Information Technology Reference
In-Depth Information
Fig. 1.1
Power shovel
reflecting jointed limbs
I will take a strictly pragmatic stance in this chapter by asking the question “What
features must be present that make behavior intelligent?” I will demonstrate that the
often-quoted Turing Test, where machine intelligence is compared with a person,
can be shown to be insufficient to support any useful discussion. Even intelligence
measures such as IQ tests only suggest problem-solving specializations and little else.
There is an alternative view, backed by experimental evidence, of practical intel-
ligence by Jaques et al. (
1978
) which shows that intelligence in children develops
in stages and is not continuous as originally supposed. This calls into question the
validity of the IQ test, which assumes a smooth uninterrupted growth with age. How-
ever, Jaques' Discontinuity Theory does identify the notion of 'insight' in problem
solving. I will go on to show that information theory, developed by engineers to quan-
tify communication systems, can also provide a means of measuring the practical
consequence of 'insight' as well as providing an argument for the need of 'purpose'
in intelligent behavior.
Another powerful tool to help us understand human problem-solving is found in
the work of Charles Saunders Peirce (Peirce
1958
,
1966a
,
b
see Tursman
1987
), in
which he introduces the three types of inference: Induction, Deduction, and Abduc-
tion. These types support a range of specialization for different aspects of reasoning.
I will suggest that each of these aspects can be improved through experience, leading
to the notion of 'wisdom' and a practical measure of intelligence in both machines
and people. In Chap. 3 a simple kind of intelligence is constructed as a computer
program, illustrating that intelligent machines, as they are currently conceived, are
unlikely to function outside of their human context. The reason for this will be
explored in later chapters.
The problem with the pursuit of programming machines to behave intelligently
is that in practice, as soon as it is done the program no longer seems to be intelligent
because we know it is only a defined and understandable procedure. It usually is
assumed that any understandable procedure cannot truly reflect human intelligence.
This view of retaining the mystery of human thinking is partially explained by the
fact that the results of all attempts to construct machine intelligence have been so dis-
appointing that the 2013 British Computer Society Machine Intelligence competition
