by Hugh M. Lewis

I want to make a case for the incorporation of randomizing functions in software systems that purport to be "intelligent" by design. I find far too little randomization of such systems whenever I encounter them, and I for one have always made an effort to incorporate such randomization into the very design of the kinds of "intelligent systems" is seek to construct. I would say that randomization depends upon certain degrees of freedom being available in the possible patterning of a system, and leads to flexibility in the patterning of systems that can exponentially increase to infinite proportions. I would suggest that in the naturally intelligent systems we are most familiar with, namely ourselves, our language, and our behavior. 

If we define randomization of information as the substitutability of one bit of information for an equivalent, alternative bit of information, from a range of such alternative bits of information, then at each step in a sequence of possible substitutions we dramatically increase the possible number of patterned outcome of such a sequence. We can apply this to the theory of automata, as this is represented by the Turing Machine, and say that for each unit of input in an automata machine, there is in principle an infinite range of possible outputs produced. The theory of automata ties directly to logical relations, the manipulation of sets and matrices, and set theory.

Of course, for natural systems, we cannot have completely random systems. Random systems in nature are completely chaotic, and hence, any totally randomized system would have no logical sense of order--it could not be a system as such, if a system is defined as some set of relationships between parts that have a sense of overall order. We say such systems are partly determined or constrained, rendering them to some extent non-random and therefore sensible and available to our descriptive and predictive analysis, even if only in a statistical sense for fairly complex systems. And all natural systems are in the final analysis complex.

But randomization, if controlled, can enhance and create the complexity of an artificial system resembling natural intelligence, and hence more closely approximating conditions of the Turing Test for artificial intelligence, namely the Chinese room experiment.

If we can redefine randomization from the standpoint of information theory as the equal probability or likelihood of occurrence at any give point of a set or paradigm of alternates, however large, then we can in our groupings and relational sets insert a sense of non-random order to our framework. If we can further restrict the alternates by certain principles, for instance as in linguistic coding, the choice between alternates is no longer completely random, but biased upon some principle of selection or prioritization. 

We must be careful here when referring to artificial or machine intelligence to distinguish between genuine intelligence as a form of intuitive feeling, understanding and self-awareness, and the mere reflective mimicry of such understanding that we ultimately project from ourselves into the behavior of such machines when the trick us into believing that they are naturally responsive to our behavior in an intelligent manner. Building such randomization of such systems into a basic level of their design and articulation may be a way of achieving a sense of mimicry of intelligence, and helping us to develop a working, functional model of such intelligence, but it does not substitute for what we could call deliberate, willful, intentional thought and sensibility.

General Systems Essays, Vol. I