On Generality and Problem Solving: A Case Study Using the DENDRAL Program

In discussing the capability of a problem-solving system, one should distinguish between generality and expertness. Generality is being questioned when we ask: how broad a universe of problems is the problem solver prepared to work on? Expertness is being questioned when we ask: how good are the answers and were they arrived at with reasonable cost? Generality has great utility in some ways, but is not often associated with superior performance. The experts usually are specialists. In analytical chemistry, there is a domain of inductive inference problems involving the determination of molecular structure by analysis of certain physical spectra of the molecule. We have written a problem-solving program (Heuristic DENDRAL) that is prepared to attempt to solve any problem in this very large domain. By now, it has solved hundreds of structure-determination problems in many different chemical families. For some families of molecules, it is an expert, even when compared with the best human performance. For the other families, i.e., most of chemistry, it performs as a novice, or worse. This paper will use the design of Heuristic DENDRAL and its performance on many different problems it has solved as raw material for a discussion of the following topics : 1. the design for generality; 2. the performance problems attendant upon too much generality; 3. the coupling of expertise to the general problem-solving processes; 4. the symbiotic relationship between generality and expertness, and the implications of this symbiosis for the study and design of problem-solving systems. We conclude the paper with a view of the design for a general problem-solver that is a variant of the 'big switch' theory of generality. 165