Two speakers talked about rule-based applications they have put together with ESE, the "Expert Systems Environment" from IBM.

The first guy did a tiny - almost trivial - rule based application and was rewarded with a 40% reduction in response time for some of their jobs. They are heavy users of DCF, a batch text formatter similar to SCRIPT. When a user submits a DCF job, it is queued to a VM machine that runs DCF. During periods of low online activity, they can afford to run multiple DCF machines, but when many terminals are active they can't run more than one or else terminal response time goes down the tubes.

So they wrote a small, 15-rule "expert system" to decide when to start up more DCF machines, or to shut them down based on the queue length and system load at the time. They improved DCF response time by 40%. I suspect that they could have done just about as well in some other language, but it was a good experience for them.

The second speaker's application was not nearly as trivial. He is from an artificial intelligence group within Amoco, and his job is to do expert systems. He described the politics, design, implementation and verification of a rule-based system that describes an oil field based on peripheral data. Politics gets involved because you have to spend an inordinate amount of time with a human expert in order to translate his knowledge into rules for the database. His system ended up with several hundred rules, and has been placed into general use. Users seeking routine advice from the expert now use the expert system, and telephone calls to the expert have been reduced by 70%.