Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!zaphod.mps.ohio-state.edu!mips!dimacs.rutgers.edu!aramis.rutgers.edu!athos.rutgers.edu!christian From: sc1u+@andrew.cmu.edu (Stephen Chan) Newsgroups: soc.religion.christian Subject: Re: AI Message-ID: Date: 14 Nov 90 08:59:37 GMT Sender: hedrick@athos.rutgers.edu Lines: 103 Approved: christian@aramis.rutgers.edu > >Excerpts from netnews.soc.religion.christian: 13-Nov-90 AI > starpath@athena.mit.edu (4352) > > >Now that is a pretty strong statement. In fact, it is pretty mistaken. > >I really am not familiar with Rutgers CS program, but I am somewhat > familiar > >with MIT's CS program, it includes/has included Marvin Minsky and > Seymour > >Papert, two rather big names in AI. (No, I'm not CS, I'm EE, so I'll > >have to paraphrase.) Actually, from what I've heard, the Rutgers Undergrad CS program is pretty darn good: there are lots of computer oriented companies in the N. Jersey area which rate the Rutgers people highly. Don't be fooled by the big names "Minsky" and "Papert". The are definitely partisans in the AI field - they are brilliant men, but they don't have a monopoly on truth. My understanding is that the Minsky folks believe human intelligence is fundamentally a symbol processing system (I may be wrong). There are lots of folks who disagree. Besides, Minsky and (I think) Papert put out a paper in the late 60's or early 70's which "proved" that the whole class of connectionist networks were incapable of modelling human intelligence, because they couldn't even handle an XOR operation. As a result of that "proof", funding got diverted to the symbol processing field of AI. Well, about 5 years ago, derivatives of the early connectionist networks started tearing up the landscape in field related to pattern recognition, fields where symbol processing AI were just not cutting it. In fact, the connectionist networks were capable of fairly accurate "generalization"; they created their own internal representation for input and could identify other members of a category without having seen it before. Minsky and Papert called it wrong, and being a Big Name doesn't mean you are always right. > >Excerpts from netnews.soc.religion.christian: 13-Nov-90 AI > starpath@athena.mit.edu (4352) > > >> I suspect that the decisions these computers make will be EXACTLY > the same > >>decisions that humans in the same situation would make (only faster). > Thus, > >>if you want to enforce morality, you have to ensure that the laws are > moral. > > >Cute, but not true for real AI. This would be true for an expert > system, > >one which takes data only from it's creator. An AI computer is not so > >intrinsically linked to it's creator. Well, you are assuming that *real* AI is possible. The other fellow is assuming that real AI is impossible. No use arguing about conclusions when you start out with different givens. > >Excerpts from netnews.soc.religion.christian: 13-Nov-90 AI > starpath@athena.mit.edu (4352) > > > >But the computers you mention here are not making 'decisions.' The > answers > >the spew are not powerful enough to call them that; it is merely the > >end result of an algorithm. A 'judgement' is weighing the issues and > >making a decision based on the issues. The judgement is made when the > >code was written, not when the answer is processed. An AI machine would > >really make a judgement. I was taught decision making under the Rational Actor model. Basically, I make a list of different alternative, then the possible outcomes associated with each alternative. I then assign probabilities to each outcome, and a payoff associated with each outcome. I multiply the payoff by the probability, this is the Expected Utility. I then choose the alternative associated with the outcome assigned the highest Expected Utility. That is an algorithm, and I learned it from someone else. Does this mean that any decision made by that system is non-intelligent? You're getting a top rate education at MIT, you go there to have people teach you how to be an EE. If you use what they teach, then it does not detract from your intelligence. Sure, computers are programmed to do things a certain way, but weren't we all "programmed" in how to solve algebra problems, how to treat other people, etc... ? There's an aspect of free will and intentionality which needs to be considered. If an AI doesn't have these, then you can't talk about making the AI moral, you can only talk about making the AI _safe_. By the way, folks have been trying to use reason and problem solving skills to develop morality for a long time. In the end, it seems they merely provide systems to justify our own moral beliefs, but pure reason cannot come up with morality by itself. - Stephen Chan User Consultant Distributed Workstation Services Carnegie Mellon University EMAIL: sc1u@andrew.cmu.edu PHONE: (412)268-6267 [Again, let me say that no participant in this discussion is from Rutgers except me, and I'm just the moderator. (I happen to have a Ph.D. in Artificial Intelligence from CMU, but I haven't been actively involved in AI for a number of years.) --clh]