Sunday, May 08, 2005

HAL? You There?

I was thinking, in my normal disorganized way, about artificial intelligence. Disorganized is meant to imply one of those random thought processes that indicate an amazing depth and breadth of thought, one of those minds that cannot be constrained by the usual niceties of how people think. Of course, anyone who's read more than a couple of lines of what I write knows that disorganized simply means messy, with just a soupcon of 'then a miracle occurs'.

When I was looking into my network problem (and oh, by the way: now that I am using the other protocol for file and printer sharing, the PC boots faster. I guess it used to be trying to do something and failing, and now it just does it. Or doesn't even try.) I thought on more than one occasion that this ought to be something that a canned artificial intelligence program could handle for me. I was not thinking exactly of the AIs that appear in commercial science fiction novels (Roboticon, make some dinner for me, please, something light, and get my broker on the phone, too) but something closer to reality. The steps I took, the things that I looked at were for the most part limited to this PC on which I'm writing this screed, so it ought to be possible to write a program, or a series of programs, to do that analysis automatically. Further, it ought to be possible for such a program to know all of the possibilities that could result in the condition being analyzed. (Okay, already; if the network wasn't working because someone had dropped a bowling ball on the keyboard, I wouldn't have expected the AI to know that. ) But all of the likely candidates, yes, I think it would.

Now, bear in mind that I still do not know what it was that changed which causes my file and printer sharing to stop working. That means that there is something that I don't know (possibly more than one) that could and in this case did cause my problem. But I do know of several things that could have caused it, and did not. For example, it isn't because all of the computers involved weren't in the same workgroup. It isn't because more than one computer in the workgroup had the same name. It isn't because the GUEST account wasn't active. It isn't because the account in use wasn't a Computer Administrator. It isn't because a firewall was inhibiting communication. It isn't because a crossover cable was bad. It isn't because a network card was in the process of failing or had failed. Is isn't because the Domain Name Server was inaccessible. All of these possibilities, I had to tease out of various articles. Most of the articles agreed about details; when they did not, it seemed that they still agreed about the major things to check.

So what the Network Problem Analysis AI would do, then, is to run through these possibilities and check the state of all of these conditions. And, what the heck, download from Redmond Command the current set of possibles for the configuration in question (which means it asked me How Many Computers in the Workgroup? Which are hardwired into the router? Which use wireless cards?). And then it would say, in my ideal world, "TCP/IP File Sharing is inoperative for an unknown reason. IPX/SPX protocol is available but not installed. Recommend installation of IPX/SPX protocol."

Even I see that one of the problems with that scenario is that its very focused. The next guy who comes along might have a problem that is somewhere else entirely -- perhaps not one of those 'could be the network, could be your PC' deals -- so the AI has to figure out what steps would solve that guys problem. What it'll need is a generalized problem analysis routine, just as an expert would do it. And, being an expert myself in some areas, I know that frequently the expert -- at least, the field expert -- doesn't know all that there is to know, and many times just tries the most likely things first -- the ones that come up most often. So the AI would do that, trying the top ten picks from Redmond first, only later delving into more arcane analyses.

There's nothing inherently magical about AI. It COULD be done.

I wish it were. And I wish I could help it happen.

No comments: