Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!dimacs.rutgers.edu!aramis.rutgers.edu!athos.rutgers.edu!nanotech From: dmocsny@minerva.che.uc.edu (Daniel Mocsny) Newsgroups: sci.nanotech Subject: Re: Some problems of super-intelligence Message-ID: Date: 4 Jan 91 04:24:41 GMT Sender: nanotech@athos.rutgers.edu Organization: University of Cincinnati, Cin'ti., OH Lines: 92 Approved: nanotech@aramis.rutgers.edu In article rjenkins@.com (Robert Jenkins) writes: >Suppose we develop direct brain/computer interfaces. I imagine this means >that accessing computer memory would seem the same to us as remembering >things on our own. This depends, of course, on just how direct "direct" turns out to be. The more I use computers, the less optimistic I become about the prospects for ever associating the word "direct" with a solution that involved a computer. Rather, when I think of involving a computer in a problem, I picture the street vendor from the film "Life of Brian" who refused to sell anything outright, but instead insisted to his customers: "You must haggle with me." A "direct" brain/computer interface would be useful, but unless by "computer" we are talking about something more robust than the word connotes today, such an interface might be a prescription for insanity. >Then, if we build computers that can think faster than us, we could link to >them, tell them our problems, then "remember" the computer's solutions. >We could even remember the steps the computer used to reach those solutions. >If we teach the computer to think like we do, how could we distinguish this >from just solving the problems ourselves? How does the situation you describe differ from having a very smart professor following you around and whispering in your ear the solution to every problem? I think you would be very aware that the professor was solving the problems, and not you. However, if you knew that you could *always* count on the professor being there for you, you might start to internalize her abilities, in a sense. Just as you know that when you reach your hand out it will grasp things, you might come to view the professor's intellect as part of your own. Come to think of it, something like this does happen to every person who works in a management-type position. Even if the manager can't solve the problem himself, he becomes quite good at matching the abilities of his subordinates to problems. If we teach the computer to think like we do, the computer may not be able to report the "steps" it used to solve many (perhaps most) problems. That is because we don't usually think in "steps". Or because the steps we can report do not always capture the essence of problem-solving, whatever that is. For example, suppose you solve a problem by numerically integrating an equation that you derived from first principles. You can describe the steps you took. You probably can't describe exactly how you inferred, from the problem statement, that these were the steps to take. You don't have to look very far into your own thought processes before you see no further conscious "steps". At some point, and a very close one, you just "know". >For that matter, we could download all brainstorming, reasoning, and even >judging of alternatives into computers of our own design, then just >remember the appropriate results. The bandwidth required for remembering >final results would be fairly small. This is only true if the computer is super-intelligent enough to feed us only the appropriate final results. And also if the problems we are considering have succinct answers. Many interesting problems do not. For example, how does one fly to Jupiter? I don't think a plan to get to Jupiter is going to be very short, at least in light of currently-available technology. Someday, we may have technological infrastructure that makes flying to Jupiter as simple as flying to France. But you can't make a very good name for yourself today by solving the problem of flying to France (the way anybody else does). Therefore I suspect we will always want to solve the remaining problems that lack succinct solutions. > The human nervous system would remain >relatively intact, yet people could think (and invent and code) as fast as >the top-of-the-line nanocomputers. I have no doubt that computers can increase intellectual efficiency, but the greatest progress to date has been in solving routine, repetitive problems. Invention does involve some repetitive steps, but much of it seems unique to every new problem and domain. Since we do not seem able to discover any generally applicable principles of invention and discovery, we have an enormous disparity between intellectual leverage in well-understood vs. poorly-understood domains. To see this, one only needs to try to solve a problem that nobody knows how to solve today. Once a generation of scientists and engineers have trampled a problem domain sufficiently, an average person can work wonders in it. But at the frontiers, our productivity is very low. Years seem to go by in which many workers generate only enough results collectively to fill a few college course hours. -- Dan Mocsny Snail: Internet: dmocsny@minerva.che.uc.edu Dept. of Chemical Engng. M.L. 171 dmocsny@uceng.uc.edu University of Cincinnati 513/751-6824 (home) 513/556-2007 (lab) Cincinnati, Ohio 45221-0171