non­monotonic change
processes.
This observation suggests a particular type of evolutionary explanation: the
emergence of new functionality as a side effect of changes in prior structures.
This type of explanation is familiar in evolutionary theory, in part due to some
long­standing challenges to that theory: What good is half a wing or one­third
of an eye? natural selection is blind to purpose and can only “select” among
phenotypes that already exist; it cannot select “for” a function that will be use­
ful to future generations, once that function has grown to sufficient power or
versatility. so what drives change during the initial phase of the emergence of
25
a novel adaptation? A common theoretical move for such cases is to assume
distinct evolutionary phases in which the emerging adaptation fulfills different
functions and hence is subject to different selective pressures. To return once
again to the evolution of flight, half a wing might be quite useful as an enlarger
of apparent body size, jump stabilizer or thermoregulator. selective pressure
might increase the size of the proto­wing to better fulfill those functions. once
those pressures have provided two­thirds of a wing, the possibilities of, for
example, long jumps and glides create a new set of selective pressures that
shape the emerging wing to better support flight.
A similar two­stage scenario might have played out with respect to the
human capability for cognitive change. The unique evolutionary pathway of
human beings – the reliance on learned instead of innate behavior patterns to
get through the day – created selective pressures in the direction of increased
cognitive ability. if you are going to live by your wits instead of your claws, those
wits had better be good. As a result, the different cognitive functions – percep­
tion, memory, problem solving, planning and plan execution – increased in
complexity and power. As a side effect of this increase in processing power, the
mechanisms for routine processing and monotonic learning came to exhibit
all the properties specified by the principles in Table 11.2. At that point, non­
monotonic change became a possibility. The survival advantages associated with
being able to override the imperatives of prior experience in turbulent environ­
ments brought new selective pressures to bear that pushed the cognitive system
in a new direction. The resulting capability enabled Homo sapiens to survive
even in environments that changed too quickly and too chaotically for species
that rely on innate skills or the monotonic projection of prior experience.