Tuesday, November 30, 2004

Solving Problems

One of the computational tasks that's intrigued me for quite some time is how you teach a computer system to solve a problem. For basic problems, the path is pretty obvious -- make up or determine an algorithm, code it, and then apply it to problems for which it's suited. As problems become more complex and textured, it becomes a question of finding out which algorithm is suitable, or which is more suitable. At some point, it starts to evolve into a system for the creation of algorithms, because you can't make one for every possible situation. At some point after that, the system starts to do interesting things, like claiming that the AE-35 unit is about to fail. And we all know where that leads.

Last night, as I was putting the new PC away, having finally gotten it to work, I thought about this a bit. In retrospect, I could have shaved hours off the resolution time if I'd realized that the migration software tool that I'd used had carried along certain registry settings, and then second, with the modem, where it somehow trashed (I don't know exactly where) the ability of the system to use the perfectly fine connection. The Dell people seemed to be working through a checklist, but not a particularly good one -- at one point, they said they'd replace the system, but I thought this seemed overkill, given the results of some experimentation I'd done, and eventually, they suggested that I just do a system restore. They recommended the Dell version, but I chose to do the Microsoft one, which says it won't mess with your program files (mostly true) . The Dell one says that it most certainly will. And it worked. Registry looks fine, modem connects with no problem.

It would have been nice to have a software solution that could help me to that solution, though. I tried thinking backwards -- okay, here's the solution, now how could I have best gotten here, and done so in a way that didn't rule out other, equally viable, options -- but I didn't get very far. I've used some of the Microsoft canned solution engines that are in the Help function, and they almost uniformly are too lame to consider as an actual 'engine'. They probably do work for people who need to be reminded that in order to work, the modem (if its an external one) must be plugged in. But past that level of abstraction, they tend to flame out. I get the sense of an infinite loop. Since my own patience is considerably less than infinite, I can't swear that if I'd Just Stuck With It, it wouldn't have eventually come up with the answer. But I don't think so. An experienced person usually knows when to try quick fixes first, and my sense it that the software wouldn't do that. It tries easy stuff.

Basically, I want a bright little AI engine popping up, advising me what might work best, next, and why. Something that learns over time, and can be downloaded and upgraded at will. (For a nice example of what an AI service can do, see the Twenty Questions site I mention in the post just below this. What's particularly interesting is that it shows how it got the answer. Some of its suppositions are totally hosed -- yet they work. Intruiging.) . I'd buy a service that helped me with this kind of question -- even a subscription service. (Now, there's an idea...hmmm....) .

Of course, if the tool were built by Microsoft, it would sneer at any software on the PC that was more than six months old, and simply recommend replacing all of it. The hardware, too.
.
---------------------
Side note: Last night, I mentioned to my wife that what I wanted was an answer to this problem, and I wanted it RFN. Whereupon my daughter, who had been reading, looked up and asked what RFN meant. Umm....its an acronym, like ASAP. Know that one?

2 comments:

Roger Stevens said...

Algy Rhythms? Lost me on this one.

BTW did you know that BPWBP?

Cerulean Bill said...

...and you lost me on THAT one.

Are you saying that the concept of algorithms has you lost? Or the AE-35 unit? Or something else entirely?