The inherent problem with algorithms whether they are based on neural networks,
genetic algorithms, knowledge discovery systems, or any other type of pattern
recognition software is the paradox in automated pattern discovery systems in
extracting random data sets and the association of interestingness as a measure
of the usefulness of the data patterns revealed because there is no prior
knowledge of the likely interesting data associations to be found before
automation. “If you do not expect it, you will not find the unexpected,
for it is hard to find and difficult.” (Padmanabhan & Tuzhilin, 1999)
So, you see the dilemma for artificial intelligence is the probability, that a rule exists, that maps one
association to another with the subjective attributes of unexpectedness is
present in the fact that the unexpected is expected in the probability of
chance relationships existing in any given time frame.
Many algorithms exist in determining patterns of interestingness, these then
can be used as a measure of the expected quality of the data retrieval, the
the main problem with this approach would seem the probability of randomness in any
given set of data. The random nature of data aggregation assumes that while
some methods would filter interesting patterns, it would also be conceivable
that many more one-dimensional patterns would not be recovered without
techniques that would isolate odd patterns.
Clearly, only understandable patterns can qualify as new
knowledge, hence, the importance of interestingness measures in finding and
tuning search heuristics in this quest for artificial intelligence.
