Information Technology Reference
In-Depth Information
Chapter 12
Implementing Intelligence
Routine, in an intelligent man, is a sign of ambition.
W. H. Auden
,
(1958, 'The life of That-There Poet')
12.1
Features of Intelligence
I have specified in Chap. 11 the different concepts as defined by Zakaria (
1994
)inhis
thesis which characterize sequence types. However, this does not help in identifying
what concept should be tried when given an IQ test sequence. For this it is required
to identify features of sequences and associate them with the range of generating
concepts. The problem is that the features assigned can apply to more than one
concept. This issue can be readdressed as a form of pattern recognition in which
the pattern of features will identify the most likely concept to apply to a particular
sequence.
The problem of inferring a concept from a sequence of numbers is very similar
to identifying a hypothesis from experimental observations, as described in Chap. 6.
Here we found that the running confidence test is an important choice in a learning
mechanism in that it can limit the amount of computation. When the degree of
confidence in a particular hypothesis reaches a certain low level then the computation
on that hypotheses will stop. The remaining hypotheses are further examined until
there is a winner; then this hypothesis is put forward. This avoids having to perform
deduction on all potential hypotheses. In addition it is this mechanism that provides
part of the tension between the three inference mechanisms through an exchange of
information on how well the competing hypotheses fit their respective criterion.
The abstraction of features from the facts gives rise to the formation of a hypothesis
generated from one of the several predefined concepts. The abstracted features are
a computationally simple test made on the data that will give some indication of the
underlying series generator. A simple Bayesian decision-making system is then used
to select the concept to be deployed in the hypotheses generator (see Chap. 6). The
selection is either made under a simple pre-learning stage or learning as combined
in a running window system. The learning that leads to a final choice of a concept
depends not only upon the features that go towards the generation of the hypotheses
but also upon a different set of features that show the general characteristics of, in
