Abstract :
Many people have experienced computers not doing what they were expected to do or, perhaps more often, met with computers that deliver faulty results. Frequently, if not always, such behavior is blamed the lack of a complete requirements specification. In this assertion, it is tacitly understood that if a specification is correct and comprehensive enough then it is possible to avoid computing errors since the designer in those cases is able to construct a correct program whose behavior in any situation is well known: every detail in the program is known and so is its behavior. In this paper I will show that if there is a truth in such a conjecture then it is valid only for not very complicated programs. In explaining this, I will follow an interesting observation of von Neumann that there is a good deal in (formal) logic which indicates that the description of the functions of a computer (automata) is simpler than a description of the computer itself, as long as the computer is not very complicated, but when we get to high complications, the actual object is simpler than the literary description. This is not a consequence of an emergent property, like an evolving organ, but is a restriction bound up with languages and not an intrinsic property originating from physical complications. Its consequences are that with very high complexity, we can never truly know how the computer in every situation will behave. As long as computing was mostly on site, perhaps this state of things was of minor importance but with the advent of an improved Web, the, so-called, Semantic Web, this indeterminism will make a difference. There will be an element of uncertainty and it may very well be a question of confidence.
