ON THE RELAmONSHIl? BETWEEN STRONG AND WEAK PROBLEM SOLWRS

The basic thesis put forth in this article is that a problem solver is essentially an interpreter that carries out computations implicit in the problem formulation A good problem formulation gives rise to what is conventionally called a strong problem solver; poor formulations co*respond to weak problem solvers Knowledge-based systems are discussed in the context of this thesis We also make some observations about the relationship between search strategy and problem formulation DURING THE LAST DECADE the distinction between strong and weak problem solvers has been emphasized in the AI literature. Weak problem solvers are those that are relatively easy. Strong problem solvers, on the other hand, can solve relatively difficult problems but are specialized to a particular application domain. The usual explanation for the performance of strong problem solvers is that they can bring specialized knowledge from the application domain to bear on the problem. This distinction between problem solvers dates back to Newell (1969). The question addressed in this article is “what is the relationship between weak and strong problem solvers?” One possible answer is that there are two different theories of problem solving: one for weak problem solving, the other for strong problem solving. We do not subscribe to this answer. This research was partially supported by the United States Air Force, Some aspects of MYCIN don’t fit the problem reduction RADC Contract F30602-82K-0045 paradigm as naturally as the above, of course. For example, a However, if it is incorrect, there must be some relationship between the two that allows them to live harmoniously within a single theory. The nature of this relationship is the focus of this article. In passing we note that the theory of weak problem solvers has been well-developed for over a decade; see Kilsson (1971) for example. MYCIN as a Weak Problem Solver To start off the discussion, let us make a statement just to make a point: many expert systems can naturally be viewed as weak problem solvers As a concrete example, consider MYCIN (Shortliffe, 1976). Its state space is the set of atomic formulae of the form < predicate function > (< object >, < attribute >, < value >) Each MYCIN production can be viewed as an operator in the problem reduction paradigm (Nilsson, 1971). For example, the production “if A & B then C” corresponds to the operator whose input state is C and whose output is the AND of the two states A, B. The goal states are patient data such as the results of lab tests. THE AI MAGAZINE Summer 1983 25 AI Magazine Volume 4 Number 2 (1983) (© AAAI)