On the Relation between Abduction and Inductive Learning

This Handbook volume is devoted to abduction and learning as they appear in various subfields of artificial intelligence. Broadly speaking, abduction aims at finding explanations for, or causes of, observed phenomena or facts. Learning occurs when an agent adapts its state or behaviour on the basis of experience. Both processes have a number of common characteristics. They both aim at improving their picture or model of the universe of discourse; they are both hypothetical, in the sense that the results may be wrong; and they can both be seen as reasoning processes, with the observations and the current knowledge of the world as input statements and the learned or abduced hypotheses as output statements.1 In the case of learning from examples, which is the most common form of learning studied in artificial intelligence, this form of reasoning is called induction, and that is the term we will be mostly using in this chapter. Given these common characteristics of abduction and induction, it makes sense to study them together. Once we have a clear picture of their similarities as well as their differences, understanding one contributes to the understanding of the other. Such an integrated study is the subject of this introductory chapter. As abduction and induction have been studied in philosophy and logic as well as artificial intelligence, we review selected approaches from each of these disciplines before attempting to come up with an integrated perspective. Even a casual investigation of the literature on abduction and induction will reveal that there is a significant amount of controversy and debate. For instance, Josephson writes that ‘it is possible to treat every good (...) inductive generalisation as an instance of abduction’ [Josephson, 1994, p.19], while Michalski has it that ‘inductive inference was defined as a process of generating descriptions that imply original facts in the context of background knowledge. Such a general definition includes inductive generalisation and abduction as special cases’ [Michalski, 1987, p.188]. One can argue that such incompatible viewpoints indicate that abduction and induction themselves are not well-defined. Once their definitions have been fixed, studying their relation becomes a technical rather than a conceptual matter. However, it is not self-evident why there should exist absolute, Platonic ideals of abduction and induction, waiting to be discovered and captured once and for all by an appropriate definition. As with most theoretical notions, it is more a matter