Information Technology Reference
In-Depth Information
Table 12.4
Used reason types and their formal limitations
Intelligence function
Characteristic
Abstraction
Closed, predefined features, two levels
Retroduction
Closed, predefined concepts, infinite and uncountable
Deduction
Closed, predefined heuristic
Induction
Closed, predefined single criteria
By employing learning and training to create a ranking of hypotheses (a ranking
list), we can see a high percentage of successfully formed hypotheses that matched
the intended hypotheses. This goes to show that learning and training can improve
the intelligence by providing better or more appropriate hypotheses. Learning does
not increase the ability to solve harder problems and hence the IQ value of the system
remains the same.
In both cases, the system managed to find hypotheses for 70 of the samples. The
other 15 samples cannot be fitted into any of the concepts employed.
12.5
Conclusion
What kind of intelligence are we implementing in our sequence recognition? We cat-
egorized that intelligence in the manner described in Sect. 2 as follows (Table
12.4
):
12.5.1
Intelligence Quality Specification
Since all intelligence functions in this model are closed and predefined, the potential
for a sensible response to a generation of a new domain and new concept is zero. This
is an example of a 'closed' inferential system. There is no new insights (concepts)
created here. What actually occurs is that hypotheses are built as a function of a
quantitative concept selected from a list of predefined concepts. Once a concept is
selected, its generator will derive all of its parameters. Induction will then evaluate
the viability of that hypothesis against a predefined criteria.
It is important to note that the employment of multiple concepts only improves
the range of intelligence; it does not improve the quality of intelligence. The more
concepts being employed, the more choices we can have to build our hypotheses. The
choice of concepts to build a hypothesis is based on Bayesian statistics. No doubt there
are other techniques that may be used for the classification of concepts. However,
the best technique should be based on sound statistical techniques. Bayesian is well
known for being the most comprehensive and widely used technique. The attraction
of Bayesian technique can be seen from its simplistic rule:
