RE: [agi] A point of philosophy, rather than engineering
Arthur Murray wrote: If Ben Goertzel and the rest of the Novamente team build up an AI that mathematically comprehends mountains of data, they may miss the AI boat by not creating persistent concepts that accrete and auto-prune over time as the basis of NLP. No, even before the Novamente system understands natural language, it will still have persistent concepts accreting and auto-pruning over time. In fact, we're right now doing some primitive testing of accretion auto-pruning processes. Accretion auto-pruning are part of a chimp's mind-brain, they're prior to language... The Mentifex Mind-1.1 AI, primitive as it may be, has since 1993 (nine years ago) gradually built up a sophisticated group of about 34 mind-modules now barely beginning to achieve NLP results. I enter these thoughts here not confrontationally but from a point-of-view that NLP is not otherwise being sufficiently represented among all these mathematicians and computationalists. NLP is obviously very important. Historically, it has often been associated with an overly rigid rule-based approach to AI, which is perhaps why it's not so fashionable among AGI people. I agree that amenability-to-NLP should be an important consideration of any AGI design process. We've designed Novamente specifically so that it will be able to learn language when the time comes. Our experience working with computational linguistics at Webmind allowed us to do that. On the other hand, Peter Voss has often put forth the following view (this is a paraphrase, not a quote): Our brains are a lot like chimps' brains. If someone designed an AGI with chimp level intelligence, making the modifications to turn this chimp-AGI into a human-level AGI with linguistic ability would be a *relatively* small trick compared to the original trick of designing the chimp-AGI. I think Peter has a certain point, but it's not the approach I'm taking. My approach is more linguistic than his but less than yours. Your approach puts linguistics at the center, it seems; but I don't think you can FOUND an AGI system on linguistics. I think linguistic ability has to mostly emerge from more generic cognitive functionality, in order to be full and genuine linguistic ability that involves deep semantic understanding -- Ben G --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/
Re: [agi] A point of philosophy, rather than engineering
Ben Goertzel wrote: Hi, Personally, I believe that the most effective AI will have a core general intelligence, that may be rather primitive, and a huge number of specialized intelligence modules. The tricky part of this architecture is designing the various modules so that they can communicate. It isn't clear that this is always reasonable (consider the interfaces between chess and cooking), but if the problem can be handled in a general manner (there's that word again!), then one of the intelligences could be specialized for message passing. In this model the core general intelligence will be for use when none of the hueristics fit the problem. And it's attempts will be watched by another module whose specialty is generating new hueristics. Plausible? I don't really know. Possibly to complicated to actually build. It might need to be evolved from some simpler precursor. It's clear that the human brain does something like what you're suggesting. Much of the brain is specialized for things like vision, motion control, linguistic analysis, time perception, etc. etc. The portion of the human brain devoted to general abstract thinking is very small. Novamente is based on an integrative approach sorta like you suggest. But it's not quite as rigidly modular as you suggest. Rather, we think one needs to -- create a flexible knowledge representation (KR) useful for representing all forms of knowledge (declarative, procedural, perceptual, abstract, linguistic, explicit, implicit, etc. etc.) This probably won't work. Thinking of the brain as a model, we have something called the synesthetic gearbox which is used to relate information in one modality of senstation with another modality. This is a part of the reason that I suggested that one of the hueristic modules be specialized for message passing (and translation). -- create a number of specialized mind agents acting on the KR, carrying out specialized forms of intelligent processes -- create an appropriate set of integrative mind agents acting on the KR, oriented toward creating general intelligence based largely on the activity specialized mindagents Again the term general intelligence. I would like to suggest that the intelligence needed to repair an auto engine is different from that needed to solve a calculus equation. I see the General Intelligence as being the primarily to handle problems for which no hueristic can be found, and would suggest that nearly any even slightly tuned hueristic is better than the general intellligence for almost all problems. E.g., if one is repairing an auto engine, one hueristic would be to remember the shapes of all the pieces you have seen, and to remember where they were when you first saw them. Just think how that one hueristic would assist reassembling the engine. Set up a knowledge base involving all these mind agents.. hook it up to sensors actuators give it a basic goal relating to its environment... Of course, this general framework and 89 cents will get you a McDonald's Junior Burger. All the work is in designing and implementing the KR and the MindAgents!! That's what we've spent (and are spending) all our time on... May I suggest that if you are even close to what you are attempting, that you have the start of a dandy personal secretary. With so much correspondence coming via e-mail these days, this would create a very simplified environment in which the entity would need to operate. In this limited environment you wouldn't need full meanings for most words, only categories and valuations. I have a project which I am aiming at that area, but it is barely getting started. -- Ben --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/ -- -- Charles Hixson Gnu software that is free, The best is yet to be. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/
RE: [agi] A point of philosophy, rather than engineering
Charles Hixson wrote (in response to me): -- create a flexible knowledge representation (KR) useful for representing all forms of knowledge (declarative, procedural, perceptual, abstract, linguistic, explicit, implicit, etc. etc.) This probably won't work. Thinking of the brain as a model, we have something called the synesthetic gearbox which is used to relate information in one modality of senstation with another modality. This is a part of the reason that I suggested that one of the hueristic modules be specialized for message passing (and translation). There are both significant differences, and significant similarities, between the representations used by different parts of the human brain. They all use neurons and synapses, frequencies of neural firing, neurotransmitter chemistry, etc., in fairly similar ways. Of course there are also some major differences in neural architecture btw brain regions -- different types of neurons, different neurotransmitter concentrations, different connective arrangementc, etc. Similarly there are significant similarities differences btw the representations used by different parts of Novamente. They all use Novamente Nodes and Links, and all use similar quantitative parameters of Nodes and Links, and there's a lot of overlap in the MindAgents (dynamical processes) they use. But there are also significant differences, in the frequency of different node and link types, the parameters of the different MindAgents, etc. Again the term general intelligence. I would like to suggest that the intelligence needed to repair an auto engine is different from that needed to solve a calculus equation. Of course it is different in many ways. It's also similar in many ways. I believe that those two forms of intelligence consist of basically the same set of processes, acting on the same basic sort of knowledge. But the two cases have very different underlying parameter settings. In the brain case, different types of neural connectivity patterns, perhaps different concentrations of neurotransmitters in different brain regions, perhaps even different amounts of different types of neurons -- all of which leads to different emergent structures/dynamics. I see the General Intelligence as being the primarily to handle problems for which no hueristic can be found, and would suggest that nearly any even slightly tuned hueristic is better than the general intellligence for almost all problems. E.g., if one is repairing an auto engine, one hueristic would be to remember the shapes of all the pieces you have seen, and to remember where they were when you first saw them. Just think how that one hueristic would assist reassembling the engine. Yes, but what allows a human mind to learn that heuristic? Our general (reasonably general, but far from absolutely general) intelligence. Set up a knowledge base involving all these mind agents.. hook it up to sensors actuators give it a basic goal relating to its environment... Of course, this general framework and 89 cents will get you a McDonald's Junior Burger. All the work is in designing and implementing the KR and the MindAgents!! That's what we've spent (and are spending) all our time on... May I suggest that if you are even close to what you are attempting, that you have the start of a dandy personal secretary. With so much correspondence coming via e-mail these days, this would create a very simplified environment in which the entity would need to operate. In this limited environment you wouldn't need full meanings for most words, only categories and valuations. As I said in a recent post, I prefer to stay away from natural language processing at this stage, until the system has acquired a rudimentary understanding of natural language thru its own experience. We're not quite there yet ;) ben --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/
RE: [agi] A point of philosophy, rather than engineering
On Tue, 12 Nov 2002, Ben Goertzel wrote: Charles Hixson wrote (in response to me): [...] May I suggest that if you are even close to what you are attempting, that you have the start of a dandy personal secretary. With so much correspondence coming via e-mail these days, this would create a very simplified environment in which the entity would need to operate. In this limited environment you wouldn't need full meanings for most words, only categories and valuations. BenG: As I said in a recent post, I prefer to stay away from natural language processing at this stage, until the system has acquired a rudimentary understanding of natural language thru its own experience. We're not quite there yet ;) That's where the Mentifex AI and Novamente differ (and probably also where A.T. Murray the linguist and Ben Goertzel the mathematician differ). If you're not aiming for language, you're aiming for a smart animal. A.T. Murray -- http://www.scn.org/~mentifex/aisource.html is the cluster of Mind programs described in the AI textbook AI4U based on AI Mind-1.1 by Arthur T. Murray which may be pre-ordered from bookstores with hardcover ISBN 0-595-65437-1 and ODP softcover ISBN 0-595-25922-7. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/
Re: [agi] A point of philosophy, rather than engineering
The problem with a truly general intelligence is that the search spaces are too large. So one uses specializing hueristics to cut down the amount of search space. This does, however, inevitably remove a piece of the generality. The benefit is that you can answer more complicated questions quickly enough to be useful. I don't see any way around this, short of quantum computers, and I'm not sure about them (I have this vague suspicion that there will be exponentially increasing probabilities of error, which require hugely increased error recovery systems, etc.). This doesn't mean that we have currently reached the limits of agi. It means that whatever those limits are, there will always be hueristicly tuned intelligences that will be more efficient in most problem domains. Of course, here I am taking a strict interpretation of general, as in General Relativity vs. Special Relativity. Notice that while Special Relativity has many uses, General Relativity is (or at least was until quite recently) mainly of theoretical interest. Be prepared for a similar result with General Intelligence vs. Special Intelligence. (The difference here is that Special Intelligence comes in lots of modules adapted for lots of special circumstances.) Personally, I believe that the most effective AI will have a core general intelligence, that may be rather primitive, and a huge number of specialized intelligence modules. The tricky part of this architecture is designing the various modules so that they can communicate. It isn't clear that this is always reasonable (consider the interfaces between chess and cooking), but if the problem can be handled in a general manner (there's that word again!), then one of the intelligences could be specialized for message passing. In this model the core general intelligence will be for use when none of the hueristics fit the problem. And it's attempts will be watched by another module whose specialty is generating new hueristics. Plausible? I don't really know. Possibly to complicated to actually build. It might need to be evolved from some simpler precursor. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/
Re: [agi] A point of philosophy, rather than engineering
On Mon, 2002-11-11 at 14:11, Charles Hixson wrote: Personally, I believe that the most effective AI will have a core general intelligence, that may be rather primitive, and a huge number of specialized intelligence modules. The tricky part of this architecture is designing the various modules so that they can communicate. It isn't clear that this is always reasonable (consider the interfaces between chess and cooking), but if the problem can be handled in a general manner (there's that word again!), then one of the intelligences could be specialized for message passing. In this model the core general intelligence will be for use when none of the hueristics fit the problem. And it's attempts will be watched by another module whose specialty is generating new hueristics. This is essentially what we do, but it works a little differently than you are suggesting. The machinery and representation underneath the modules is identical, where each module is its own machine which has become optimized for its task. In other words, if you were making a module on chess and a module on cooking you would start with the same blank module machinery and they would be trained for their respective tasks. If you looked at the internals of the module machinery after the training period, you would notice marked macro-level structural differences between the two that relate to how the machinery self-optimizes for its task. The computational machines, which are really just generic Turing virtual machines that you could program any type of software on, use a pretty foreign notion of computation/processing -- the processor model looks nothing like a von Neumann-variant architecture. Despite notable differences in structure, it is really just two modules of the same machine that have automatically conformed structurally to their data environment. The interesting part is the integration of the modules. There are actually a number of ways to do it, all of which have advantages and disadvantages. One advantage of having simple underlying machinery controlling the representation of data is that all modules already deeply understand the data of any other module. You COULD do a hard merge of the cooking module with the chess module into one module, and automatically discover the relations and abstract similarities between the two (whatever those might be) without any special code, but there are lots of reasons why this is bad in practice. In implementation, we typically do what we would call a soft merge, where the machines are fully integrated for most purposes and can use each others space, but where external data feeds are localized to specific modules within the cluster (even though these modules have access to every other module for the purposes of processing the data feed). From the perspective of external data streams it looks like a bunch of independent machines working together, but from the perspective of the machine the entire cluster is a single machine image. There are good theoretical reasons for doing things this way which I won't go into here. In short, we mostly do what you are talking about, but you've actually over-estimated the difficulty of integration of domain-specific modules (using our architecture, at least). Actually building modules is more difficult, mostly because the computing architecture uses assumptions that are very strange; I think my programmer's mind works against me some days, and teaching/training modules by example is easier than programming them directly most times. Once they are done, you pretty much can do plug-n-play on-the-fly integration, even on a hot/active cluster of modules (resource permitting, of course). An analogy would be how they learned new skills on-the-fly in The Matrix. The integration is a freebie that comes with the underlying architecture, not something that I spent much effort designing. Cheers, -James Rogers [EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/
