Robotics Reference
In-Depth Information
the Nile delta. Please find a cure for those already affected by it and a
vaccination to guard against it.”
Knowledge Discovery
The sum total of recorded data and human knowledge has become ab-
solutely staggering in the Internet age. The School of Information Man-
agement and Systems at the University of California at Berkeley pub-
lish an annual report entitled
How Much Information?
, summarizing the
amount of information produced since mankind began, and in particular
during the year prior to that year's report. The total amount of informa-
tion that existed on printed, film, magnetic or optical storage media in
2002 is estimated to be roughly 18,000,000,000,000,000,000 bytes of
data (18 exabytes), of which more than one-quarter was created in 2002
alone.
With this ever-growing electronic mine of knowledge, available in a
digital form that can be processed electronically, come opportunities to
employ computers to unearth useful knowledge that we might not oth-
erwise have noticed. Unsurprisingly, this discipline, which is often called
Data Mining or Knowledge Discovery in Databases, is burgeoning.
The development of the necessary technology for Data Mining started
in the field of cognitive psychology in the mid-late 1950s at Yale, with the
work of Carl Hovland and his research student Earl Hunt, on the com-
puter simulation of concept creation. In 1966 Hunt, together with two
other cognitive psychology researchers, Janet Marin and Philip Stone,
published a model of concept formation called Concept Learning System
(CLS). This model was based the idea that, in order to isolate discrete
concepts from cluttered data, the human mind tends to divide complex
concepts into groups of smaller similar concepts that have common char-
acteristics.
Within this field of AI research a concept is defined formally as a
classification rule that divides a set of examples into two classes, such as
those examples that satisfy the concept, and those that do not. As an
illustration, if the CLS is set to learn how to play the Chess endgame of
king and rook versus a lone king, as played by Torres y Quevedo's rule-
based machine,
44
the system might begin by dividing all such positions
into those in which the lone king is within two horizontal or vertical
44
See the section “Torres y Quevedo's Chess Endgame Machine” in Chapter 1.
