On the role of abduction
Pietro Torasso, Luca Console, Luigi Portinale, Daniele Theseider Dupré · ACM Computing Surveys · 1995
The notion of explanation is basilar in many human behaviors and indeed in many elds (such as philosophy or psychology) there is a long tradition in the study of such a notion. In particular, reasoning towards explanation is a basic task in many of the problem solving activities investigated by the AI community. For example, it is the core of diagnostic problem solving (whose goal is to explain symptoms observed in a system with the presence of some faults) or of interpretation activities (e.g., image or story interpretation or plan recognition), and it is a basic component of machine learning (where the goal is to produce some theory that accounts for and generalizes some phenomena). Following a logicist tradition, many AI researchers working on explanation reduced such a notion to deduction: a phenomenon is explained when it can be deduced from a domain theory, possibly after some assumptions. Indeed, the interesting case is the one where the observed phenomenon does not follow from the original theory and some assumptions have to be made � in particular, one may either assume the truth of some facts mentioned in the theory or extend