The next wave in computing is something that was predicted fifty years ago. Marvin Minsky, Herbert Simon, and other luminaries in Artificial Intelligence predicted that machines that equaled or improved upon people would soon be upon us. But, as Minsky recently pointed out, this prediction entailed the assumption that lots of people would be working on making machines intelligent and this simply was never the case. Now it is time for AI to move out of the realm of definable games and into the realm of real human needs in well-defined domains. People who know the domain, combined with computer scientists and cognitive scientists who can model that domain, allows for the possibility of really intelligent software in that domain.

What should that software seek to accomplish? Should it try to be the smartest entity operating in that domain? This seems like a foolish goal.

A more important goal is to create software that would make the people who work in that domain more efficient, more knowledgeable, and more unlikely to make critical mistakes. To accomplish this, the computer would have to know a great deal about a domain and a great deal about the problems in that domain. This constitutes a plan of attack.

Consider the problem of diagnosis. Who would want generic diagnosis software? Such software would tell you how to approach a problem in general, and how to consider evidence. But without real knowledge of what you were trying to diagnose it wouldn’t be that useful. It might be good for a first year medical student, but you would want a cancer specialist to diagnose cancer on the basis of having seen a lot of it.

To put this another way, an intelligent diagnosis tool would only be useful if it knew an awful lot about the domain in which it was operating. Like the chess program, it would have to know how to play the game and know what has worked or failed in the past.

Enterprises today are in need of channeling the right information to the right person at the right time. This would enhance productivity and more effectively manage risk. Enterprises today use e-mail for co-ordination instead of thinking about how to effctively organize information.

Software needs to model the key knowledge structures of the enterprise in order to deliver information just in time, while someone is trying to do something, so that it facilitates that doing. So, if diagnosis of a problem is the issue at hand, the question is, can the software find relevant knowledge and quickly deliver it to the decision maker(s) to help make the proper diagnosis? In other words, can the software replicate a sage old expert and have him be available in simulation every time you need him?

Men working on a shipping vessel where intelligent enterprise software is now being used. Doing this depends on both having the right knowledge in the software in the first place and properly indexing that knowledge so that it is called up just in time. Indexing knowledge properly, the basis of what intelligent systems are all about, means knowing the goals, expectations, plans, mistakes, conflicts and so on that categorize useful information.

These subconscious categorizations of what we know are the basis of human intelligence. The basis of machine intelligence is the same. While we can’t collect and properly index all of human knowledge just yet, because it is a daunting task, we can select well-defined domains of knowledge that we think are worth the effort.