Re: [agi] What Must a World Be That a Humanlike Intelligence May Develop In It?
2009/1/12 Ben Goertzel b...@goertzel.org: The problem with simulations that run slower than real time is that they aren't much good for running AIs interactively with humans... and for AGI we want the combination of social and physical interaction There's plenty you can do with real-time interaction. OTOH, there's lots you can do with batch processing, e.g. tweak the AI's parameters, and see how it performs on the same task. And of course you can have a regression test suite of tasks for the AI to perform as you improve it. How useful this sort of approach is depends on how much processing power you need: if processing is very expensive, it makes less sense to re-run an extensive test suite whenever you make a change. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=126863270-d7b0b0 Powered by Listbox: http://www.listbox.com
Re: [agi] What Must a World Be That a Humanlike Intelligence May Develop In It?
2009/1/9 Ben Goertzel b...@goertzel.org: Hi all, I intend to submit the following paper to JAGI shortly, but I figured I'd run it past you folks on this list first, and incorporate any useful feedback into the draft I submit Perhaps the paper could go into more detail about what sensory input the AGI would have. E.g. you might specify that its vision system would consist of 2 pixelmaps (binocular vision) each 1000x1000 pixels, in three colours and 16 bits of intensity, updated 20 times per second. Of course, you may want to specify the visual system differently, but it's useful to say so and make your assumptions concrete. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=126863270-d7b0b0 Powered by Listbox: http://www.listbox.com
Re: [agi] Universal intelligence test benchmark
2008/12/29 Matt Mahoney matmaho...@yahoo.com: --- On Mon, 12/29/08, Philip Hunt cabala...@googlemail.com wrote: Incidently, reading Matt's posts got me interested in writing a compression program using Markov-chain prediction. The prediction bit was a piece of piss to write; the compression code is proving considerably more difficult. Well, there is plenty of open source software. http://cs.fit.edu/~mmahoney/compression/ If you want to write your own model and just need a simple arithmetic coder, you probably want fpaq0. Most of the other programs on this page use the same coder or some minor variation of it. I've just had a look at it, thanks. Am I right in understanding that the coder from fpaq0 could be used with any other predictor? -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Universal intelligence test benchmark
2008/12/27 Matt Mahoney matmaho...@yahoo.com: --- On Fri, 12/26/08, Philip Hunt cabala...@googlemail.com wrote: Humans are very good at predicting sequences of symbols, e.g. the next word in a text stream. Why not have that as your problem domain, instead of text compression? That's the same thing, isn't it? Yes and no. What i mean is they may be the same in principle, but I don't think they are in practice. I'll illustrate this by way of an analogy. The Turing Test is considered by many to be a reasonable definition of intelligence. And I'd agree with them -- if a computer can fool sophisticated alert people into thinking it's a human, it's probably at least as clever as a human. Now consider the Loebner Prize. IMO this is a waste of time in terms of advancement of AI because we're not anyway near advanced enough to build a machine that can think as well as a human. So programs that are good at the Loebner prize as so not because they have good AI architectures, but because threy employ clever tricks to fool people. But that's all there is -- clever tricks with no real substance. Consider compression programs. I have several on my computer: zip, compress, bzip2, gzip, etc. These are all quite good at compression (they all seem to work well on Python source code, for example), but there is not real intelligence or understanding behind them -- they are clever tricks with no substance (where by substance I mean intelligence). Now, consider if I build a program that can predict how some sequences will continue. For example, given ABACADAEA it'll predict the next letter is F, or given: 1 2 4 8 16 32 it'll predict the next number is 64. (Whether the program works on bits, bytes, or longer chunks is a detail, though it might be an important detail.) Even though the program is good at certain types of sequences, it doesn't do compression. For it to do so, I'd have to give it some notation to build a compressed file and then uncompress it again. This is a lot of tedious detail work and doesn't add to it's intelligence. IMO it would just get in the way. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Universal intelligence test benchmark
2008/12/28 Philip Hunt cabala...@googlemail.com: Now, consider if I build a program that can predict how some sequences will continue. For example, given ABACADAEA it'll predict the next letter is F, or given: 1 2 4 8 16 32 it'll predict the next number is 64. (Whether the program works on bits, bytes, or longer chunks is a detail, though it might be an important detail.) Even though the program is good at certain types of sequences, it doesn't do compression. For it to do so, I'd have to give it some notation to build a compressed file and then uncompress it again. This is a lot of tedious detail work and doesn't add to it's intelligence. IMO it would just get in the way. Furthermore, I don't see that a sequence-predictor should necessarily attempt to guess the next in the sequence by attempting to generate thre shortest possible Turing machine capable of producing the sequence (certainly humans don't work that way). If sequence-predictor uses this method and is good at predictinbg sequences, good; but if it uses anotherm ethod and is good at predicting sequences, it's just as good. What matters is a program's performance, not how it does it. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Universal intelligence test benchmark
2008/12/29 Matt Mahoney matmaho...@yahoo.com: Please remember that I am not proposing compression as a solution to the AGI problem. I am proposing it as a measure of progress in an important component (prediction). Then why not cut out the middleman and measure prediction directly? I.e. put the prediction program in a test harness, feed it chunks one at a time, ask it what the next value in the sequence will be, tell it what the actual answer was, etc. The program's score is then simply the number it got right divided by the number of predictions it had to make. Turning a prediction program into a compression program requires superfluous extra work: you have to invent an efficient file format to hold compressed data, and you have to write a decompression program as well as a compressor. Furthermore there are bound to be programs that're good at compression but not good at prediction. Whereas all programs that're good at prediction are guaranteed to be good at prediction. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Universal intelligence test benchmark
2008/12/29 Philip Hunt cabala...@googlemail.com: 2008/12/29 Matt Mahoney matmaho...@yahoo.com: Please remember that I am not proposing compression as a solution to the AGI problem. I am proposing it as a measure of progress in an important component (prediction). [...] Turning a prediction program into a compression program requires superfluous extra work: you have to invent an efficient file format to hold compressed data, and you have to write a decompression program as well as a compressor. Incidently, reading Matt's posts got me interested in writing a compression program using Markov-chain prediction. The prediction bit was a piece of piss to write; the compression code is proving considerably more difficult. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Universal intelligence test benchmark
2008/12/26 Matt Mahoney matmaho...@yahoo.com: I have updated my universal intelligence test with benchmarks on about 100 compression programs. Humans aren't particularly good at compressing data. Does this mean humans aren't intelligent, or is it a poor definition of intelligence? Although my goal was to sample a Solomonoff distribution to measure universal intelligence (as defined by Hutter and Legg), If I define intelligence as the ability to catch mice, does that mean my cat is more intelligent than most humans? More to the point, I don't understand the point of defining intelligence this way. Care to enlighten me? -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Universal intelligence test benchmark
2008/12/26 Matt Mahoney matmaho...@yahoo.com: Humans are very good at predicting sequences of symbols, e.g. the next word in a text stream. Why not have that as your problem domain, instead of text compression? Most compression tests are like defining intelligence as the ability to catch mice. They measure the ability of compressors to compress specific files. This tends to lead to hacks that are tuned to the benchmarks. For the generic intelligence test, all you know about the source is that it has a Solomonoff distribution (for a particular machine). I don't know how you could make the test any more generic. It seems to me that you and Hutter are interested in a problem domain that consists of: 1. generating random turing machines 2. running them to produce output 3. feeding the output as input to another program P, which will then guess future characters based on previous ones 4. having P use these guesses to do compression May I suggest that instead you modify this problem domain by: (a) remove clause 1 -- it's not fundamentally interesting that output comes from a turing machine. Maybe instead make output come from a program (written by humans and interesting to humans) in a normal programming language that people would actually use to write code in (b) remove clause 4 -- compression is a bit of a red herring here, what's important is to predict future output based on past output. IMO if you made these changes, your problem domain would be a more useful one. While you're at it you may want to change the size of the chunks in each item of prediction, from characters to either strings or s-expressions. Though doing so doesn't fundamentally alter the problem. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Universal intelligence test benchmark
2008/12/27 Ben Goertzel b...@goertzel.org: And this is why we should be working on AGI systems that interact with the real physical and social world, or the most accurate simulations of it we can build. Or some other domain that may have some practical use, e.g. understanding program source code. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Levels of Self-Awareness?
2008/12/24 Steve Richfield steve.richfi...@gmail.com: Clearly, it would seem that no AGI researcher can program a level of self-awareness that they themselves have not reached, tried and failed to reach, etc. This is not at all clear to me. It is certainly prossible for programmers to program computer to do tasks better than they can (e.g. play chess) and I see no reason why it shouldn't be possible for self awareness. Indeed it would be rather trivial to give an AGI access to its source code. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Relevance of SE in AGI
2008/12/21 Valentina Poletti jamwa...@gmail.com: I have a question for you AGIers.. from your experience as well as from your background, how relevant do you think software engineering is in developing AI software and, in particular AGI software? If by software engineering you mean techniques for writing software better, then software engineering is relevant to all production of software, whether for AI or anything else. AI can be thought of as a particularly hard field of software development. Just wondering.. does software verification as well as correctness proving serve any use in this field? I've never used formal proofs of correctness of software, so can't comment. I use software testing (unit tests) on pretty much all non-trivial software thast I write -- i find doing so makes things much easier. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] AGI Preschool: sketch of an evaluation framework for early stage AGI systems aimed at human-level, roughly humanlike AGI
2008/12/20 Ben Goertzel b...@goertzel.org: Well, it's completely obvious to me, based on my knowledge of virtual worlds and robotics, that building a high quality virtual world is orders of magnitude easier than making a workable humanoid robot. I guess that depends on what you mean by high quality and workable. Why does a robot have to be humanoid, BTW? I'd like a robot that can make me a cup of tea, I don't particularly care if it looks humanoid (in fact I suspect many humans would have less emotional resistance to a robot that didn't look humanoid, since it's more obviously a machine). On the other hand, making a virtual world such as I envision, is more than a spare-time project, but not more than the project of making a single high-quality video game. GTA IV cost $5 million, so we're not talking about peanuts here. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] AGI Preschool: sketch of an evaluation framework for early stage AGI systems aimed at human-level, roughly humanlike AGI
2008/12/20 Ben Goertzel b...@goertzel.org: It doesn't have to be humanoid ... but apart from rolling instead of walking, I don't see any really significant simplifications obtainable from making it non-humanoid. I can think of several. For example, you could give it lidar to measure distances with -- this could then be used as input to its vision system making it easier for the robot to tell which objects are near or far. Instead of binocular vision, it could have 2 video cameras. It could have multiple ears, which would help it tell where a sound is coming from. The the best of my knowledge, no robot that's ever been used for anything practical has ever been humanoid. Grasping and manipulating general objects with robot manipulators is very much an unsolved research problem. So is object recognition in realistic conditions. What sort of visual input do you plan to have in your virtual environment? So, to make an AGI robot preschool, one has to solve these hard research problems first. That is a viable way to go if one's not in a hurry -- but anyway in the robotics context any talk of preschools is drastically premature... On the other hand, making a virtual world such as I envision, is more than a spare-time project, but not more than the project of making a single high-quality video game. GTA IV cost $5 million, so we're not talking about peanuts here. Right, but that is way cheaper than making a high-quality humanoid robot Is it? I suspect one with tracks, two robotic arms, various sensors for light and sound, etc, could be made for less than $10,000 -- this would be something that could move around and manipulate a blocks world. My understanding is that all, or nearly all, the difficulty comes in programming it. Which is where AI comes in. Actually, $$ aside, we don't even **know how** to make a decent humanoid robot. Or, a decently functional mobile robot **of any kind** Is that because of hardware or software issues? -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] AGI Preschool: sketch of an evaluation framework for early stage AGI systems aimed at human-level, roughly humanlike AGI
2008/12/20 Derek Zahn derekz...@msn.com: Ben: Right. My intuition is that we don't need to simulate the dynamics of fluids, powders and the like in our virtual world to make it adequate for teaching AGIs humanlike, human-level AGI. But this could be wrong. I suppose it depends on what kids actually learn when making cakes, skipping rocks, and making a mess with play-dough. I think that the important cognitive abilities involved are at a simpler level than that. Consider an object, such as a sock or a book or a cat. These objects can all be recognised by young children, even though the visual input coming from trhem chasnges from what angle they're viewed at. More fundamentally, all these objects can change shape, yet humans can still effortlessly recognise them to be the same thing. And this ability doesn't stop with humans -- most (if not all) mammalian species can do it. Until an AI can do this, there's no point in trying to get it to play at making cakes, etc. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] AGI Preschool: sketch of an evaluation framework for early stage AGI systems aimed at human-level, roughly humanlike AGI
2008/12/20 Ben Goertzel b...@goertzel.org: However, I have long been perplexed at the obsession with so many AI folks with vision processing. I wouldn't say I'm obsessed with it. On its own vision processing does nothing, the same as all other input processing -- its only when a brain/AI used that processing to create output that it is actually doing any work. Theimportnat thing about vision, IMO, is not vision itself, but the way that vision interfaces with a mind's model of the world. And vision isn't really that different in principle from the other sensory modalities that a human or animal has -- they are all inputs, that go to building a model of the world, through which the organism makes decisions. But, it's not obvious to me why so many folks think vision is so critical to AI, whereas other aspects of human body function are not. I don't think any human body functions are critical to AI. IMO it's a perfectly valid approach to AI to build programs that deal with digital symbolic information -- e.g. programs like copycat or eurisko. For instance, the yogic tradition and related Eastern ideas would suggest that *breathing* and *kinesthesia* are the critical aspects of mind. Together with touch, kinesthesia is what lets a mind establish a sense of self, and of the relation between self and world. Kinesthesia/touch/movement are clearly important sensory modalities in mammals, given that they are utterly fundamental to moving around in the world. Breathing less so -- I mean you can do it if you're unconscious or brain dead. Why then is there constant talk about vision processing and so little talk about kinesthetic and tactile processing? Possibly because people are less conscious of it than vision. Personally I don't think one needs to get into any of this sensorimotor stuff too deeply to make a thinking machine. Me neither. But if the thinking machine is to be able to solve certain problems (when connected to a robot body, of course) it will have to have sophisticated systems to handle touch, movement and vision. By certain problems I mean things like making a cup of tea, or a cat climbing a tree, or a human running over uneven ground. But, if you ARE going to argue that sensorimotor aspects are critcial to humanlike AI because they're critical to human intelligence, why harp on vision to the exclusion of other things that seem clearly far more fundamental?? Say I asked you to imagine a cup. (Go on, do it now). Now, when you imagined the cup, did you imagine what it looks like, or what it feels like to the touch. For me, it was the former. So I don't think touch is clearly more fundamental, in terms of how it interacts with our internal model of the world, than vision is. Is the reason just that AI researchers spend all day staring at screens and ignoring their physical bodies and surroundings?? ;-) :-) -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] AGI Preschool: sketch of an evaluation framework for early stage AGI systems aimed at human-level, roughly humanlike AGI
2008/12/19 Ben Goertzel b...@goertzel.org: What I'd like to see is a really nicely implemented virtual world preschool for AIs ... though of course building such a thing will be a lot of work for someone... Why a virtual world preschool and not a real one? A virtual world, if not programmed accurately, may be subtly differernet from the real world, so that for example an AGI is capable of picking up and using a screwdriver in the virtual world but not real real world, because the real world is more complex. If you want your AGI to be able to use a screwdriver, you probably need to train it in the real world (at least some of the time). If you don't care whether your AGI can use a screwdriver, why have one in the virtual world? -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] AGI Preschool: sketch of an evaluation framework for early stage AGI systems aimed at human-level, roughly humanlike AGI
2008/12/20 Ben Goertzel b...@goertzel.org: I.e., I doubt one needs serious fluid dynamics in one's simulation ... I doubt one needs bodies with detailed internal musculature ... but I think one does need basic Newtonian physics and the ability to use tools, break things in half (but not necessarily realistic cracking behavior), balance things and carry them and stack them and push them together Lego-like and so forth... Needs for what purpose? I can see three uses for a virtual world: 1. to mimic the real world accurately enough that the AI can use the virtual world instead, and by using it become proficient in dealing with the real world, because it is cheaper than a real world. Obviously to program a virtual world this real is a big up-front investment, but once the investment is made, such a world may well be cheaper and easier to use than our real one. 2. to provide a useful bridge between humans and the AGI, i.e. the virtual world will be similar enough to the real world that humans will have a common frame of reference with the AGI. 3. to provide a toy domain for the AI to think about and become proficient in. (Of course there's no reason why a toy domain needs to be anything like a virtual world, it could for example be a software modality that can see/understand source code as easily and fluently as humans interprete visual input.) AIUI you're mostly thinking in terms of 2 or 3. Fair comment? -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] AGI Preschool: sketch of an evaluation framework for early stage AGI systems aimed at human-level, roughly humanlike AGI
2008/12/20 Derek Zahn derekz...@msn.com: And yet, in your paper (which I enjoyed), you emphasize the importance of not providing a simplistic environment (with the screwdriver example). Without facing the low-level sensory world (either through robotics or through very advanced simulations feeding senses essentially equivalent to those of humans), I wonder if a targeted human-like AGI will be able to acquire the necessary concepts that children absorb and use as much o f the metaphorical basis for their thought -- slippery, soft, hot, hard, rough, sharp, and on and on. Evolution has equipped humans (and other animals) have a good intuitive understanding of many of the physical realities of our world. The real world is not just slippery in the physical sense, it's slippery in the non-literal sense too. For example, I can pick up an OXO cube (a solid object), crush it so it become powder, pour it into my stew, and stir it in so it dissolves. My mind can easily and effortlessly track that in some sense its the same oxo cube and in another sense it isn't. Another example: my cat can distinguish between surfaces that are safe to sit on, and others that are too wobbly, even if they look the same. An animals intuitive physics is a complex system. I expect that in humans a lot of this machinery isd re-used to create intelligence. (It may be true, and IMO probably is true, that it's not necessary to re-create this machinery to make an AGI). -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] AGI Preschool: sketch of an evaluation framework for early stage AGI systems aimed at human-level, roughly humanlike AGI
2008/12/20 Ben Goertzel b...@goertzel.org: 3. to provide a toy domain for the AI to think about and become proficient in. Not just to become proficient in the domain, but become proficient in general humanlike cognitive processes. The point of a preschool is that it's designed to present all important adult human cognitive processes in simplified forms. So it would be able to transfer its learning to the real world and (when given a robot body) be able to go into a kitchen its never seen before and make a cup of tea? (In other words, will the simulation be deep enough to allow that). -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] AGI Preschool: sketch of an evaluation framework for early stage AGI systems aimed at human-level, roughly humanlike AGI
2008/12/20 Ben Goertzel b...@goertzel.org: Baking a cake is a harder example. An AGI trained in a virtual world could certainly follow a recipe to make a passable cake. But it would never learn to be a **really good** baker in the virtual world, unless the virtual world were fabulously realistic in its simulation (and we don't know how to make it that good, right now). Being a really good baker requires a lot of intuition for subtle physical properties of ingredients, not just following a recipe and knowing the primitive basics of naive physics... A sense of taste would probably help too. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Religious attitudes to NBIC technologies
2008/12/8 Bob Mottram [EMAIL PROTECTED]: People who are highly religious tend to be very past positive according the Zimbardo classification of people according to their temporal orientation. [...] I agree that in time we will see more polarization around a variety of technology related issues. You're probably right. Part of the problem is that these people [correctly] believe that science and technology are destroying their worldview. And as the gaps in scientific knowledge decrease, there's less roo for the God of the gaps to occupy. Having said that, I'm not aware that nanotechnology or AI are specifically prohibited by any of the major religions. And if one society forgoes science, they'll just get outcompeted by their neighbours. -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] Lamarck Lives!(?)
2008/12/3 Richard Loosemore [EMAIL PROTECTED]: http://www.newscientist.com/article/mg20026845.000-memories-may-be-stored-on-your-dna.html are saying is that memories can be stored as changes in the DNA inside neurons? No. They are saying memories might be stored as changes *on* the DNA. Imagine a big long DNA molecule. It has little molecules attached to bits of it, which regulate which genes are and aren't expressed. That's how a cell knows it's a skin cell, or an eye cell or a liver cell. Apparently the same mechanism is used in neurons are part of the mechanism for laying down new memories. Would it mean that memories (including cultural adaptations) could be passed from mother to child? No, for two reasons: (1) the DNA isn't being changed. (2) even if the DNA was being changed, it isn't in the germ-line. (Incidently, my understanding is[*] that DNA in various cells in the mammalian immune system does change as the immune system evolves to cope with infectious agents; but these changes aren't passed along to the next generation.) * if there are any molecular biologists reading, feel free to correct me. -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] Lamarck Lives!(?)
2008/12/3 Richard Loosemore [EMAIL PROTECTED]: Implication for neuroscientists proposing to build a WBE (whole brain emulation): the resolution you need may now have to include all the DNA in every neuron. Any bets on when they will have the resolution to do that? No bets here. But they are proposing that elements are added onto the DNA, not that changes are made in arbitrary locations within the DNA, so it's not /quite/ as bad as you suggest It would be pretty embarrassing for people gearing up for scans with a limiting resolution at about the size of one neuron, though. IIRC that was the rough order of magnitude assumed in the proposal I reviewed here recently. It might well be. It is anyway apparent that there are different mechanisms in the brain for laying down long-term memories and for short-term thinking over the order of a few seconds. -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: RE: FW: [agi] A paper that actually does solve the problem of consciousness
2008/12/1 Ben Goertzel [EMAIL PROTECTED]: And, science cannot tell us whether QM or some empirically-equivalent, wholly randomness-free theory is the right one... If two theories give identical predictions under all circumstances about how the real world behaves, then they are not two separate theories, they are merely rewordings of the same theory. And choosing between them is arbitrary; you may prefer one to the other because human minds can visualise it more easily, or it's easier to calculate, or you have an aethetic preference for it. -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] AIXI
That was helpful. Thanks. 2008/12/1 Matt Mahoney [EMAIL PROTECTED]: --- On Sun, 11/30/08, Philip Hunt [EMAIL PROTECTED] wrote: Can someone explain AIXI to me? AIXI models an intelligent agent interacting with an environment as a pair of interacting Turing machines. At each step, the agent outputs a symbol to the environment, and the environment outputs a symbol and a numeric reward signal to the agent. The goal of the agent is to maximize the accumulated reward. Hutter proved that the optimal solution is for the agent to guess, at each step, that the environment is simulated by the shortest program that is consistent with the interaction observed so far. Hutter also proved that the optimal solution is not computable because the agent can't know which of its guesses are halting Turing machines. The best it can do is pick numbers L and T, try all 2^L programs up to length L for T steps each in order of increasing length, and guess the first one that is consistent. If there are no matches, then it needs to choose larger L and T and try again. That solution is called AIXI^TL. It's time complexity is O(T 2^L). In general, it may require L up to the length of the observed interaction (because there is a fast program that outputs the agent's observations from a list of length L). In a separate paper ( http://www.vetta.org/documents/ui_benelearn.pdf ), Legg and Hutter propose defining universal intelligence as the expected reward of an AIXI agent in random environments. The value of AIXI is not that it solves the general intelligence problem, but rather it explains why the problem is so hard. It also justifies a general principle that is already used in science and in practical machine learning algorithms: to choose the simplest hypothesis that fits the data. It formally defines simple as the length of the shortest program that outputs a description of the hypothesis. For example, to avoid overfitting in neural networks, you should use the smallest number of connections and the least amount of training needed to fit the training data, then stop. In this case, the complexity of your neural network is the length of the shortest program that outputs the configuration of your network and its weights. Even if you don't know what that program is, and haven't chosen a programming language, you may reasonably expect that fewer connections, smaller weights, and coarser weight quantization will result in a shorter program. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] Mushed Up Decision Processes
2008/11/30 Ben Goertzel [EMAIL PROTECTED]: Could you give me a little more detail about your thoughts on this? Do you think the problem of increasing uncomputableness of complicated complexity is the common thread found in all of the interesting, useful but unscalable methods of AI? Jim Bromer Well, I think that dealing with combinatorial explosions is, in general, the great unsolved problem of AI. I think the opencog prime design can solve it, but this isn't proved yet... Good luck with that! In general, the standard AI methods can't handle pattern recognition problems requiring finding complex interdependencies among multiple variables that are obscured among scads of other variables The human mind seems to do this via building up intuition via drawing analogies among multiple problems it confronts during its history. Yes, so that people learn one problem, then it helps them to learn other similar ones. Is there any AI software that does this? I'm not aware of any. I have proposed a problem domain called function predictor whose purpose is to allow an AI to learn across problem sub-domains, carrying its learning from one domain to another. (See http://www.includipedia.com/wiki/User:Cabalamat/Function_predictor ) I also think it would be useful if there was a regular (maybe annual) competition in the function predictor domain (or some similar domain). A bit like the Loebner Prize, except that it would be more useful to the advancement of AI, since the Loebner prize is silly. -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
[agi] AIXI (was: Mushed Up Decision Processes)
2008/11/29 Matt Mahoney [EMAIL PROTECTED]: The general problem of detecting overfitting is not computable. The principle according to Occam's Razor, formalized and proven by Hutter's AIXI model, is to choose the shortest program (simplest hypothesis) that generates the data. Overfitting is the case of choosing a program that is too large. Can someone explain AIXI to me? My understanding is that you've got some black-box process emitting output, and you generate all possible programs that emit the same output, then choose the shortest one. You then run this program and its subsequent output is what you predict the black-box process will do. This has the minor drawback, of course, that it requires infinite processing power and is therefore slightly impractical. I've read Hutter's paper Universal algorithmic intelligence, A mathematical top-down approach which amusingly describes itself as a gentle introduction to the AIXI model. Hutter also describes AIXItl of computation time Ord(t*2^L) where I assume L is the length of the program and I'm not sure what t is. Is AIXItl something that could be practically written or is it purely a theoretical construct? In short, is there something to AIXI or is it something I can safely ignore? -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] Mushed Up Decision Processes
2008/11/30 Ben Goertzel [EMAIL PROTECTED]: Hi, I have proposed a problem domain called function predictor whose purpose is to allow an AI to learn across problem sub-domains, carrying its learning from one domain to another. (See http://www.includipedia.com/wiki/User:Cabalamat/Function_predictor ) I also think it would be useful if there was a regular (maybe annual) competition in the function predictor domain (or some similar domain). A bit like the Loebner Prize, except that it would be more useful to the advancement of AI, since the Loebner prize is silly. -- Philip Hunt, [EMAIL PROTECTED] How does that differ from what is generally called transfer learning ? I don't think it does differ. (Transfer learning is not a term I'd previously come across). -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] DARPA funds using memsistors to model synapses in neuromorphic computing
2008/11/27 Matt Mahoney [EMAIL PROTECTED]: This is probably not a serious problem for neural networks because the connections could be written in parallel. It's actually much faster than the write times in the human brain, probably 10^4 seconds in the hippocampus and 10^8 seconds in the cortex. 10^8 seconds is 3 years! I think that number's wrong. -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com