Managing Knowledge with Artificial Intelligence by Kevin C. Desouza
Posted on January 20th, 2004
A few weeks back I finished reading Kevin C. Desouza's Managing Knowledge with Artificial Intelligence . The subtitle says it all-- this is "An Introduction with Guidelines for Nonspecialists." Nonspecialist, that's me. Interested in knowledge management and AI, that's me.
Each chapter of the book covers a different technique for managing information. Managing, that is, so you or your business can either recall it at some point in the future, exploit it in the here-and-now, or basically just do something more sophisticated than sitting on it.
The structure of each chapter is fairly formulaic-- you start of with a what-is-this-thing explanation, then saunter into an overview of where the technique is used, the kinds of problems it helps solve, and why.
A lot of techniques are covered, including all the big-name, buzzword-worth ones. Neural networks, got it. Fuzzy logic, got it. Case-based reasoning, got it. I doubt anyone without an interest in AI will be particularly fascinated with these topics, because they're far-afield from ready-to-go answers. They're possible solutions waiting for problems.
Still, if you can sit through the whole thing or at least skip through the tedious sections, you'll probably end up with a new awareness of all the knowledge management systems around you, and possibly some insight into why they might suck. Take the Windows Help system, for instance. In my experience it's not helpful at all. It seems comprehensive enough, but it never seems smart enough to bring me solutions to my problems. Granted, my problems might be more intricate than someone who needs to find the firewall settings in XP. For those kinds of predicaments, the Help system excels.
After you've used the Help system a few times and found it to be unhelpful, though, you may not keep coming back for reiteration. Past experience has built a model in my mind of what information the Help system has. Whether or not the model is true is of secondary importance. If I'm sure about this model, odds are I'll rarely be surprised by the help I get-- I have advance expectations. In other words, I may be stuck with a problem but I'm still the smartest thing in the picture.
Time travel a few decades forward, though. And leave behind all your preconceived ideas about what a computer can and can't do. The help systems of 2050 might be far closer to living up to their name. They might use a smorgasbord of all the AI techniques in Desouza's nook, plus a few new ones, to give machine something closer to human intelligence. That could mean better memory, as in a reminder of how you were asking about a similar problem a few months ago, and here was the solution you found then. It could mean better inference, as in a search query for "startup" bringing back information on things most likely to go wrong with your machine's startup, respective of its age and hardware, and general history. Or it could mean self healing-- a system where the computer is able to correct problems on its own, without you having to worry about it.
The appeal of all this to me is that's got potential to make substantial improvements in day-to-day life. Not many other things related to computer science-- at least that I can think of-- can stack up quite as well in that department.