Engineers Don't Search

This paper is on the automation of knowledge-intensive tasks in engineering domains; here, the term “task” relates to analysis and synthesis tasks, such as diagnosis and design problems. In the field of Artificial Intelligence there is a long tradition in automated problem solving of knowledge-intensive tasks, and, especially in the early stages, the search paradigm dictated many approaches. Later, in the modern period, the hopelessness in view of intractable search spaces along with a better problem understanding led to the development of more adequate problem solving techniques. However, search still constitutes an indispensable part in computer-based diagnosis and design problem solving—albeit human problem solvers often gets by without: “Engineers don ́t search” is my hardly ever exaggerated observation from various relevant projects, and I tried to learn lessons from this observation. This paper presents two case studies. 1. Diagnosis problem solving by model compilation. It follows the motto: “Spend search in model construction rather than in model processing.” 2. Design problem solving by functional abstraction. It follows the motto: “Construct a poor solution with little search, which then must be repaired.” On second sight it becomes apparent that the success of both mottos is a consequence of untwining logic-oriented reasoning (in the form of search and deduction) and approximation-oriented reasoning (in the form of simulation).