Re: [agi] The Test
Mike Tintner wrote: Richard: Consider yourself corrected: many people realize the importance of generalization (and related processes). People go about it in very different ways, so some are more specific and up-front about it than others, but even the conventional-AI people (with whom I have many disagreements on other matters) realize the importance of it, and are trying to do something about it. As for what "AGI systembuilders" are doing, you can take it from me that my own system is deeply rooted in the concept of generalization. Richard, We have another of your misreadings in haste here - something of a Q.E.D. misreading. I can do no better than requote my opening lines (please read carefully) : "There's a simple misreading here. No way am I saying nobody is looking at the problem! I am saying nobody is offering a solution! And none of the AGI systembuilders present or past have *acknowledged* that they haven't offered a solution - otherwise they wouldn't have made such large claims. And I am not aware of anyone even offering an equivalent of the General Test I just offered." Like nearly everyone else, you are indeed looking at the problem, and even claiming again a solution, but have so far actually offered bupkes :) - cetainly in relation to the generalization problem/ test. And until you do, it remains to be seen whether you are even actually addressing the problem. I am suggesting - and I shall be delighted to eat my words - that this is the central one of what Mark Waser identified as the unacknowledged, "a-miracle-will-happen-here" holes in not just Ben's but everyone's project plans. And indeed it is also the central reason why as Wozniak, pace Storrs Hall, more or less identified - computers & AGI's and, to some extent, robots are "Tommy's" (and then some) - deaf, dumb and blind quadriplegics, who while they may be extraordinary autistic savants, are still unable to deal with the real world. Perhaps better to wait for my next post before replying. I'm sorry, but I think this argument is losing coherence. If you are complaining that no-one has solved the problem of generalization then you are (to coin a phrase) saying bupkes :-). According to that way of thinking, nobody has 'solved" anything until Delivery Day. If, on the other hand, you are saying that someone has a part of their plans that belongs in the "a-miracle-will-happen-here" category (an dyou do indeed say this, no?), then you are saying that that person is ignoring it, trying to pretend they don't need it, not aware of the fact that it is missing-but-crucial, etc etc. In a nutshell, they are not working on it, and they should be. Those two types of critique are not the same. Once again I am deeply confused about what you are criticising. Perhaps the fault is mine, but when I read what you write, I get the feeling that the left hand paragraphs knoweth not what the right hand paragraphs sayeth... Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93139505-4aa549
Re: [agi] The Test
Richard: Consider yourself corrected: many people realize the importance of generalization (and related processes). People go about it in very different ways, so some are more specific and up-front about it than others, but even the conventional-AI people (with whom I have many disagreements on other matters) realize the importance of it, and are trying to do something about it. As for what "AGI systembuilders" are doing, you can take it from me that my own system is deeply rooted in the concept of generalization. Richard, We have another of your misreadings in haste here - something of a Q.E.D. misreading. I can do no better than requote my opening lines (please read carefully) : "There's a simple misreading here. No way am I saying nobody is looking at the problem! I am saying nobody is offering a solution! And none of the AGI systembuilders present or past have *acknowledged* that they haven't offered a solution - otherwise they wouldn't have made such large claims. And I am not aware of anyone even offering an equivalent of the General Test I just offered." Like nearly everyone else, you are indeed looking at the problem, and even claiming again a solution, but have so far actually offered bupkes :) - cetainly in relation to the generalization problem/ test. And until you do, it remains to be seen whether you are even actually addressing the problem. I am suggesting - and I shall be delighted to eat my words - that this is the central one of what Mark Waser identified as the unacknowledged, "a-miracle-will-happen-here" holes in not just Ben's but everyone's project plans. And indeed it is also the central reason why as Wozniak, pace Storrs Hall, more or less identified - computers & AGI's and, to some extent, robots are "Tommy's" (and then some) - deaf, dumb and blind quadriplegics, who while they may be extraordinary autistic savants, are still unable to deal with the real world. Perhaps better to wait for my next post before replying. Mike Tintner wrote: Benjamin:When I read your post, claiming that generalization is important, I think to myself "yeah, that is what everybody else is saying and attempting to solve -- I even gave you several examples of how generalization could work", so I then find myself surprised that you claim that nobody is looking at it! Quick response for now. There's a simple misreading here. No way am I saying nobody is looking at the problem! I am saying nobody is offering a solution! And none of the AGI systembuilders present or past have *acknowledged* that they haven't offered a solution - otherwise they wouldn't have made such large claims. And I am not aware of anyone even offering an equivalent of the General Test I just offered. Yeah, it's an incredibly obvious test - almost a redefinition (although just a little more, too) of "Artificial GENERAL Intelligence." But you'd be amazed how often people ignore the obvious. Look also at how strenuously Ben objected when I suggested that his definition of intelligence as "achieving *complex* goals in complex environments" should be replaced by one focussing on the *general* aspect (for AGI), which is what he really seemed to mean in one passage, (although in another text of his, the general aspect simply gets lost). I do believe though - and I stand to be corrected - that nobody has fully identified the central importance of this problem - i.e. I agree with Joseph Gentle's: "I think making a representation of the world which can be generalised and abstracted is the emergent crux of AGI". Consider yourself corrected: many people realize the importance of generalization (and related processes). People go about it in very different ways, so some are more specific and up-front about it than others, but even the conventional-AI people (with whom I have many disagreements on other matters) realize the importance of it, and are trying to do something about it. As for what "AGI systembuilders" are doing, you can take it from me that my own system is deeply rooted in the concept of generalization. Richard Loosemore Yes, it's still *emerging* AFAIK. If you want to correct me here, though, you'll have to quote some literature. Yes, I'm developing & will set out a much larger argument here - later today/tomorrow. When I do, I think you'll see why people are, however subtly, avoiding the problem, .[BTW I will want to attach a photo file - can one do that?] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?&; -- No virus found in this incoming message. Checked by AVG Free Edition. Version: 7.5.516 / Virus Database: 269.19.21/1265 - Release Date: 2/7/2008 11:17 AM - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=9
Re: [agi] The Test
Mike Tintner wrote: Benjamin:When I read your post, claiming that generalization is important, I think to myself "yeah, that is what everybody else is saying and attempting to solve -- I even gave you several examples of how generalization could work", so I then find myself surprised that you claim that nobody is looking at it! Quick response for now. There's a simple misreading here. No way am I saying nobody is looking at the problem! I am saying nobody is offering a solution! And none of the AGI systembuilders present or past have *acknowledged* that they haven't offered a solution - otherwise they wouldn't have made such large claims. And I am not aware of anyone even offering an equivalent of the General Test I just offered. Yeah, it's an incredibly obvious test - almost a redefinition (although just a little more, too) of "Artificial GENERAL Intelligence." But you'd be amazed how often people ignore the obvious. Look also at how strenuously Ben objected when I suggested that his definition of intelligence as "achieving *complex* goals in complex environments" should be replaced by one focussing on the *general* aspect (for AGI), which is what he really seemed to mean in one passage, (although in another text of his, the general aspect simply gets lost). I do believe though - and I stand to be corrected - that nobody has fully identified the central importance of this problem - i.e. I agree with Joseph Gentle's: "I think making a representation of the world which can be generalised and abstracted is the emergent crux of AGI". Consider yourself corrected: many people realize the importance of generalization (and related processes). People go about it in very different ways, so some are more specific and up-front about it than others, but even the conventional-AI people (with whom I have many disagreements on other matters) realize the importance of it, and are trying to do something about it. As for what "AGI systembuilders" are doing, you can take it from me that my own system is deeply rooted in the concept of generalization. Richard Loosemore Yes, it's still *emerging* AFAIK. If you want to correct me here, though, you'll have to quote some literature. Yes, I'm developing & will set out a much larger argument here - later today/tomorrow. When I do, I think you'll see why people are, however subtly, avoiding the problem, .[BTW I will want to attach a photo file - can one do that?] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93139505-4aa549
Re: [agi] The Test
Benjamin:When I read your post, claiming that generalization is important, I think to myself "yeah, that is what everybody else is saying and attempting to solve -- I even gave you several examples of how generalization could work", so I then find myself surprised that you claim that nobody is looking at it! Quick response for now. There's a simple misreading here. No way am I saying nobody is looking at the problem! I am saying nobody is offering a solution! And none of the AGI systembuilders present or past have *acknowledged* that they haven't offered a solution - otherwise they wouldn't have made such large claims. And I am not aware of anyone even offering an equivalent of the General Test I just offered. Yeah, it's an incredibly obvious test - almost a redefinition (although just a little more, too) of "Artificial GENERAL Intelligence." But you'd be amazed how often people ignore the obvious. Look also at how strenuously Ben objected when I suggested that his definition of intelligence as "achieving *complex* goals in complex environments" should be replaced by one focussing on the *general* aspect (for AGI), which is what he really seemed to mean in one passage, (although in another text of his, the general aspect simply gets lost). I do believe though - and I stand to be corrected - that nobody has fully identified the central importance of this problem - i.e. I agree with Joseph Gentle's: "I think making a representation of the world which can be generalised and abstracted is the emergent crux of AGI". Yes, it's still *emerging* AFAIK. If you want to correct me here, though, you'll have to quote some literature. Yes, I'm developing & will set out a much larger argument here - later today/tomorrow. When I do, I think you'll see why people are, however subtly, avoiding the problem, .[BTW I will want to attach a photo file - can one do that?] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93139505-4aa549
Re: [agi] The Test
On Feb 8, 2008 7:12 AM, Benjamin Johnston <[EMAIL PROTECTED]> wrote: > > 4. If you're trying to develop your own argument, then I'd recommend > taking a look at some of the more philosophical works in the research > literature - not just in AGI but also in areas like embodied robotics, > commonsense reasoning, cognitive science, qualitative reasoning and > cognitive robotics. I personally found that writings on the symbol > grounding problem were very helpful in clarifying a lot of my own > thoughts (and in understanding how my own opinion relates to established > positions). I'm sure there's something out there that would do the same > for you, whether it be in the grounding problem (like me) or something > completely different. Ben, Could you say a couple of words about specifics of what you found helpful and in which writings? There are plenty, and I mainly found them long-winded and unhelpful (although significant part of what I've developed so far can be charachterised as 'philosophy' of how to build an AGI, and it would have greatly helped if I could just read it). Closest thing to inspiration-generating writings that I found is basic cognitive science, and Hofstadter. -- Vladimir Nesovmailto:[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93139505-4aa549
Re: [agi] The Test
Thankyou for another really constructive response - and I think that I, at any rate, am really starting to get somewhere. I didn't quite get to the nub of things with my last post. I think I can do a better job this time & develop the argument still more fully later. Hi Mike, I have five comments. 1. You seem to be using a more specific definition of AGI than I. I don't believe that all AGI work must necessarily focus on real-world embodiment. Don't you think it is possible to have an artificial general intelligence (such as an AGI info-bot) that inhabits a virtual symbolic world (such as a database); a world in which initial classification of objects is irrelevant to the agent? I think AGI can encompass a range of different kinds of intelligences that inhabit not just real world environments, but also virtual environments, language-based environments or even purely formal symbolic environments. Some approaches might be better suited to particular environments. 2. I don't believe it it right to say that nobody is looking at generalization. I illustrated how generalization might be achieved automatically by a mutation operator in a GA biased towards generalization (for all instances of a given symbol, substitute it with a more general symbol), or how GA might be used to automatically acquire categorizations of abstract concepts from raw sensory features. Generalization lies at the very core of machine learning and AGI and there are plenty of formal and informal attempts to describe it. 3. It certainly is my own experience that I got into this area because I was intrigued by feelings that true intelligence is different from classical logical deduction or the standard kinds of machine learning algorithms. I suspect that most people here have felt (and still do) the same way, and it looks like you feel that way too. When I look at various approaches, if I focus on the similarities instead of the differences, it strikes me that we're all attempting to attack the same deep issue from different perspectives. When I read your post, claiming that generalization is important, I think to myself "yeah, that is what everybody else is saying and attempting to solve -- I even gave you several examples of how generalization could work", so I then find myself surprised that you claim that nobody is looking at it! I'll illustrate my point with fuzzy/uncertain logics, because you directly attacked them in a previous post... My own initial reaction to modified logics that support "fuzzy" propositions was also that it didn't match my intuitions of how intelligence works - that they're "not even looking at the same fundamental problem". But, if - as you also say - the problem is that formal methods can't be used until after "you've classified the real world and predigested its irregularities into nice neat regular units", then I realize that maybe this fuzzy approach really does make sense: they're trying to use the universality of logic but they're also trying to skip over the need for "nice neat regular units" by letting the logic natively accept "ugly messy irregular units". You might not buy this particular reasoning and so you may need to find one of your own that maps their objectives to your own view of the challenge of intelligence; but I think you will find that with an open mind, you really will start seeing that there are connections to your own ideas. That is, everybody does have some kind of grasp on the same fundamental problem, but they're just looking at it from different angles. When you start to formulate your ideas into a coherent argument (that doesn't use vague words like "structured" without definition), you might then start forming your own ideas of how to approach the problem. Hypothetically... you might reason that generalization is fundamental, so you could (again, hypothetically) start off by experimenting with a translation of this abstract idea into a concrete computational model where self-modifying programs can take their own subroutines and automatically search for generalizations of those subroutines (and maybe you also have another process of hierarchical learning to discover "x is a generalization of y" patterns). At that point you'll have your own AGI system building program, at then maybe you'll come across somebody else who sees what you are doing, who ignores the background work that got you there and your long term vision of where you want to go with it, but simply claims "hey, no, intelligence isn't self-modifying subroutine abstraction, duh! why don't you come up with a crux idea?". 4. If you're trying to develop your own argument, then I'd recommend taking a look at some of the more philosophical works in the research literature - not just in AGI but also in areas like embodied robotics, commonsense reasoning, cognitive science, qualitative reasoning
Re: [agi] The Test
On Feb 7, 2008 11:53 AM, Mike Tintner <[EMAIL PROTECTED]> wrote: > And I think it's clear, if only in a very broad way, how the human mind > achieves this (which I'll expound in more detail another time) - it has what > you could call a "general activity language" - and learns every skill *as an > example of a general activity*. Every particular skill is learned in a > broad, general way in terms of concepts that can be and are applied to all > skills/ activities (as well as more skill-specific terminology). Very > general, "modular" concepts. My approach to the whole problem of AGI is to think "What is the hard bit, that noone has really gotten right at all?". I completely agree with the problem you pose here - I think making a representation of the world which can be generalised and abstracted is the emergent crux of AGI. Neural networks and other machine learning methods like decision trees don't have representations which support the kind of operation we're talking about. How can we solve this? I think it requires making a really simple associative language of sorts. We need something in which I can trivially represent things like: "A is somehow related to B" "A and B have some common properties. I will call these common properties P and make A and B specialisations of P" "C is similar to A and B somehow. C might have the properties of P." If the representation just makes a graph of links between things (/objects/concepts/cortical columns) then finding the common links between two objects doesn't actually seem that hard a problem. I think. -J - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=94603346-a08d2f
Re: [agi] The Test
Benjamin, Thankyou for another really constructive response - and I think that I, at any rate, am really starting to get somewhere. I didn't quite get to the nub of things with my last post. I think I can do a better job this time & develop the argument still more fully later. Why are those ideas not crux ideas - those schools of programming not true AGI? You almost hit the nail on the head with: "My point is, however, that general purpose reasoning is possible - I think there are plenty of signs of how it might actually work." i.e. none of those approaches actually show true "general purpose reasoning," you only hope and believe that some new ones will in the future (and have some good suggestions about how). What all those schools lack, to be a bit more precise, is an explicit "generalizing procedure" - let's call it a "real world generalization procedure." They don't tell you directly how they are going to generalize across domains - how, having learnt one skill, they can move on to another. The GA's, if I've understood, didn't generalize their skills - didn't recognize that they could adapt their walking skills to water - their minders did. An explicit real-world generalization procedure must tell you how the system itself is going to recognize an unfamiliar domain as related to the familiar one(s). How the lego construction system will recognize irregular-shaped rocks as belonging to a larger class that includes the lego bricks. How Ben's pet who, say, knows all about navigating neat, flat office buildings will be able to recognize a very different, messy bomb site or rocky hillside as nevertheless all examples of navigable.terrains. How a football playing robot will recognize other games such as rugby, hockey etc as examples of "ball games [I may be able to play]". How in other words the AGI system will recognize unfamiliar (and not obviously classifiable) problems as having something in common with familiar ones. And how those systems will have general ways of adapting their skills/ solutions. How the lego system will adapt its bricklaying movements to form rocklaying movements, or the soccer player will adapt its arm and leg movements to rugby. I think you'll find that all the schools of programming only wave at this... they don't offer you an explicit method. I'll take a bet, for example, that Ben G cannot provide you with even a virtual world generalization procedure. The AGI systems/agents, it must be stressed, have to be able to recognize *independently* that they can move on to new domains - even though they will of course also need to seek help to learn the rules etc, as we humans do. And I think it's clear, if only in a very broad way, how the human mind achieves this (which I'll expound in more detail another time) - it has what you could call a "general activity language" - and learns every skill *as an example of a general activity*. Every particular skill is learned in a broad, general way in terms of concepts that can be and are applied to all skills/ activities (as well as more skill-specific terminology). Very general, "modular" concepts. But such human powers of generalization are still way, way beyond current computers. The human mind's ability to cross domains is dependent on the ability, for example, to generalize from something as concrete as "taking steps across a field" to something as abstract as "taking steps to solve a problem in philosophy or formal logic". And the reason that I classify all this as *real world* generalization is that it cannot be achieved by logic or mathematics, which is what all the schools you mention depend on, (no?) They can't help you classify the bricks and rocks as alike, or rugby as like football, or a rocky bomb site as like an office floor, let alone steps across a field as like steps in an argument. They can only be brought into play *after* you've classified the real world and predigested its irregularities into nice neat regular units that they can operate on. That initial classification/ generalization requires the general skill that is still beyond all AGI's - and actually, I think, doesn't even have a name. bENJAMIN: mt:>> I think your approach here *is* representative - &, as you indicate, the details of different approaches to AGI in this discussion, aren't that important. What is common IMO to your and the thinking of others here is that you all start by asking yourselves : what kinds of programming will solve AGI? Because programming is what interests you most and is your life. Actually, that isn't necessarily accurate. I'm currently collaborating with a cognitive scientist, and I've seen other people here hint at drawing their own inspiration from cognitive science and other non-programming disciplines. I reason the problem like this: 1. I know intelligence is possible, by looking at the animal kingdom. 2. I don't believe that the animal k
Re: [agi] The Test
Benjamin Johnston wrote: Very briefly, my focus a while back in attacking programs was not on the sign/ semiotic - and more particularly, symbolic - form of programs, although that is v. important too. My focus was on the *structure* of programs - that's what they are: structured and usually sequenced sets of instructions.No matter how sophisticated their structure, and/or their capacity to adapt their structure, they are still structured. I'm unclear what you mean by structure. Interpretaton 1: - Every program in a modern computer language is a structured and sequenced set of instructions. It isn't possible to write an unsequenced set of instructions, because the language itself imposes that structure. If structured programs cannot be intelligent, then if I understand you correctly, it follows that what you are saying is that it is *impossible* to write intelligent systems in modern computer programming languages. Given that modern computer languages are Turing complete (modulo space and time limitations), your claims would therefore be equivalent to saying that intelligence is not computable. Interpretation 2: - May be you mean something a little stronger by structure? That the way that human beings engineer software is very structured, and software that has been engineered by humans with that kind of structure cannot possibly solve unstructured problems. Do you think, then, that it is possible for a human to write a structured program that generates unstructured programs that have general intelligence? Ben, I feel compelled to help out here, because (as I said in my post to Mike), he is using words in a way that causes confusion ... and since Mike and I have had the same conversation/debate at least twice before, it might help if I explain what I have already understood from those previous conversations. The key thing s that he does not mean "structured" in any of the senses that most others would use the term. What Mike is trying to say is that he has great objections to the style of Artificial Intelligence system in which the intelligence process is supposed to be very narrowly rule-governed, with simple symbols (no internal structure to the symbols) and very deterministic processing. Unfortunately, he often uses the word "program" to describe this, although he has now also called it "structured". I would tend to call that approach to AI something like "simple, logical symbol-processing", or some such term. Other people would make the same distinction between different types of AI, but use different language. What Mike is demanding is that people recognize the limitations of that style of AI, and move to something that allows for fluidity, creativity, unpredictability (non-deterministic reasoning?), and perhaps most important of all, some degree of emergence. In my previous debates with him I have tried to explain that there are many, many people who already accept the limitations of simple, logical symbol-processing, and that approaches such as genetic algorithms, neural nets, the FARG-type systems of the Hofstadter school, and also my own "molecular" approach (closely related to Hofstadter's), all have at least some of teh characteristics that he is asking for. In particular, I have stresed that there is no black and white distinction between systems that are rigid (in the way that he complains of) and systems that are fluid and unpredictable (in the way that he prefers), but rather there is a continuum of types. And even more important, "programs" are completely neutral on this score: you can use "programs" to build systems that are rigid or systems that are labile. Mike: I know you do not accept this analysis of your position, but I believe that whenever you try to explain your position, it always come out as equivalent to this. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=94169430-374467
Re: [agi] The Test
Very briefly, my focus a while back in attacking programs was not on the sign/ semiotic - and more particularly, symbolic - form of programs, although that is v. important too. My focus was on the *structure* of programs - that's what they are: structured and usually sequenced sets of instructions.No matter how sophisticated their structure, and/or their capacity to adapt their structure, they are still structured. I'm unclear what you mean by structure. Interpretaton 1: - Every program in a modern computer language is a structured and sequenced set of instructions. It isn't possible to write an unsequenced set of instructions, because the language itself imposes that structure. If structured programs cannot be intelligent, then if I understand you correctly, it follows that what you are saying is that it is *impossible* to write intelligent systems in modern computer programming languages. Given that modern computer languages are Turing complete (modulo space and time limitations), your claims would therefore be equivalent to saying that intelligence is not computable. Interpretation 2: - May be you mean something a little stronger by structure? That the way that human beings engineer software is very structured, and software that has been engineered by humans with that kind of structure cannot possibly solve unstructured problems. Do you think, then, that it is possible for a human to write a structured program that generates unstructured programs that have general intelligence? - -Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=94090141-147bec
Re: [agi] The Test
I think your approach here *is* representative - &, as you indicate, the details of different approaches to AGI in this discussion, aren't that important. What is common IMO to your and the thinking of others here is that you all start by asking yourselves : what kinds of programming will solve AGI? Because programming is what interests you most and is your life. Actually, that isn't necessarily accurate. I'm currently collaborating with a cognitive scientist, and I've seen other people here hint at drawing their own inspiration from cognitive science and other non-programming disciplines. I reason the problem like this: 1. I know intelligence is possible, by looking at the animal kingdom. 2. I don't believe that the animal kingdom is doing something that is formally uncomputable (i.e., intelligence is computable). 3. I can see the things that intelligence can do, and have ideas about how it may work. 4. I recognize that biological computing machinery is vastly different to artificial computing machinery. 5. I assume that it is possible to build intelligence on current artificial computing machinery (i.e., intelligence is computable on current computers) 5. So, my goal is to translate those ideas about intelligence to the hardware that we have available. Programming comes into it not because we are obsessed with programming, but because we have to make do with the computing machinery that is available to us. We're attempting to exploit the strengths of computing machinery (such as its ability to do fast search and precise logical deduction) to make up for the weaknesses of the machinery (such as the difficulty in analogizing or associative learning). I don't believe there is only one path to intelligence, and we must be very conscious of the platform that we are building on. What you have to do in order to produce a true, crux idea, I suggest, is not just define your approach but APPLY IT TO A PROBLEM EXAMPLE OR TWO of general intelligence - show how it might actually work. Well, that is what many of us are doing. We have these plausible crux ideas, and we're now attempting to apply it to problems of general intelligence. It takes time to build systems, and the more ambitious the demonstration the longer it takes to build. I have my own challenge problems in the pipeline (I have to start very small, and have been using the commonsense problem page*), and I know most serious groups involved in system building have their own problems too. * http://www-formal.stanford.edu/leora/commonsense/ I've mentioned Semantic Web reasoning and General Game Playing. Even something like the Weka toolkit could be seen as a kind of general intelligence - you can run their machine learning algorithms on any kind of dataset and it will discover novel patterns. I admit that those are weak examples from an AGI perspective because they are purely symbolic domains, but it seems that AGI comes in where those kind of examples end. My point is, however, that general purpose reasoning is possible - I think there are plenty of signs of how it might actually work. You have to show how, for example, your GA might enable your lego-constructing system to solve an unfamiliar problem about building a dam of rocks in water. You must show that even though it had only learned about regularly-shaped bricks, it could neverthless recognize irregularly-shaped rocks as, say, "building blocks"; and even though it had only learned to build on solid ground, it could nevertheless proceed to build on ground submerged in water. [I think BTW, when you try to do this, you will find that GA's *won't* work] Why not? Genetic algorithms have been used in robots that learn how to move. You can connect a GA up to a set of motors and set up the algorithm so that movement is rewarded. Attach the motors to legs and put it on land, and the robot will eventually learn that walking maximizes its goals. Put the motors into fins and a tail and put it in water, and the robot will eventually learn that swimming maximizes its goals. Isn't this a perfect example of how GAs can problem-solve across domains? Or to address your specific (but more challenging) problem directly... Lets say, instead, that we're using GAs to generate high level strategies, plans and reasoning... the GA may evolve, on land, some wall-building strategies: 1. Start with the base 2. Put lego blocks on top of other lego blocks 3. Make sure lego blocks are stacked at an even height 4. Make sure there are no gaps When we give the robot the goal of building a dam, and it may then take those existing strategies and evolve generalizations: Here's one: 1. Start with the base 2. Put things on top of other things 3. Make sure things are stacked at an even height 4. Make sure there are no gaps This could happen by a cross-over or mutation that generalizes categories (Lego block -> Thing) -- and it may be the case that an AGI-op
Re: [agi] The Test
Benjamin [as in Johnston :)], Thankyou for a detailed response which is totally constructive. (An uncommon thing and I appreciate it). And therefore v. helpful. It's helps me understand how you & others think. I can see more clearly why you believe - reasonably from your POV - that crux ideas have been offered. I hope I can show you why they're not really crux ideas. I think your approach here *is* representative - &, as you indicate, the details of different approaches to AGI in this discussion, aren't that important. What is common IMO to your and the thinking of others here is that you all start by asking yourselves : what kinds of programming will solve AGI? Because programming is what interests you most and is your life. And in assessing the value of different approaches, you reason logically, as you do, for example, about GA's: "If you have a "genetic language" that is sufficiently general, and infinite computing power, then a good genetic algorithm can eventually solve any computable problem." Well, put like that, how can GA's fail? Even if you take a more specific logical formulation like - (loosely off the top of my head) - "GA's can mix a given set of elements any which way to arrive at new, unforeseen approaches to any problem" - it can still sound good, as if it might solve AGI. However, logical reasoning proves nothing - and can be just as easily used to "disprove" all these approaches, as indeed it has been. What you have to do in order to produce a true, crux idea, I suggest, is not just define your approach but APPLY IT TO A PROBLEM EXAMPLE OR TWO of general intelligence - show how it might actually work. You have to show how, for example, your GA might enable your lego-constructing system to solve an unfamiliar problem about building a dam of rocks in water. You must show that even though it had only learned about regularly-shaped bricks, it could neverthless recognize irregularly-shaped rocks as, say, "building blocks"; and even though it had only learned to build on solid ground, it could nevertheless proceed to build on ground submerged in water. [I think BTW, when you try to do this, you will find that GA's *won't* work] You don't just have to tell me in general terms what your programming approach can do, you have to apply it to specific true AGI END-PROBLEMS - and invite additional tests. I suggest you look again at any of the approaches you mention, as formally outlined, and I suggest you will not find a single one, that is actually applied to an end-problem, to a true test of its AGI domain-crossing potential. And I think if you go through the archives here you also won't find a single attempt in relevant discussions to do likewise. On the contrary, end-problems are shunned like the plague. (And you see yet another example of this general philosophy in Arthur Murray's recent formulation of his system/approach - no attempt to apply it to a general intelligence end-problem, only the non-AGI problems that he has carefully selected. Happens again and again. Yet another reason why that "General Test" is so important). Without application to AGI problem examples, you don't have crux ideas, you only have "hand-waving around the problem" And I quote an eloquent post from a Slashdot discussion of the McKinstry/Singh suicides - which underlines my points - it testifies to the long history of different AI/AGI schools of programming, which all, I suggest, were never really applied to AGI end-problems, or a true AGI test, as they should have been from the very beginning. The post also offers hope because it shows that when you really pressure AI/AGI-ers to apply themselves to end-problems, as with DARPA, you start to get real results - but you do really have to pressure. (I appreciate DARPA's AGI status is debatable): """It's discouraging reading this. Especially since I knew some of the Cyc [cyc.com] people back in the 1980s, when they were pursuing the same idea. They're still at it. You can even train their system [cyc.com] if you like. But after twenty years of their claiming "Strong AI, Real Soon Now", it's probably not happening. I went through Stanford CS back when it was just becoming clear that "expert systems" were really rather dumb and weren't going to get smarter. Most of the AI faculty was in denial about that. Very discouraging. The "AI Winter" followed; all the startups went bust, most of the research projects ended, and there was a big empty room of cubicles labeled "Knowledge Systems Laboratory" on the second floor of the Gates Building. I still wonder what happened to the people who got degrees in "Knowledge Engineering". "Do you want fries with that?" MIT went into a phase where Rod Brooks took over the AI Lab and put everybody on little dumb robots, at roughly the Lego Mindstorms level. Minsky bitched that all the students were soldering instead of learning theory. After a deca
Re: [agi] The Test
Richard,:Mike, When you say "I just believe that our thinking works on different mechanistic/ computational principles to those of programs" ... What you are really trying to say is that intelligence is not captured by a certain type of rigid, pure symbol-processing AI. The key phrase is "symbol-processing", which has connotations a certain approach to the representation of knowledge Richard, Thankyou for a sympathetic response, but I suggest - in a well-meaning way - that it would be worth your while giving me credit, if only provisionally, for a little more intelligence and awareness than you do. Very briefly, my focus a while back in attacking programs was not on the sign/ semiotic - and more particularly, symbolic - form of programs, although that is v. important too. My focus was on the *structure* of programs - that's what they are: structured and usually sequenced sets of instructions.No matter how sophisticated their structure, and/or their capacity to adapt their structure, they are still structured. So what I am saying - v. loosely for the moment - is that you *cannot* employ a programmed/ *structured* approach to *ill-structured* problems - and there isn't any evidence that humans actually do, or that AGI's can successfully. Hence it was that the great Herbert Simon himself distinguished between "programmed" and "NONPROGRAMMED" decisions - his term, which still obtains to this day in management science, and is not about symbol-processing. And ill-structured problems, I suggest, are the stuff of AGI. As I said, I will set out one last, v. different and systematic presentation of this POV in a while, which people can ignore or not - I did not mean to reignite the argument now, and there's no need to comment for the moment. P.S. Here's one analogy and also much-more-than-analogy of what I am talking about. As I said in singularity, the new genetics of Venter & co is changing everything, and will change the way we think about programs too. It's fundamentally changing paradigms. One way that it's doing this, (which I didn't mention), is making us think in terms of "self-assembling" genomes. Now clearly "self-assembly" is a totally different paradigm for thinking about everything - a paradigm we haven't even begun to master. We don't know how to create self-assembling machines, only ones pre-assembled according to a rigid blueprint.,(although we are starting) - and its' pre-assembled machines that have shaped science's entire view of the world. Nature mastered self-assembly long ago - with life. But it didn't, I suggest, just master self-assembling *forms*, it mastered self-assembling *behaviour*. Computers currently are only capable of programs - pre-assembled behaviour which must follow a structured blueprint. Human courses of action are by contrast, self-assembled, as they happen - still more so than biological forms - examples of "making it up as you go along" without any structured blueprint. That's what your post to me was. That's what the next minute of your life and every minute after that will be. And that's what AGI's will need to succeed and survive. Later. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93872219-b642cb
Re: [agi] The Test
Mike Tintner wrote: I believe we are thinking machines and not in any way magical. I just believe that our thinking works on different mechanistic/ computational principles to those of programs - which someone apart from me, surely should at least question. It has to be a serious *possibility* that programs equal narrow AI, and are the wrong paradigm for AGI. Mike, You are repeating a statement that you have made before (and which I have addressed before), and this is just going to cause great confusion again. When you say "I just believe that our thinking works on different mechanistic/ computational principles to those of programs" you are using the word "programs" in a misleading way. "Programs" in general are capable of implementing any type of AI whatsoever, ranging from the most stupid-brained AI that you hate, to the most flexible, creative, unpredictable (etc) AI that you would like to see. What you are really trying to say is that intelligence is not captured by a certain type of rigid, pure symbol-processing AI. The key phrase is "symbol-processing", which has connotations a certain approach to the representation of knowledge. The way you phrase your position, you look like one of the "computers cannot do intelligence because intelligence is not COMPUTATION" crowd. These people believe that there is something magical and non-computational about thought. You are not the first person to complain about the problems associated with the narrow symbol-processing approach, not by a long way: many of the experts on this list already bought that message decades ago. So: I already agree that intelligence is not going to happen that way! But every time you say "intelligence is more than just programs" I can only shake my head and watch while many other people on this list take your words the wrong way and a huge, pointless debate kicks off again. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93829695-4b54c3
Re: [agi] The Test
Benjamin Johnston wrote, among other things: I like to think about Deep Blue a lot. Prior to Deep Blue, I'm sure that there were people who, like you, complained that nobody has offered a "crux" idea that could make truly intelligent computer chess system. In the end Deep Blue appeared to win largely by brute force computing power. What I find most interesting is that Kasparov didn't say he was beaten by a particularly strong computer chess system, but claimed to see deep intelligence and creativity in the machine's play. That is, he didn't think Deep Blue was merely a slightly better version than the other chess systems, but he felt it had something else. He was surprised by the way the machine was playing, and even accussed the IBM team of cheating. You know, this gets me thinking that may the idea of intelligence is misleading. Maybe it's not really something like power or strength that is objective, but something more like deliciousness, that exists only as something we say about something else and isn't really a characteristic of the object. andi - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93812712-fac443
Re: [agi] The Test
On Feb 5, 2008 11:36 PM, Benjamin Johnston <[EMAIL PROTECTED]> wrote: > Well, as I said before, I don't know which will directly produce general > intelligence and which of them will fail. > My point, again, is that we don't know how the first successful AGI will > work - but we can see many plausible ideas that are being pursued in the > hope of creating something powerful. Some of these are doomed to fail; but > we don't really know which ones they are until we try them. It doesn't seem > fair for you to say that nobody has offered a "crux" idea, and I'd prefer > that people follow their passions rather than insist that everybody should > get hung up on the centuries/millennia old question of what exactly is > intelligence. Thankyou. Your list is very informative. I think its worth mentioning the dangerous phenomenon you touched on here. For some reason, people get religious about their approaches. "No, my idea is better. I can't prove why yet, but it'll work." The problem with this line of reasoning (as we've all experienced) is it ends with "Lets just not argue about which approach is better." I think we all agree that some approaches _are_ better than others. We might not agree on which ones are which, but I don't want to run away from that discussion. You mentioned passion -- I'm passionate about solving strong AI, not about pursuing my ideas even if they're wrong. I don't think any of us want to waste our time working on a flawed idea because nobody told us. The other reason I think discussing this stuff is worthwhile is thus: I think eventually what we all want is the same. We want a machine into which we plug a reward function and maybe a webcam or something and then we can teach it to talk and think. Thats what I imagine anyway. Maybe any of the methods you talked about could be used to make that. Its like we're dreaming of inventing computers while we work on our different CPU designs. I want to talk about how the computer will fit together. If the cpu is the hard bit, I want as good a spec as possible, and that means knowing and discussing the infrastructure. I don't want to work on something which could never actually power an intelligent system because I didn't think big picture. Thats a real danger. If its true, tell me that I'm in the wrong forest. -J > -Benjamin Johnston > > > - > This list is sponsored by AGIRI: http://www.agiri.org/email > To unsubscribe or change your options, please go to: > http://v2.listbox.com/member/?&; > - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93801513-06c777
Re: [agi] The Test
On Feb 4, 2008 11:42 PM, Mike Tintner <[EMAIL PROTECTED]> wrote:
> The test, I suggest, is essentially; not the Turing Test or anything like
> that but "The General Test." If your system is an AGI, or has AGI potential,
> then it must first of all have a skill and be able to solve problems in a
> given doman. The "test" is then: can it a) independently learn a skill in an
> adjacent domain, and/or b) pass a problemsolving test in an adjacent domain
> (to be set by someone other than the systembuilder!). If it can play soccer,
> can it learn how to play rugby and solve problems in rugby? If it can build
> Lego constructions, can it learn to build a machine? If it can search for
> hidden items, can it learn to play hide-and-seek? The General Test then is
> simply a test of whether a system can generalize its skill(s). If it knows
> how to put together a set of elements in certain kinds of ways, can it then
> learn to put those same elements together [and perhaps some new ones] in new
> kinds of ways?
Interesting test. However, as others have mentioned it is a difficult
test to evaluate in practice. Here's my proposal:
I propose that the purpose of any 'intelligent' system / agent is to
pursue goals. These goals can be specified in any way; 'classical'
victory condition-type goals, maintainance goals, whatever. An
intelligent system should be evaluated based on how well it can pursue
goals.
In particular, the quality of an intelligent system should be
evaluated based on:
- Optimality. An intelligence should try to optimize its goal-seeking
behavior for maximum 'reward'.
- Adaptability. If the world changes and makes different strategies
optimal, the system should account for this in its behaviour.
- Generality. A general intelligence should be able to 'solve' as wide
a range of goals or subgoals as possible.
Clearly humans are classified intelligent with this metric. Dogs are
still intelligent, but less intelligent. They have much less
generality in the goals they can solve. They are also less adaptable
('creative') than people.
Is a washing machine intelligent? It certainly minimally fits the
'intelligence' criteria of being able to solve a goal. The goal of my
washing machine is to make it easy for me to wash my clothes. Does it
do this optimally? No. Is it adaptable? Not really. Can it solve any
other goals? No.
Perhaps to be flagged 'intelligent' some minimal benchmark in
optimality, adaptability and generality is required. This is not the
interesting end of the scale.
> That's what people should be doing here centrally - discussing and
> exchanging their ideas about how to solve the General Test. The fact that no
> one is discussing this (despite vast volumes of overall discussion) suggests
> very powerfully that no one *has* an idea.
I think solutions are easy. Asking the right question is hard. Here's
my favorites:
"What kind of information does a general intelligence need to store
and manipulate?"
"What are the most fundamental elements of information you need to
store? What requirements are there on these pieces of information? How
can they be combined?"
"What is the simplest goal an intelligent system could possibly learn to solve?"
"What features must a good intelligent system have? If you were
writing a software engineering spec for an AI, what would it look
like?"
... anyone?
-J
-
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=8660244&id_secret=93782378-41d2ce
Mindforth and the Wright Brothers ... [WAS Re: [agi] The Test]
A. T. Murray wrote: Mike Tintner wrote in the message archived at http://www.mail-archive.com/agi@v2.listbox.com/msg09744.html [...] The first thing is that you need a definition of the problem, and therefore a test of AGI. And there is nothing even agreed about that - although I think most people know what is required. This was evident in Richard's recent response to ATMurray's recent declaring of his "Agi" system. Richard clearly knew pretty well why that system failed the AGI "test" but he didn't have an explicit definition of the test at his fingertips. Richard Loosemore "clearly knew pretty well" nothing of the sort. His was a lazy man's response. He did not download and experiment with the MindForth program at http://mentifex.virtualentity.com/mind4th.html and http://mind.sourceforge.net/mind4th.html -- he only made a few generalizations about what he lazily _thought_ MindForth might be doing. In the archive http://www.mail-archive.com/agi@v2.listbox.com/msg09674.html Richard Loosemore vaguely compares sophisticated MindForth with the canned-reponse "Eliza" program -- which nobody ever claimed was an artificial intelligence. Richard Loosemore furthermore suggested that all of the cognitive processes in the Eysenck & Keane textbook of Cognitive Psychology would have to be implemented in MindForth before it could be said to have achieved True AI functionality. That demand is like telling Wilbur and Orville Wright that they have to demo a transatlantic French Concorde jet before they may claim to have achieved "true airplane functionality." > [snip] Well, I have had some people get mad at me before, but not when I was being so ... charming. Arthur, if there is an analogy between Mindforth and the Wright Brothers, then you, alas, are just standing on the sand at Kitty Hawk, waving your hands up and down and shouting "I can flap! I can flap!". You don't have to build Concorde at the first attempt, you just have to get your plane off the ground and show that it can travel any distance at all under its own power. I assumed that your own description of what Mindforth did was accurate (it was, wasn't it?) and on that basis I saw it merely flapping its wings in the same way that Eliza did 30 years ago. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93769008-df05ed
RE: [agi] The Test
> Fine. Which idea of anyone's do you believe will directly produce > general intelligence - i.e. will enable an AGI to solve problems in > new unfamiliar domains, and pass the general test I outlined? (And > everyone surely agrees, regardless of the test, that an AGI must have > "general" intelligence). Well, as I said before, I don't know which will directly produce general intelligence and which of them will fail. I have my own theories about which approaches are more likely to succeed than others, and about which approaches are fundamentally wrong. However, most serious ideas seem to have a plausible story, and I'm not ready to completely rule out any serious idea until it is proven wrong. I'll briefly discuss some ideas below. You may not agree with my interpretation of the approaches, and you may not fully agree with my argument about why they're plausible... but I think that you surely have to agree that a plausible argument can be made for most of this research, and it is clear that the people conducting the research can see themselves as addressing the crucial questions. That is, while I'm not the right person to be arguing the details of these approaches, I'm confident that many researchers here wouldn't be devoting their time to their research if they didn't see a coherent picture for how their work fits into the grand scheme of AGI. Many apologies to other readers if I've not included your preferred approach or have misrepresented/misinterpreted your ideas. I've just taken a quick and informal sample here. The details aren't as important as the overall message. Logic --- An automated theorem prover is an extremely general purpose intelligence. Consider, for example, how logics may be adapted to many different domains on the Semantic Web or the increasing strength of competitors in General Game Playing competitions (surely not long before they're better than the average human at any novel game?). As to whether logic can be applied to general purpose embodied intelligent systems remains to be seen - I think the symbol grounding problem points towards logic not being enough - but researchers looking into logics with uncertainty or logics that incorporate iconic representations are effectively exploring a possible "solution" to the symbol grounding problem. In other words, these researchers are saying "Logical deduction offers true 'general intelligence' in symbolic domains, and we're trying to adapt that intelligence to real life situations": a plausible crux idea and worth pursuing. Hybrid Systems --- If we just keep doing what we're doing in "Narrow AI", but look at combining many components into a coherent architecture then it seems plausible that we'll eventually end up with a system that is indistinguishable from an ideal general intelligence. It may not be an elegant answer, but it may be an answer. This gives good reason to pursue integration. Consider for example, problems like the DARPA Grand Challenges. In current systems, obstacles may be specifically identified against a hand-coded database. In the next generations, these representations might become more generic and learnt from experience. I see a plausible progression to increasingly more powerful systems. When the system can identify and learn the behavior of any new object it encounters (and the rules that govern it), it may then be able to reason about that object and construct plans that uses the object in novel ways. At first the planning algorithms seek merely to visit way-points. Future versions, with richer goals, richer models and more powerful reasoning may autonomously deduce novel behaviors beyond their explicit programming (e.g., that truck will run into the pedestrian! my higher goal of not hurting pedestrians means that the best plan is one in which I stop in front of the truck so that it crashes into me instead of the pedestrian). Genetic Algorithms and other search algorithms --- If you have a "genetic language" that is sufficiently general, and infinite computing power, then a good genetic algorithm can eventually solve any computable problem. Evolution eventually discovered human beings - given infinite computing power, then at worst you could evolve a virtual human! It seems reasonable then to consider exploring genetic or other search algorithms that have a bias towards the kinds of problems encountered by humans and AGI. Activation, Similarity, Analogizing, HTM, Confabulation and other "targeted" approaches --- There seem to be a lot of groups working on specific modes of thought. You may not be convinced that they're solving enough of the problem, but it seems plausible to me that maybe general intelligence really is easy once you've managed to solve some particular problem. That is, we might have a 80/20 rule or even a 99.9/0.1 rule at play with intelligence. Maybe the brain only does learn a few techniques for problem solv
Re: [agi] The Test
On 05/02/2008, Mike Tintner <[EMAIL PROTECTED]> wrote: > William P : I can't think > of any external test that can't be fooled by a giant look up table > (ned block thought of this argument first). > > A by definition requirement of a "general test" is that the systembuilder > doesn't set it, and can't prepare for it as you indicate. He can't know > whether the test for, say, his lego-constructing system is going to be > building a machine, or constructing a water dam with rocks, or a game that > involves fitting blocks into holes. He can't know. but he might guess. It will be hard to test between the builders lucky guess(es) and generality. > His system must be able to adapt to any > adjacent-domain activity whatsoever. That too is the point of the robot > challenge test - the roboticists won't know beforehand what that planetary > camp emergency is going to be. I think we have different ideas of what a test should be. I am looking for a scientific test, in which repeatability and fairness are important features. One last question what exactly defines adjacent in your test? Is composing poetry adjacent to solving non-linear equations. I agree that this type of testing will winnow out lots of non-general systems. But it might let a few slip through the cracks or say a general system is non-general. I would fail the test some days when I am ill, as all I would want to do is go to sleep not try and solve the problem. Will Pearson - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93723647-9e9867
Re: [agi] The Test
Benjamin: > I believe that you're misrepresenting the situation. I would guess that most people on this list have an idea that they are pursuing because they believe it has a chance at creating general intelligence. Fine. Which idea of anyone's do you believe will directly produce general intelligence - i.e. will enable an AGI to solve problems in new unfamiliar domains, and pass the general test I outlined? (And everyone surely agrees, regardless of the test, that an AGI must have "general" intelligence). Please note very carefully - I am only asking for an idea that will play a direct *part* in solving new-domain problems. Of course *many* ideas will be required to do the job completely. I am only asking for one that gives a glimmer of hope - and am saying I haven't seen a single idea that addresses that problem/goal directly. A new search algorithm, for example, does not address the problem. Neither does a new logic of uncertainty. They might be good and useful new ideas, but they don't address the problem. I have, however, seen people v. definitely avoiding the problem -and hoping that a solution will "emerge" (not a chance). And if you do address the problem, I think you'll find that it requires not just a creative idea, but a whole new creative *paradigm* of problem-solving. Benjamin: I get the impression from this posting, and your earlier posting about a "Simple mathematical test of cog sci" that you see intelligence as something "crazy and spontaneous" (to use your words) - something almost magical. With that position, it would seem logical for you to expect a solution to AGI to also appear magical. No, that impression is completely wrong - although since you're the second person to say that, maybe it's my fault. I believe we are thinking machines and not in any way magical. I just believe that our thinking works on different mechanistic/ computational principles to those of programs - which someone apart from me, surely should at least question. It has to be a serious *possibility* that programs equal narrow AI, and are the wrong paradigm for AGI. Hence, the oft-stated objection: "The problem with most robots is that they tend to be, well, robotic. They know nothing they aren't programmed to know, and can do nothing they aren't programmed to do." Robot Pals. Scientific American Frontiers (April 13, 2005). So, Ben, show me one idea from anyone that can get a program to do what it isn't programmed to do - cross into unfamiliar domains. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93712126-c7dbd9
Re: [agi] The Test
Mike Tintner wrote in the message archived at http://www.mail-archive.com/agi@v2.listbox.com/msg09744.html > [...] > The first thing is that you need a definition > of the problem, and therefore a test of AGI. > And there is nothing even agreed about that - > although I think most people know what is required. > This was evident in Richard's recent response to > ATMurray's recent declaring of his "Agi" system. > Richard clearly knew pretty well why that system > failed the AGI "test" but he didn't have an explicit > definition of the test at his fingertips. Richard Loosemore "clearly knew pretty well" nothing of the sort. His was a lazy man's response. He did not download and experiment with the MindForth program at http://mentifex.virtualentity.com/mind4th.html and http://mind.sourceforge.net/mind4th.html -- he only made a few generalizations about what he lazily _thought_ MindForth might be doing. In the archive http://www.mail-archive.com/agi@v2.listbox.com/msg09674.html Richard Loosemore vaguely compares sophisticated MindForth with the canned-reponse "Eliza" program -- which nobody ever claimed was an artificial intelligence. Richard Loosemore furthermore suggested that all of the cognitive processes in the Eysenck & Keane textbook of Cognitive Psychology would have to be implemented in MindForth before it could be said to have achieved True AI functionality. That demand is like telling Wilbur and Orville Wright that they have to demo a transatlantic French Concorde jet before they may claim to have achieved "true airplane functionality." Sorry, Richard, but the AI breakthrough functionality is, plain and simple, the ability to think -- to activate an associative string of concepts and to express the thinking in the generative grammar of Chomsky. There is no requirement that people be other than lazy, smug and self-satisfied on this AGI list. I felt that I should announce the end of the decade-long process of debugging MindForth AI. Now the controversy has spilled over to http://onsingularity.com/item/3175 and the dust has not yet settled. Richard is beginning to act like ESY! > > The test, I suggest, is essentially; not the Turing > Test or anything like that but "The General Test." > If your system is an AGI, or has AGI potential, > then it must first of all have a skill and be > able to solve problems in a given doman. [...] The skill of MindForth is spreading activation -- from concept to concept -- under the direction of a Chomksyan linguistic superstructure. Now I would like to digress and draw Ben Goertzel's math-minded attention to my latest "creative idea" at http://mind.sourceforge.net/computationalization.html#syllogism where on 30 January 2008 I thought up and loaded-up: It may be possible to endow an AI mind with the ability to think in syllogisms by creating super-concepts or set-concepts above and beyond, and yet in parallel with, the ordinary concepts. Certain words like "all" or "never" may be coded to duplicate a governed concept and to endow the duplicate with only one factual or asserted attribute, namely the special relationship modified by the "all" or "never" assertion. Take, for instance, the following. All fish have tails. Tuna are fish. Tuna have tails. When the AI mind encounters an "all" proposition involving the verb "have" and the direct object "tails", a new, supervenient concept of "fish-as-set" is created to hold only one class of associative nodes -- the simultaneous association to "have" and to the "tail" concept. Whenever the basic "fish" concept is activated, the fish-as-set concept is also activated, ready to "pounce," as it were, with the supervenient assertion that all fish have tails. Thenceforth, when any animal is identified as being a fish by some kind of "isA" tag, the "fish-as-set" concept is also activated and the AI mind superveniently knows that the animal in question has a tail. The machine reasoning could go somewhat like the following dialog. Do tuna have tails? Are tuna plants? Tuna are animals. What kind of animals? Tuna are fish. All fish have tails. Tuna have tails. The ideas above conform with set theory and with the notion of neuronal prodigality -- that there need be no concern about wasting neuronal resources -- and with the idea of "inheritance" in object-oriented programming (OOP). Whereas normally a new fiber might be attached to the fiber-gang of a redundantly entertained concept, it is just as easy to engender a "concept-as-set" fiber in parallel with the original, basic concept. For some basic concepts, there might be multiple concept-as-set structures reperesenting multiple "all" or "never" ideas believed to be the truth about the basic, ordinary concept. The AI mind thinking about an ordinary concept in the course of problem-solving, does not have to formally engage in the
Re: [agi] The Test
Er, you don't ask that in AGI. The general culture here is not to recognize the crux, or the "test" of AGI. You are the first person here to express the basic requirement of any creative project. You should only embark on a true creative project - in the sense of committing to it - if you have a creative "idea", i.e. if you have a provisional definition of the problem and a partial solution to it, one that will make people say, "Yes that might work." (Many more ideas will of course usually be required). It's one of the most extraordinary phenomena that everyone, but everyone, involved in the creative community of AGI resists doing that and has extensive rationalisations of why they're not doing that.Every AGI systembuilder has several "ideas" about how to do *other* things, that may be auxiliary to AGI, like search more efficiently, and logics to deal with uncertainty, but no one has offered a "crux" idea. I believe that you're misrepresenting the situation. I would guess that most people on this list have an idea that they are pursuing because they believe it has a chance at creating general intelligence. Some here are research students or professional academics, who enjoy the spirit of discussion but are careful about disclosing the specifics of their own ideas until they have first been 'timestamped' by publication. Others have already made their position clear on this list and in publication, but it seems that you're rejecting their ideas as not "creative" enough. That doesn't mean nobody has offered a "crux" idea, it just means that nobody has offered an idea that you believe in. Personally, I'm optimistic about many of the ideas that have been floated here. I know that many will fail, and that only one can be the *first* to create AGI; but I see sufficient cause here for people to commit to their ideas and embark on a true creative project to test those ideas and discover which ones are workable and which aren't. Not everybody is sufficiently well staffed and funded to lay out a roadmap for their entire project today, but I'm sure everybody has an idea of how their work fits into a big picture, where they ultimately see it going, and how their work relates to AGI. I get the impression from this posting, and your earlier posting about a "Simple mathematical test of cog sci" that you see intelligence as something "crazy and spontaneous" (to use your words) - something almost magical. With that position, it would seem logical for you to expect a solution to AGI to also appear magical. While human intelligence is impressive, I don't think it is inherently magical. If you look at a timeline of evolution, you'll see that it took billions of years to evolve multi-cellular life, hundreds of millions of years to evolve mammals, but the evolutionary time difference between us and apes or even between us and mice is, by comparison, very small. Creating human-like intelligence doesn't appear to take much extra work (for evolution) once you can do mouse-like intelligence. I like to think about Deep Blue a lot. Prior to Deep Blue, I'm sure that there were people who, like you, complained that nobody has offered a "crux" idea that could make truly intelligent computer chess system. In the end Deep Blue appeared to win largely by brute force computing power. What I find most interesting is that Kasparov didn't say he was beaten by a particularly strong computer chess system, but claimed to see deep intelligence and creativity in the machine's play. That is, he didn't think Deep Blue was merely a slightly better version than the other chess systems, but he felt it had something else. He was surprised by the way the machine was playing, and even accussed the IBM team of cheating. I'm certainly not saying that Deep Blue exhibited general intelligence or that it was anything more than a powerful move-searching machine (with well designed heuristics); but the fact that Kasparov had played many computer systems before, but saw an exceptional intelligence in Deep Blue suggests to me that intelligence isn't magical, but is something that can emerge when a suitable mechanism performed on a sufficient scale. Look at our own brains for example: while a single neuron is not yet 100% understood; each neuron appears to perform a minimal computation that when combined in the billions emerges to create an extremely robust intelligence. Brute force search or assembling millions of neurons might not seem like "crux" ideas, but when they are used towards a coherent vision, it is possible to create something that appears to be deeply intelligent. I don't know how the first successful AGI will work - it may be based on a special logic, a search algorithm, a neural network, a vast knowledge base, some new mechanisms, or a hybrid combination of several approachs - I think, however, that we have seen many plausible ideas that are being pursued in the hope of
Re: [agi] The Test
William P : I can't think
of any external test that can't be fooled by a giant look up table
(ned block thought of this argument first).
A by definition requirement of a "general test" is that the systembuilder
doesn't set it, and can't prepare for it as you indicate. He can't know
whether the test for, say, his lego-constructing system is going to be
building a machine, or constructing a water dam with rocks, or a game that
involves fitting blocks into holes. His system must be able to adapt to any
adjacent-domain activity whatsoever. That too is the point of the robot
challenge test - the roboticists won't know beforehand what that planetary
camp emergency is going to be.
("External testing", BTW, I suggest, should be a fundamental bottom-up part
of the culture of all AGI and robot systembuilding. Human students don't get
to set their own exams/ intelligence tests! It would be absurd).
What I think is so useful about the idea of a "general" test is that you
*don't* try to define it specifically in advance - other than as an
"adjacent domain" test. So it automatically applies to any would-be AGI
whatsoever and at any level - whether it's say a snake-like system that only
knows how to navigate through different terrains, or a complex would-be
humanoid system that claims conversational powers. I think the latter is
wildly unrealistic and unlikely to happen till the distant future, but it
doesn't matter - the general test would still be applicable. And you would
then have a focussed Turing test - if your system claims knowledge and can
converse about one domain, then it should be able to learn and converse
about a new but related domain.
If you had a general test as a focus, too, you wouldn't, I suggest, get AGI
systembuilders wasting years of their lives on ill-defined projects, as has
clearly happened and will otherwise continue to happen.
-
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=8660244&id_secret=93653402-e18623
Re: [agi] The Test
On 04/02/2008, Mike Tintner <[EMAIL PROTECTED]> wrote: > (And it's a fairly safe bet, Joseph, that no one will now do the obvious > thing and say.." well, one idea I have had is...", but many will say, "the > reason why we can't do that is...") And maybe they would have a reason for doing so. I would like to think of an external objective test, I like tests and definitions. My stumbling block for thinking of external tests is that I can't think of any external test that can't be fooled by a giant look up table (ned block thought of this argument first). That is something that when input X comes in at time t, output Y goes out. It can pretend to learn things by having poor performance early on and then "improve". Not all designs of systems use lots of external tests to prove their abilities. Take making a new computer architecture that you want to have the property of computational universality. You wouldn't try to give it a few programs see if it can run them and declare it universal you would program it to emulate a Turing Machine to prove its universality. Similarly for new chip designs of existing architecture, you want to prove them equivalent to the old ones. Generality of an intelligence is this sort of problem I think, due to the inability to capture it flawlessly with external tests. I would be interested to discuss internal requirements of systems, if anyone else is. I'd have thought that you with your desire for things to be spontaneous would be wary of any external test that can be gamed by non-spontaneous systems. Will Pearson - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93443404-a1be29
[agi] The Test
Joseph Gentle:> Eventually, you will have to write something which allows for emergent behaviour and complex communication. To me, that stage of your project is the interesting crux of AGI. It should have some very interesting emergant behaviour with inputs other than the information SLAM outputs... Why not just work on that difficult part now? Er, you don't ask that in AGI. The general culture here is not to recognize the crux, or the "test" of AGI. You are the first person here to express the basic requirement of any creative project. You should only embark on a true creative project - in the sense of committing to it - if you have a creative "idea", i.e. if you have a provisional definition of the problem and a partial solution to it, one that will make people say, "Yes that might work." (Many more ideas will of course usually be required). It's one of the most extraordinary phenomena that everyone, but everyone, involved in the creative community of AGI resists doing that and has extensive rationalisations of why they're not doing that.Every AGI systembuilder has several "ideas" about how to do *other* things, that may be auxiliary to AGI, like search more efficiently, and logics to deal with uncertainty, but no one has offered a "crux" idea. The first thing is that you need a definition of the problem, and therefore a test of AGI. And there is nothing even agreed about that - although I think most people know what is required. This was evident in Richard's recent response to ATMurray's recent declaring of his "Agi" system. Richard clearly knew pretty well why that system failed the AGI "test" but he didn't have an explicit definition of the test at his fingertips. The test, I suggest, is essentially; not the Turing Test or anything like that but "The General Test." If your system is an AGI, or has AGI potential, then it must first of all have a skill and be able to solve problems in a given doman. The "test" is then: can it a) independently learn a skill in an adjacent domain, and/or b) pass a problemsolving test in an adjacent domain (to be set by someone other than the systembuilder!). If it can play soccer, can it learn how to play rugby and solve problems in rugby? If it can build Lego constructions, can it learn to build a machine? If it can search for hidden items, can it learn to play hide-and-seek? The General Test then is simply a test of whether a system can generalize its skill(s). If it knows how to put together a set of elements in certain kinds of ways, can it then learn to put those same elements together [and perhaps some new ones] in new kinds of ways? The robotic challenge test set by the ICRA is a good one, precisely because it is a "General Test," requiring robotbuilders to solve *any* breakdown of equipment that may reasonably occur in a planetary exploration camp - and generalize their existing repair skills. That's what people should be doing here centrally - discussing and exchanging their ideas about how to solve the General Test. The fact that no one is discussing this (despite vast volumes of overall discussion) suggests very powerfully that no one *has* an idea. (And it's a fairly safe bet, Joseph, that no one will now do the obvious thing and say.." well, one idea I have had is...", but many will say, "the reason why we can't do that is...") - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=93375464-9c07fc
