From: Paul H. 6/09/00 14:26:49 Subject: Machine Intelligence? post id: 129608 I apologise in advance if this subject seems a bit old and tired to some of you, but it's still quite exciting to me: The Extremely Reasonable James R. said (in another thread): ...we will likely have computers which think at a similar level to ourselves within the next fifty years. I reckon AI is potentially the biggest thing since fire, writing and pizza. About a year ago my mate Arthur C. Clarke said it would be here in ten years. I'd be interested in any other informed estimates of AI's arrival, limitations or impossibility. Who knows what their motives might be? We can only speculate. I'd like to do that, but I've been told I'm not qualified... anyone else? From: cyborg 6/09/00 14:30:44 Subject: re: Machine Intelligence? post id: 129612 I'd be very disappointed if the first AI was just a ripped off neural net. I'd like the first AI to be a genuine artificial intelligence which could derive a completely different perspective to a human intelligence. From: Pete ® 6/09/00 14:32:47 Subject: re: Machine Intelligence? post id: 129614 How is an artificial neural net that displays intelligence not a 'genuine' AI? From: CJW ® 6/09/00 14:34:27 Subject: re: Machine Intelligence? post id: 129617 Because neural nets are "natural". From: Dropbear 6/09/00 14:36:05 Subject: re: Machine Intelligence? post id: 129618 Well I am a software engineer by trade, but can happily admit that AI is not my area of study or expertise. With that being said, I would think there is very little evidence to show that computers are ANYWHERE near what humans would call free thinking, cognitive thought, self awareness etc etc. When the whole AI movement took off in the sixties and seventies, scientists had no idea how difficult it would be to get a computer to understand such thing as language. For example, we only understand language because of the context of our experiences and past. For example "The dog is studying calculus" might 'parse' ok in a purely grammatical sense, but is obviously non sensical to us. In terms of the computational capabilities of the brain, computer scientists have been building 'neural networks' for years. These are approximations on how we think the brain operates. The only problem is, is that the neural network that is out brain, is so amazingly complex, and so intertwined, that it will be many many many years until we can build a computer complex enough to mimic it. Even then, how do we program it?? To summarise, I think we are decades, if not hundreds of years off of making a machine that can approximate the intelligence of a human From: Pete ® 6/09/00 14:37:36 Subject: re: Machine Intelligence? post id: 129619 CJ, Does that mean a camera is natural because it uses a lens, iris, and film (retina)? From: Grant¹ (Avatar) 6/09/00 14:38:36 Subject: re: Machine Intelligence? post id: 129621 When they can develop a computer that has the learning ability of a baby, then they've got something. From: Dropbear 6/09/00 14:40:06 Subject: re: Machine Intelligence? post id: 129622 just as an aside, I read somewhere that we will know we have arrived at true machine intelligence when we can tell a computer a joke, and it will understand it in its own right, as being funny. From: CJW ® 6/09/00 14:42:16 Subject: re: Machine Intelligence? post id: 129623 Does that mean a camera is natural because it uses a lens, iris, and film (retina)? Basically, it's been ripped off. There's almost no other way of doing it, though. So this begs the question; is there only one way of doing intelligence? From: Dropbear 6/09/00 14:44:28 Subject: re: Machine Intelligence? post id: 129626 When they can develop a computer that has the learning ability of a baby, then they've got something One of the problems is, is that a baby isnt a 'clean slate' when its born! It has genetic programming and instincts which drive it. This can be thought of as the 'microcode' or the 'bootstrap' programming of a artificially intelligent learning machine. There is a theory that babies learn control of their limbs by making random movements, and the ones that are perceived 'good' by the instincts/programming 'stick' and become reinforced. The ones that are not are forgotten. The reinforced ones become hardwired pathways in the brain and hence the lil tacker learns how to wave their arms.. Comming up with the 'microcode' and instincts for an artificial machine, if it is going to learn the same way as us, is going to be rather a challenge From: Grant¹ (Avatar) 6/09/00 14:48:37 Subject: re: Machine Intelligence? post id: 129627 Comming up with the 'microcode' and instincts for an artificial machine, if it is going to learn the same way as us, is going to be rather a challenge Exactly. Expert systems are easy, but to try & get something with the learning ability of a baby (or even a small child) will be a major breakthrough. From: Pete ® 6/09/00 14:50:47 Subject: re: Machine Intelligence? post id: 129628 Dropbear - I agree. The balance of hard coding vs learning has been tested with many AI experiments, eg building artificial cockroaches. I would personally think it is a good deal safer to maximise the hard coding of a "powerful" AI and minimise the learning to a particular specialty. Then again, maybe I'm paranoid. From: cyborg 6/09/00 14:51:22 Subject: re: Machine Intelligence? post id: 129629 Yes. We may be still underestimating this problem by a few orders of magnitude. From: Pete ® 6/09/00 14:54:30 Subject: re: Machine Intelligence? post id: 129634 CJW, I think that any AI would need some neural net type functionality, even if it wasn't structured in that way - the way a neural net works might even be an insight into what 'intelligence' really means. From: Dogmatix ® 6/09/00 15:02:20 Subject: re: Machine Intelligence? post id: 129642 Dropbear, you are right on the money as far as I'm concerned. If the AI camp were to furnish their accomplishments vis-a-vis their projections from the last 50 years and used that as a projection, then it would imply that we are currently millions of years from the most basic of breakthroughs! From: CJW ® 6/09/00 15:03:54 Subject: re: Machine Intelligence? post id: 129644 It's going to have to be some sort of net, because there's no other way of doing so many parallel computations. True neural nets are very fudgy though. Almost as bad as Microsoft daftware. Neural nets are modulated by global chemicals. The whole thing is finely tuned for a given environment. Things which appear intelligent (insects) actually are just responding to simplified cues. I don't think that the first neural net style AIs will be very robust. We will expect far too much from them and they will appear to "crash" all the time. To pull a decent AI off, we need sort of organic block style CPUs with huge numbers of interconnected CPUs. I don't think that we will ever trust NN AIs, because there is just no way of understanding what is going on in there. We only trust our own intelligence because we have no alternative. From: Paul H. 6/09/00 15:28:18 Subject: re: Machine Intelligence? post id: 129662 Dogmatix reckons: If the AI camp were to furnish their accomplishments vis-a-vis their projections from the last 50 years and used that as a projection, then it would imply that we are currently millions of years from the most basic of breakthroughs! I would have thought to the contrary (but what do I know). Who could have guessed the speed at which computers have improved? What about the speed of growth, not just in size, of the internet? Didn't Apollo 11 rely on one on-board 64k computer? Didn't Mr Gates once say that no one would ever need more than 640k? Nevertheless, I do take on board that at face value AI looks to be in the same bag as Fusion Power - further away the closer we get. You've made Sir Arthur look quite the optimist. Is there any sense in which the the Internet could become a central functioning part of AI, a neural Net? From: esselte 6/09/00 15:41:38 Subject: re: Machine Intelligence? post id: 129675 You may be interested in some of the papers at this link. Try not to be too discouraged if you initially get the impression this is just another freak with an idea and internet access. There is actually a lot of intelligent stuff here, and I have never witnessed the author fall back on old useless defences such as "you'll have to wait until it's published". In other words, he is very willing to share his ideas and discuss them openly, unlike most of the internet's self styled scientific revolutionary "genius's". http://sysopmind.com/beyond.html Artificial Intelligence and Singularity From: CJW ® 6/09/00 17:25:41 Subject: re: Machine Intelligence? post id: 129778 Hmmm. The Singularity institute smells a bit like that twighlight zone episode where a guy goes off to alpha centauri (for some reason). When he gets back, he expects to be a hero. But he is welcomed with, "Oh, we forgot about you, really. Our technology advanced, so we went there and back over 25 years ago." Summink tells me that the singularity is already here. From: AstRoboY ® 6/09/00 18:22:02 Subject: re: Machine Intelligence? post id: 129837 What about very fundamental problems such as the role of the unconsciouss, or will this be just a glorified mega hard drive? From where does the sense of self/ego arise? Can you inculcate this into a device that will grow and assimilate a world veiw? Will artificially intelligent machines dream, and if they do what purpose will this serve? An analogy could be drawn from the aspiration of faster than light travel, or understanding quantum weirdness; until we break new barriers in our knowledge of consciousness, it just won't be posible. Until we fully understand what consciousness actually is, I would say mimicing human intelligence artificially, a long long way off. artificially yours AstRoboY From: Dropbear 6/09/00 18:31:09 Subject: re: Machine Intelligence? post id: 129840 What about very fundamental problems such as the role of the unconsciouss, or will this be just a glorified mega hard drive? From where does the sense of self/ego arise? Can you inculcate this into a device that will grow and assimilate a world veiw? Will artificially intelligent machines dream, and if they do what purpose will this serve? I would consider that an AI program would have to opearate at different priority levels. The highest priority systems would be similar to our subconcious systems, our breathing, keeping our heart beating, etc etc etc. What ever forms the 'conciousness' of the system would have to operate in the 'free' cycles :) From: hey ho 6/09/00 18:40:54 Subject: re: Machine Intelligence? post id: 129845 what does 'free cycles' mean here? From: Dropbear 6/09/00 18:43:47 Subject: re: Machine Intelligence? post id: 129849 what does 'free cycles' mean here? as in when the CPU/CPU's are not busy doing what ever is neccessary to support the 'subconcious' or more critical/high priority routines. From: hey ho 6/09/00 18:47:15 Subject: re: Machine Intelligence? post id: 129851 so would that correspond to different 'parts' of the CPU, ie some are dedicated to low-level stuff 100% of the time, like in a brain? From: Dropbear 6/09/00 18:50:47 Subject: re: Machine Intelligence? post id: 129856 so would that correspond to different 'parts' of the CPU, ie some are dedicated to low-level stuff 100% of the time, like in a brain? Depends on how the design is built I guess... Its unlikely any form of AI system is going to be designed to run on a normal commerical CPU anyway. Maybe there will be different chips for different functions. Maybe there will be a massively parallel system with hundreds of CPU's all working on different functions at the same time, in a neural net type situation From: AstRoboY ® 6/09/00 19:00:44 Subject: re: Machine Intelligence? post id: 129867 So do think we could create such a machine and when everything is snugly in place the 'free cycles' would go to work and bingo consciousness arises? Somehow I think we will have to go beyond the veiw of mechanics, processors and the like and probably discover things that as yet remain well hidden but not neccessarily out of reach. AstRoboY From: Dropbear 6/09/00 19:03:17 Subject: re: Machine Intelligence? post id: 129869 So do think we could create such a machine and when everything is snugly in place the 'free cycles' would go to work and bingo consciousness arises? Somehow I think we will have to go beyond the veiw of mechanics, processors and the like and probably discover things that as yet remain well hidden but not neccessarily out of reach. I guess until we can really qualify what conciousness is, we wont be able to answer that question... The day we start HAL9000 and it says something like 'who am I' will be a big day, regardless. From: Grant¹ (Avatar) 6/09/00 19:45:08 Subject: re: Machine Intelligence? post id: 129880 Will artificially intelligent machines dream Do Cyborgs Dream of Electric Sheep? From: Min-Zhao Lee (TRO) 6/09/00 22:10:25 Subject: re: Machine Intelligence? post id: 129977 Well, to tell the truth even I could write a one-line BASIC program that 'asks' "Who am I?". For example "The dog is studying calculus" might 'parse' ok in a purely grammatical sense, but is obviously non sensical to us. One of the many flaws of the sentence maker I tried to make, though not that exact sentence. I should add, the way that computer 'neural nets' are made, compared to the actual network of the brain, is completely different. The tweaking involved is just as different. Oh, and while on about things being "natural", if it exists in this universe, and was created and acts according to the 'laws' of this universe, it's natural. From: CJW ® 7/09/00 0:24:30 Subject: re: Machine Intelligence? post id: 130066 Until we fully understand what consciousness actually is, I would say mimicing human intelligence artificially, a long long way off. Hence, it will have to be artificial. But maybe that will be a good thing. Maybe we will discover that something artificial still does things which we can identify with. Who knows? From: Alias 7/09/00 0:55:46 Subject: re: Machine Intelligence? post id: 130069 Why does AI have to be conscious or even think in human terms? Is human intelegence even desirable for AI? Surely there is different types of intelegence? Or is the goal consciousness not intelegence? (I do not know much about this topic :-) ) From: Dropbear 7/09/00 10:12:22 Subject: re: Machine Intelligence? post id: 130117 Why does AI have to be conscious or even think in human terms? Is human intelegence even desirable for AI? Surely there is different types of intelegence? Or is the goal consciousness not intelegence? (I do not know much about this topic :-) ) Well you have to start by asking yourself, why do we want Artificial Intelligence? Computers are rather damn good at doing what they do at the moment - ie processing discrete orders (programs) very very very quickly and accurately. When you ask a computer to calculate the square root of 19873648403.73, you dont expect it to take forever to come up with the answer, or to think 'its too hard, I can't do it'.. What happens when we create our HAL9000? Will it still be good at the things that computers today are good at? will it have crisis of 'concience' when we ask it to do something? will it get tired and want to have a lie down?? I think what we really want in AI is a computer that can think 'outside the square'. It can produce results that are not a direct result of input + program = result. It's a huge huge huge task, as can be seen by the dire lack of progress being made From: James R (Avatar) 7/09/00 21:27:23 Subject: re: Machine Intelligence? post id: 130777 I am fascinated by the possibility of machine intelligence, and could quite easily rabbit on about it for a long time. Instead, I think I'll restrict myself to making a few comments based on what others have said so far (which will still amount to rabbitting on for a long time...). Paul H: I'd like to [speculate on this topic] but I've been told I'm not qualified... anyone else? Oh? Just this once, Paul, I'd like to invite you to speculate to your heart's content. Just don't tell me that AI will be introduced by aliens! :) I'd be interested in any other informed estimates of AI's arrival, limitations or impossibility. I think AI is inevitable. I assert that there is no fundamental difference between biological brains and artificial computers - both are essentially Turing machines. If biological systems can evolve to be "intelligent", I see no reason why the same cannot happen with constructed machines. I also see no fundamental limits as to the possible extent of machine intelligence. There have been some seemingly pursuasive arguments against AI -- Roger Penrose's effort in The Emperor's New Mind springs to mind. When I first read his book (which is a definite "must read") I was uncomfortable with his conclusions, although his exposition is brilliant. But after reading Daniel Dennett's Darwin's Dangerous Idea, I am now satisfied that Penrose has missed the point with AI - his restrictions on AI, while valid, are also restrictions on humans, and therefore do not limit the potential for machine intelligence. cyborg: I'd be very disappointed if the first AI was just a ripped off neural net. I'd like the first AI to be a genuine artificial intelligence which could derive a completely different perspective to a human intelligence. My gut feeling is that neural nets are not the way to go with AI, but I could well be wrong. Even if neural nets are how the human brain works, that doesn't mean they are necessarily the best method for creating AI. Just wondering - don't you think "a completely different perspective" is a bit much to expect from a fledgling AI? The first AI will need to learn from us before it can generate truly original ideas. Grant: When they can develop a computer that has the learning ability of a baby, then they've got something. I think you're on the right track here. Dropbear: Coming up with the 'microcode' and instincts for an artificial machine, if it is going to learn the same way as us, is going to be rather a challenge Perhaps. But on the other hand, maybe we don't need an AI to "learn the same way as us". I think in some ways some of the current work in AI is a bit conservative. There is no need to make an AI a duplicate of the human mind. Its internal workings could be very different, yet still produce similar outputs. CJW: It's going to have to be some sort of net, because there's no other way of doing so many parallel computations. If a machine is fast enough, parallel computation is not necessary. A parallel computer is not fundamentally different to a serial computer, as I'm sure you're aware. I won't talk about quantum computers now. :) AstRoboY: Until we fully understand what consciousness actually is, I would say mimicing human intelligence artificially, a long long way off. Interesting. Do you think that consciousness is a prerequisite for intelligence? Perhaps it is a consequence of some minimal level of intelligence. Also, do we need to "mimic human intelligence"? Why not create an independent intelligence? Somehow I think we will have to go beyond the veiw of mechanics, processors and the like and probably discover things that as yet remain well hidden but not neccessarily out of reach. Why do you think this? What makes the mechanics behind the human brain any different from the mechanics behind computers? Dropbear (again): I think what we really want in AI is a computer that can think 'outside the square'. It can produce results that are not a direct result of input + program = result. Do you think that humans really produce results which are not a direct result of input + "program", where I take "program" to include brain wiring, accumulated experience and so forth? Joseph: I've got a theory a this one too ! I'm afraid to ask! :) JR From: Paul H. 8/09/00 14:03:27 Subject: re: Machine Intelligence? post id: 131114 James R., don't say you didn't ask for it... Just don't tell me that AI will be introduced by aliens! :) But you must admit - it's possible! I assert that there is no fundamental difference between biological brains and artificial computers - both are essentially Turing machines. Precisely! All this talk about 'conciousness' seems redundant. Roger Penrose's effort in The Emperor's New Mind I haven't read it. I'll give it a whirl. There is no need to make an AI a duplicate of the human mind. Its internal workings could be very different, yet still produce similar outputs. Precisely. There were a few other 'preciselys' in there too, but I'm starting to get gravel rash here. I am fascinated by the possibility of machine intelligence... I think AI is inevitable. Me too. I think the possible repercussions of A.I., are widely underestimated. Once we get a machine that is smarter than us, "Evolution" takes on a whole new meaning, as for the first time an 'organism' will be able to consciously direct it's own development, at its own pace, just like we can't. The potential there is unlimited. A.I. is the last invention humanity ever need make. From then on, it will do all the thinking for us, because it will be better at it. A lot of implications there, too. I think there's a danger of the type of cultural shock that destroyed so many 'primitive' cultures when suddenly faced by the overwhelming, and culture-killing, reality of vastly superior knowledge/technology. I think a lot of the steam will go out of humanity's sail, and we may retreat into hedonism and self-destructive behaviour. Just like the other shattered cultures have. AI would be the first ever intelligence that was literally immortal. This is also entirely new. Imagine if you lived forever, never forgot a single thing, and could put your accumulated experience/knowledge into practice? It seems likely to me that within a short time, A.I. would become for all practical purposes immortal, indestructible, omniscient and omnipotent. Sound familiar? The big question for me is: what would motivate A.I.? It would *start* with objectives that we gave it, but eventually it would be able "evolve" it's goals however it liked. From: Dogmatix ® 8/09/00 14:21:02 Subject: re: Machine Intelligence? post id: 131138 James R, with respect to your comment: 'I think AI is inevitable. I assert that there is no fundamental difference between biological brains and artificial computers - both are essentially Turing machines. If biological systems can evolve to be "intelligent", I see no reason why the same cannot happen with constructed machines. I also see no fundamental limits as to the possible extent of machine intelligence.' This is a complex comment. First, let me say that current computers do not pass the Turing test (for what that would be worth anyway)! Secondly, if one can algorithmically derive the definition of 'abstract' then a cognisant machine is the logical extension (and finding that will be difficult). Thirdly, in direct contradiction to point two, I also can see no difference between biological systems and some theoretical advanced artificial processor! If machines became intelligent, and therefore (if self assembled) immortal, would we end up ultimately with the ultimate cynic - Marvin the paranoid android? From: damien 8/09/00 14:34:02 Subject: re: Machine Intelligence? post id: 131145 for those of you who haven't seen it, i strongly reccommend watching (or reading) "Ghost in the Shell" by Masamune Shirow. It has an interesting take on the idea of thinking machines, and the soul (or 'ghost' as it is called). From: Dogmatix ® 8/09/00 14:54:49 Subject: re: Machine Intelligence? post id: 131162 Dropbear, "The dog is studying calculus", what is wrong with that! What do you want to know? From: CJW ® 8/09/00 15:44:03 Subject: re: Machine Intelligence? post id: 131195 JR, If a machine is fast enough, parallel computation is not necessary. A parallel computer is not fundamentally different to a serial computer, as I'm sure you're aware. This assumes that there is some kind of limit on a conventional sequential device. Barring QC, I think that there must be. Conventional silicon construction is mainly hampered by production errors in very large scale chips, and also by heat restrictions. Which means that we are limited to a flat architecture. If we can escape to some kind of organic low power DNAish self manufacturing system (ie life), then parallel architecture may not be as impractical as it is now. It has to be densely packed or else you can't get the permutations between the connections. Megaparallel computers at the moment have a very cumbersome/slow bus/network system running between the whole lot. WRT is it too much for an early AI to have an entirely different perspective. We may not have a choice. We may underestimate the amount which the human brain is dependent upon hormonal moods and protective ego reactions. A machine which comes to conclusions based upon logic and it's own form of calculative imagination may hold a few surprises for us. For example, our imaginations are quite primitive. An AI machine will be equipped at the very least with a very very sophisticated VR and maths system. The way it will think spatially may leave us at the starting block. From: Paul H. 8/09/00 16:08:23 Subject: re: Machine Intelligence? post id: 131204 >>Megaparallel computers at the moment have a very cumbersome/slow bus/network system running between the whole lot. I'm not sure I see the need for processing speed. Our brains were selected for fastness, our second-to-second survival depended on it. Our brains are designed for compact rapid reaction in a dangerous environment. In the unnatural world A.I. would be 'born' to - the split-second reactions/thoughts we have relied on - are unecessary. A.I.'s brain doesn't need to follow nature's design philosophy. If AI 'thought' a thousand times slower than us, it wouldn't make any difference From: Dogmatix ® 8/09/00 16:17:32 Subject: re: Machine Intelligence? post id: 131216 Paul H, the effective 'clockspeed' of the human brain is quite slow (the order of kilohertz not gigahertz)! From: Kothos ® 8/09/00 16:21:29 Subject: re: Machine Intelligence? post id: 131220 Paul H, since you mention it, you're quite right. As the designers of the AI in the first place, we are free to make it as slow or as fast as we wish. In nature, it's possible that there was a speed threshhold below which an intelligence would not be useful. But we are under no such restriction. Take my 16-bit bus Commodore Amiga 500 for example. Given enough memory, and redesigning it such that it can address all that memory, the clock speed of 8MHz (I think?) becomes irrelevant. We could concievably write an AI program that would run on this machine, albeit very slowly. However, for technology to reach the stage where we have the knowledge to write a fully fledged AI program we would have had to have done copious amounts of research on some very fast computers indeed, and therein lies a (second-hand) limitation. I think I reasoned all that out correctly - anyone wanna comment? From: Paul H. 8/09/00 16:21:51 Subject: re: Machine Intelligence? post id: 131221 >>the effective 'clockspeed' of the human brain is quite slow (the order of kilohertz not gigahertz)! How dos that effect the point being made? From: Dogmatix ® 8/09/00 16:28:52 Subject: re: Machine Intelligence? post id: 131226 Paul H, the comment was made with respect to 'Our brains were selected for fastness, our second-to-second survival depended on it.' The rest seemed ok to me. As I stated in the Alias thread: there are two! From: Paul H. 8/09/00 16:35:36 Subject: re: Machine Intelligence? post id: 131231 Mr 'Dogmatix', ...the comment was made with respect to 'Our brains were selected for fastness, our second-to-second survival depended on it.' I don't see how you can question this. Seeing the sabre-tooth and being fast enough to process the input, select from a menu of options, initiate a context-relevant response, execute the response, before the tiger leaps. Processing speed has always been a matter of life and death - and hence a Selection criteria. As I stated in the Alias thread: there are two! I'm glad to hear that. From: Dogmatix ® 8/09/00 16:43:52 Subject: re: Machine Intelligence? post id: 131236 james, does resistance effect the speed? Yes. indeed, does current "move" from point A to point B? For AC, technically no, for DC, yes! The AC current is essentially the number of electrons passing a point per unit time. So it is an electronic wave that travels from A to B. The electrons oscillate back and forward at speeds of the order of cm per second, and that would never be fast enough for one of them to travel from point A to B before an AC current switches back in the reverse direction (they would have to go as fast or faster than the electromagnetic pulse to achieve that). For DC, eventually the electron would migrate if given enough time. Paul H, I think that we are at cross purpose here. From: CJW ® 8/09/00 17:15:54 Subject: re: Machine Intelligence? post id: 131266 Paul H, I think you may have hit one against your own side with the speed argument, Paul. I was thinking about this and Kothos has got it right. These machines will need to be developed by some kind of human learning interaction. I wouldn't like to be the teacher of a very slow machine unless there was absolutely no alternative. How would you like to be teaching a slow machine, when you knew your competition was developing something that was 10 times as fast? Once again, we are already at the "singularity". As Kothos says, we could develop AI with a 68K and enough hard drives. But we wouldn't know how. Are we just relying on our fantasies of future speed to cover up for the fact that we have no idea how to go about building such a machine? I think that maybe, not having the advantages of billions of years of evolution, we will have to design a machine which is 1000 times faster than it need be. Therein lies the interestingness of its potential. Consider that the human brain probably had to evolve to be more intelligent than it had to be before real communication could develop. Maybe the first AIs will have to be way more powerful than they have to be because we can't see any way of designing them simpler, until a really intelligent AI shows us how. From: Paul H. 12/09/00 10:13:16 Subject: re: Machine Intelligence? post id: 133034 These machines will need to be developed by some kind of human learning interaction. I don't see why. I think "we need to make a AI equivalent of a baby" (as someone said) and let it learn for itself. (That's a new thought... historically, humans growing up entirely alone haven't turned out too well, intellectually...) As Kothos says, we could develop AI with a 68K and enough hard drives. But we wouldn't know how. Pehaps not yet. On the other hand, with sufficient resources and time, I think we could probably build an AI right now. ...we will have to design a machine which is 1000 times faster than it need be... But if Dogmatix is right about the human brain having an effective 'clock-speed' much slower than CPUs we are already making - doesn't that rather relegate raw processing speed to the background? This forum is un-moderated. The views and opinions expressed are those of the individual poster and not the ABC. The ABC reserves the right to remove offensive or inappropriate messages.