ABSTRACT
When AI systems make inferences based on the facts they have at hand, the issue arises of what to do if one of the facts turns out not to be true after all. For example, if the database has the forward-chaining rule (- > (ISA ?X DOG) (ISA ?X MAMMAL)), then (ISA FIDO MAMMAL) is added as a side effect of adding (ISA FIDO DOG). If we now find out that (ISA FIDO DOG) is not true, we had better also remove (ISA FIDO MAMMAL), unless it can be proved by another means.
