Greg Detre
Sunday, 24 June, 2001
in The philosophy of AI, (ed) Boden
He discusses AI as
the study of complex
information-processing problems that often have their roots in some aspect of
biological information-processing. The goal of the subject is to identify
interesting and solvable information-processing problems and solve them.
He divides the solution to an information-processing problem into two parts:
the �theory� of a computation � an abstract formulation of what is being computed and why
particular algorithms for implementing a computation, the how
He divides AI problems into two types:
type-1 problems can and have been specified in terms of (distilled into) an underlying, pure computational theory, that tells us what the algorithm is achieving, e.g. Horn�s (1975) method for obtaining shape from shading
type-2 problems can only be solved by the simultaneous action of a considerable number of processes, whose interaction is its own simplest description, where all you can do is map out and run this complex mess (e.g. protein folding)
We want to find type-1 solutions to problems, but that�s hard. Just because we haven�t found a type-1 solution doesn�t mean it doesn�t exist, and there may be a type-1 solution which is masked by a type-2 implementation. We should also be careful not to over-simplify, thinking that a problem is type-1 when it may actually require a type-2 solution to fully capture it (e.g. perhaps language). We shouldn�t be pleased with hack solutions that seem to mimic results in a type-2 fashion without making any effort to get at what the underlying processes must be (e.g. Weizenbaum�s �Eliza�)
He considers the following emerging ideas: