Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!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: 7 Dec 90 08:35:21 GMT Sender: nanotech@athos.rutgers.edu Organization: University of Cincinnati, Cin'ti., OH Lines: 99 Approved: nanotech@aramis.rutgers.edu >[Actually, the million mark in increased intelligence probably is > the level we can expect to get *without* some fundamental increase > in knowledge about intelligence, simply by simulating the existing > structure but making it faster. Combine the raw speed with a > built-in library, and the resulting entity can apply to any problem > in 5 minutes the effect of 10 years of in-depth study and research, > current human scale. > --JoSH] I like this idea, but I think the term "simply" looks a little out of place on the third line. :-) No system that can accurately simulate a human brain is going to satisfy the ordinary definition of "simple". However, assuming that we can "simply" simulate the underlying structure and get the same result, simply turning up the clock is going to change a lot of things, I think. Remember that intelligence as we (hardly) know it appears to have a significant real-time component. This could create the following problems: 1. Even though the characteristic times of the underlying physical components of the system may appear to be slow, they may depend intimately and subtly on physical processes that occur very rapidly. This could make the system as a whole difficult to simulate at higher speeds. 2. An entity capable of simulating a human brain at 1e+6 times greater speed would lead a very curious existence indeed. For example, humans seem to require sleep, and this may play a non-trivial role in converting short-term memories to long-term. Simply speeding up the clock without any understanding of what is going on would require preserving all possible aspects of present slow brain function. Thus the mega-brain would oscillate continuously between sleep and waking modes. 3. Speeding up the brain would be equivalent to locking the hapless entity into a prison of agonizing slowness. Surrounding phenomena, as well as the entity's own physical mobility, would effectively slow down by 1e+6 times. Imagine living in a world where reaching out your arm and picking up an object for inspection takes not 1 second of perceived time, but 1e+6 seconds = 11.6 days. By the time the object was where you wanted it, would you still remember why you wanted to pick it up in the first place? (Remember, you went through over 10 sleep/wake cycles in the meantime.) How long can a person sit still without going nuts? This would be like being paralyzed. 4. The above point is not trivial. The human brain has apparently evolved to solve problems of interacting with the physical world within definite time constraints. Upsetting those time relationships may cause things to start breaking. My guess is that the mega-brain would avoid going insane by giving up on most interaction with the real world, and instead withdraw into an introspective, simulated world where things happened at a satisfying speed. 5. This wouldn't necessarily hamstring our mega-brain, but it might cut its effectiveness appreciably. Effectively using a 1e+6 times faster brain would probably require a simulated world of commensurate speed for that brain to play in. However, this is sounding like a much harder problem than "simply" speeding up a brain. And how will we persuade the mega-brain to come out of dreamland and divulge its latest findings to the slow world? 6. Even if we can simulate both a researcher's brain and a physical world for that brain to do its research in, the resulting research output will "only" be a computational experiment. How many researchers can produce 10 years of results with only 5 minutes of real-world experimenting? Would a researcher have the patience and dedication to pursue a train of thought that might require *millennia* of perceived time to verify? Because of the above problems, I suspect that "simply" speeding up a simulated brain will be anything but simple. The simulators will not be able to get away with being naive. I suspect that a direct simulation will not work very well. Rather, the designers will have to speed up different components of thought to different degrees. To do this and yield a working system will necessitate a detailed understanding of the mechanisms of thought. You can't make a balloon fly faster by strapping a jet engine onto it. You can't make a light bulb brighter by increasing the applied voltage by 1e+6 (for very long, anyway :-) The human brain is the most complex system known. Its internal and external interactions are devilishly complicated. Those interactions may be time-dependent and brittle. Changing one global design variable may start breaking things left and right. -- 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 [As you note, speeding the brain up is not "simple" in some absolute sense. However, even with all the quite valid complications you have pointed out, it is vastly simpler than changing the organization of the brain to obtain enhanced intelligence. --JoSH]