lansky@VENICE.AI.SRI.COM (Amy Lansky) (10/30/87)
DEFAULTS AND CONJECTURES:
HYPOTHETICAL REASONING FOR EXPLANATION AND PREDICTION
David Poole (dlpoole%watdragon.waterloo.edu@relay.cs.net)
Logic Programming and Artificial Intelligence Group
University of Waterloo
11:00 AM, MONDAY, November 2
SRI International, Building E, Room EJ228
Classical logic has been criticised as a language for common sense
reasoning as it is monotonic. In this talk I wish to argue that the
problem is not with logic, but with how logic is used. An alternate
way to use logic is by using theory formation; logic tells us what a
theory implies, an inconsistency means that the theory cannot be true
of the world. I show how the simplest form of theory formation, namely
where the user supplies the possible hypotheses, can be used as a
basis for default reasoning and model-based diagnosis. This is the
basis of the "Theorist" system being built at the University of
Waterloo. I will discuss what we have learned from building and using
our system. I will also discuss distinctions which we have found to
be important in practice, such as between explaining observations and
making predictions; and between normality conditions (defaults) and
abnormality conditions (prototypes, conjectures, diseases). The
effects of these distinctions on recognition and prediction problems
will be presented along with algorithms, theorems and examples.