EMYCIN CF vs. Probability Theory

EMYCIN certainty factor calculations differ in significant ways from standard probability theory. These differences have attracted criticism, but often have good practical motivations.

CF Threshold: EMYCIN generally considers anything with CF < .2 to be false.

Q: ``Can't this prevent several pieces of weak support from adding up to a significant level of support?''

A: Yes, it is possible, but doesn't seem to be a problem in practice.

The benefit of the CF threshold is that it keeps the system from asking a lot of dumb questions:

If: A (.01) 
and B (.02) 
and C (.001) 
and D (as yet unknown) ...

In such a case, we don't want the system to ask questions about D, which is a most unlikely prospect. Asking dumb questions will quickly discourage potential users of the system.

Contents    Page-10    Prev    Next    Page+10    Index