Page 115 - Deep Learning

98 Creativity

Discussion
The key principles of combinatorial, layered processing, mutual constraints
within each layer due to horizontal excitatory and inhibitory links and context
effects implemented via downward feedback links are shared between visual
perception and discourse comprehension, and presumably by perceptual pro-
cessing generally. The application of these basic principles to problem solving
is straightforward. Whether in the psychologist’s laboratory, in school or in
work situations, people are confronted with situations and problem materials
that require visual inspection and interpretation.
For a situation or a set of materials to constitute a problem, it must be
paired with a goal, a mental representation of a desired (but, by definition, not
yet realized) state of affairs. If the goal is posed and communicated by someone
else, the constructive nature of comprehension has similar consequences as
the constructive nature of vision: Goal comprehension is not uniquely deter-
mined by the verbal input. The problem solver’s representation of that goal is
an interpretation based on prior experience.
The representations of the initial situation and the goal constitute the
starting points for analytical problem solving. One might object that this per-
spective applies only to material problems, such as a flat tire, or to problems
posed by someone else, such as an employer or a teacher. But many impor-
tant problems are found and posed by the problem solver himself. Also, many
problems do not seem to refer to any concrete situation. Problems in mathe-
matics and other formal domains are examples. However, it remains true that
to have a problem is to entertain some initial representation of some state of
affairs, even if it is found rather than imposed and even if it is abstract and
imagined instead of concrete and situational. Also, to be engaged in problem
solving, a person must have posed some goal. It might be abstract, vague or
incomplete, but it has to contain enough information so that the person can
tell when he or she has solved the problem.

Knowledge Retrieval
A person’s knowledge store is vast. At any one moment, only a small num-
26
ber of all the knowledge elements in long-term memory are active. When a
person is faced with a problem, particularly an unfamiliar one, every piece of
his prior knowledge is potentially relevant. To be applied to the situation at
27
hand, a knowledge element has to be retrieved through spread of activation.
To visualize this process, it is useful to conceptualize long-term memory as a
network in which knowledge elements are nodes and the relations between