(I've had some interest in Artificial Intelligence for a couple of years and have spent a fair percentage of this last year's take-home pay on AI textbooks. I tell my friends that I have one of the few platinum B. Dalton's cards around. After years in the laboratories, AI is beginning to come out of the closet with a few - very few - rudimentary applications.)
IBM's head marketeer for AI products talked about IBM's present product line and what we could expect in the future. This guy appeared to know what he was talking about, and was refreshingly candid. He characterized AI marketing as being "an awful lot of hype and hot air", and noted that IBM was not making the extraordinary claims of many of the other firms marketing so-called AI products.
IBM's offering consists mostly of a Common Lisp implementation and an Expert-System shell. Lisp is nearly the oldest high-level language in the industry; only FORTRAN is older. It was invented by John McCarthy in the 1950s at MIT, and has become the language of choice among AI researchers throughout the world. (It turns out that programs and data are the same thing in LISP, and it is a relatively simple matter to write programs that write programs.)
Expert Systems are also called "knowledge based" systems, but both phrases are really inventions of AI Marketters - few systems perform at the level of a human expert. In the academic community they are usually called "rule based" or "production" systems. A rule-based system consists of two parts: a database full of rules, and an "inference engine" that executes rules. Here are some sample rules for a simple rule-based system:
Rule databases contain related rules in an IF-THEN format. The inference engine is responsible for evaluating all the rules in a database when you ask it questions, such as:
Databases for certain applications might easily run into thousand of rules. You can see that performance could become an issue when each rule in a database has to be examined for a match whenever a condition changes. Although Expert Systems technology has been around for a long time, it is relatively expensive to run, and only recently has processor power been sufficient for rule-based systems to become commercially feasible.
A program that contains an inference engine, a database manager, and a user interface is called an "expert system shell". IBM's is called Expert System Environment, or ESE. It runs under MVS or VM.
IBM has lots of other AI systems in the lab that require a LOT of hardware. "Artificial Intelligence" is a catchall term for many different areas: natural language understanding, vision systems, learning systems, robotics, automatic programming, and of course expert systems. (Have you seen the IBM commercial for speech recognition? "Please write Mrs. Wright right away" is very difficult to parse correctly, but IBM does it on a desktop - with a 50,000 word vocabulary and a 27 MIPS processor. It still has bugs, and is nowhere ready for production.) IBM said that vision systems will probably be ready for commercial use before speech systems due to military requirements.
Rule-based systems have saved money, but ther are difficult to develop and literally IMPOSSIBLE to prove correct. IBM has an expert system with 1600 rules that tests 3380s before they are shipped. When an error is detected, the expert systems runs diagnostics and determines which card or component is in error. It is right 98% of the time, and it has saved IBM somewhere around 8 million dollars. It was written by two factory floor technicians over a period of about seven months, and will eventually be deployed for use by Field Engineers. Other successful applications for rule-based systems have been:
IBM has over 125 internal projects doing AI! Their labs are headquartered in Cambridge (home of the MIT AI laboratory) and Palo Also (home of the Stanford AI laboratory). They will do a turnkey expert system for you if you have the bucks.
Expert in the near future an expert system for SNA network diagnosis.