Information Technology Reference
In-Depth Information
IQ tests are used to measure human intelligence rather than machine intelligence.
This is of paramount importance since rather than constructing a machine with intel-
ligence capability, I am attempting to describe a model that was devised by Zakaria
(
1994
) of the human decision-making process within a problem IQ domain. Since
no predefined framework for this decision-making process existed in this domain at
the time the model was constructed, its form was based upon IQ tests. The comfort-
ing component about this very practical approach is that these tests reflect human
intelligence derived from empirical evidence. A definition of intelligence may be
disagreed with but performance cannot be denied.
For this purpose, a subset of intelligence tests was selected, to measure the success
of this model. The range of tests seeks to assess intelligence in a direct and reliable
manner. These tests contain a number of miscellaneous problems, since it is only in
this way that the test can yield a measure of that general ability (the g factor) which
is intelligence. However, these problems aimed for the fundamental objective: the
demand for relational and constructive thinking which involves the discovery of
relationships and the induction of correlates. From the set of 18 different tests (each
with between 35 and 40 problems), only those that deal only with numerical sequence
extrapolation were selected, since they served the purpose and objective of testing
the model proposed.
11.2
Sequence Extrapolation as a Model
Given a sequence of numbers, the system is organized to find a simple rule from
which such a sequence might have been generated. The creation and validation of
hypotheses are performed by interacting and co-operating retroductive, deductive
and inductive inferences.
Note that retroduction is closely related to 'abduction' as described in Chap. 2
(see Fig. 2.9). Retroduction selects a concept drawn from a set of concepts and
constructs (abduces) a specific hypothesis to 'explain' the observed facts. In this
case the observed facts are a sequence of numbers. Abstraction is not needed here
since a computer easily recognizes numbers and the basic transform functions for
numbers.
In its simplest form, the process is similar to the simple 'generate and test' proce-
dure (see Chap. 2, Fig. 2.9). However, the process is more complex than this simple
cycle in that the results at each stage influence the way in which each element in the
cycle behaves. There is a “tension” among the three inferences and this “tension”
provides feedback data from one inference to another in order to improve the quality
and credibility of a potential hypothesis. Figure
11.1
illustrates this tension.
Note that Fig. 2.9 in Chap. 2 and Fig. 4.4 in Chap. 4 there is an extra inference
of 'abstraction'. Abstraction infers the relevant features that form the appropriate
hypotheses. In Chaps. 2 and 4, I considered 'abstraction' as a fourth inference.
However, I will not consider it here because, first, this is not usual in the literature
and second, I am dealing with only numbers and the sequences that use the notion of
