[IxDA Discuss] Success with inference engines?

[Bruce Esrig]

Yes, I once implemented the rule-firing part of a sophisticated and 
rigorous inference engine that was ultimately too heavy to fly.

On the front end was a heuristics based graphical interface that did its 
own reasoning. It showed a little network and could chase four values 
around the network and tell you where they could get to. It was great.

To make inference work, you need:
  - a really limited problem domain to reason about
  - a simple set of inference rules

It takes some experience with inference to know what "simple" is. It's 
surprising how fast this stuff can get out of control.

For a simple rule of thumb, watch out for "or". It's the choices that lead 
to the combinatorial explosions.

Best wishes,

Bruce Esrig