When to filter what, or prediction isn't explanation
Michael Reinfrank · 2002
Discusses the minimal entailment relation proposed by E. Sandewall (1989). It is argued that Sandewall's notion of entailment is well suited for prediction but is incapable of accounting for evidence contradicting the predicted 'normal' course of events. The author does not describe a complete solution to this problem. However it is suggested that a solution has to be based on distinguishing between different purposes of the reasoning process, such as prediction or explanation. A problem solver must then make the assumptions underlying a particular mode of reasoning explicit in order to be able to reason about them and to retract them in the light of conflicting evidence. Sandewall's combination of logic and differential equations and its relevance to the area of intelligent control are described. The minimal entailment relation is introduced, and a problem with the logic when it comes to multiple modes of reasoning is presented. A solution based on mode assumptions is suggested.>