Path: utzoo!news-server.csri.toronto.edu!cs.utexas.edu!samsung!umich!sharkey!msuinfo!galaxy.cps.msu.edu!dailey From: dailey@galaxy.cps.msu.edu (Chris Dailey) Newsgroups: comp.ai.philosophy Subject: Re: computer life? Keywords: Survival, instincts Message-ID: <1991Mar3.024846.16626@msuinfo.cl.msu.edu> Date: 3 Mar 91 02:48:46 GMT References: <1991Feb28.235517.20218@news.larc.nasa.gov> <1991Mar1.143222.29977@msuinfo.cl.msu.edu> <1991Mar1.205136.10670@news.larc.nasa.gov> Sender: news@msuinfo.cl.msu.edu Reply-To: dailey@galaxy.cps.msu.edu (Chris Dailey) Organization: Dept. of Computer Science, Michigan State University Lines: 102 Originator: dailey@galaxy.cps.msu.edu [sorry for the long message] In article <1991Mar1.205136.10670@news.larc.nasa.gov> kludge@grissom.larc.nasa.gov ( Scott Dorsey) writes: >In article <1991Mar1.143222.29977@msuinfo.cl.msu.edu> dailey@buster.cps.msu.edu (Chris Dailey) writes: >>I would tend to agree that we could not do software mutation and weeding >>out of bad variants with what are today's acceptable software >>engineering practices. Maybe this will become a new specialization? I >>think it is needed to accomplish 'AI' (I'll leave the reason >>unexplained). >No comment. Except to say that if we do mutation and weeding, than we need >to develop an adequate mutation system and also a criterion for weeding out. >Evolution (in the sense of biological systems) is very, very slow. It took >quite a few million years to construct man, even though it was operating in >a parallel fashion (across many different people in large populations). But man had to restart from the beginning with each new birth. With each death, infinite discoveries are lost. This would not need be the case with computers, as death would not necessarily occur, and they would not need to evolve a library system to remember important discoveries as humans did -- it will already have that capability. >I >think that this is not a very efficient way of developing anything, and it >also tends to result in the development of systems which are not well- >understood. (Maybe that's the crux of the thing.) Under controlled circumstances, I think it would be most advantageous for a computer program to be able to evaluate its performance for various processes. If some process is costly, it could then try to create a new method to achieve the same outcomes. If it comes up with a more efficient method, it should be able to replace the old with the new -- it could keep the old method around if it [or the original programmer] wanted. This would be the program doing its own mutation and weeding. [Current software engineering practices dictate that the programmer does all mutation and weeding -- a practice staunchly supported by most software engineers. :) ] Agreed, there would need to be much work done on the performance evaluation functions and verifying that new methods are equivalent to old ones. >The point is that humans don't have brains which are suited to playing chess >like computers. [...] Nor computers like humans, etc., etc. Come to think of it I don't see how this has much to do with whether something is intelligent or not [except that chess is a specialized case where a computer is not really being intelligent, but being a computer following it's programmer's commands]. [..my reference to cellular automata and chaos deleted..] >Yes. But since the system is calculable, it's behaviour can be predicted. >It's just more difficult to predict, and the pattern which occurs is not >obvious at all to the naked eye. (Sorry if I tend to have a rather >deterministic, nineenth century point of view on the world, but I think >it's a rather good one). If it's almost impossible to predict, that just >means that the tools for predicting it haven't been devised yet. Someone mentioned a recent issue of Popular Science (I'm sorry, I just saw it at a friend's place and don't have more info on the reference) in which there was an article about MIT's insect labs. In going from point A to point B, the robot 'insect' would take different paths every time. This is the type of unpredictability I'm talking about. You might be able to get the 'insect' to take the exact same path every time, but you would have to have the 'insect' start out in exactly the same place every time -- an impossibility. (Plus the terrain that the 'insect' went over would be [minutely] different by the time he went over it again due to the fact that it went over that terrain!) I strongly believe that you can model pretty much anything in the real world with mathematics. The more accurate you want to be, the more mathematics you need to do so (and the more accurate results that you can get). The real world has a nearly infinite [you know what I mean] number of particles, so in order to simulate the universe, how much computer would you need to keep track of every single particle? There's be a catch-22 because the computer itself is a part of the universe it was trying to keep track of. But I have digressed. >>>This sounds very much like we are agreeing. >>I think you're right. >The whole point of the first article that I wrote was to point out that >because humans and computers have fundamental differences in the way they >work inside, that tasks are performed differently by computers and human >beings to obtain the same result, and therefore if a living computer system >is built that it probably won't be built in the same way that a living organic >system is built. I don't know at all how it will be built, and if I did I'd >be making a fortune on the lecture circuit. I guess we were just splitting hairs on that particular point. Although it would be based on human ideas, the architecture would be much different. Kinda like speaking a different language -- you can get an equivalent idea across (although not the exact same idea), but you use different words and a different grammar. >--scott -- Chris Dailey dailey@(frith.egr|cps).msu.edu __ __ ___ | "A line in the sand." -- The Detroit News __/ \/ \/ __:>- | \__/\__/\__/ | "Allein in der sand." -- me