An `AI in Design' View of Design


Does the system do any learning, and, if so, what is learned?

Discussion. Learning extends the capabilities of a design system without the direct involvement of the user [Duffy 1997]. It is extremely useful to have the system learn to deal with new situations, whether these lead to success or failure. New plans offer different ways to construct designs. New designs imply that the system has expanded its space of solutions. New attributes enhance the existing descriptions of the objects used by the system. New constraints are associated with a better focus in the search for a solution.

What type of learning method is used?

Discussion. The type of learning determines the method through which the new knowledge is created [Lenat 1978] [Lenat 1982]. Compilation assembles existing pieces of knowledge into new entities, relating causes to effects in a more direct way [Araya & Mittal 1987] [Brown 1991] [Chabot & Brown 1991] [Keller et al. 1992] [Spillane & Brown 1992] [Tong & Franklin 1989]. Analogy describes the transfer of a problem-solving method from a source area into a target domain [Huhns & Acosta 1992]. Explanation-based learning uses the generalized explanation of a problem-solving method to cover a larger application domain.

What are the sources of learning?

Discussion. The sources of learning determine what will be learned and what triggers the learning process. Failures are a very good source of learning, since it is desirable to avoid them in the future. Histories, as repositories of system experience, offer an opportunity to process knowledge for more efficient use [Mitchell et al. 1985] [Mostow et al. 1992]. In distributed design systems, conflicts should be analyzed to determine possibilities to alleviate them, to solve them or to avoid them altogether [Grecu & Brown 1996].

Up : Next

/Papers/IJDC/sect5.21.html Wed May 28 14:46:09 EDT 1997