RE: Language modeling (was Re: [agi] draft for comment)
From: Matt Mahoney [mailto:[EMAIL PROTECTED] --- On Sat, 9/6/08, John G. Rose [EMAIL PROTECTED] wrote: Compression in itself has the overriding goal of reducing storage bits. Not the way I use it. The goal is to predict what the environment will do next. Lossless compression is a way of measuring how well we are doing. Predicting the environment in order to determine which data to pack where, thus achieving higher compression ratio. Or compression as an integral part of prediction? Some types of prediction are inherently compressed I suppose. John --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] A NewMetaphor for Intelligence - the Computer/Organiser
On Thursday 04 September 2008, Mike Tintner wrote: You start v. constructively thinking how to test the non-programmed nature of - or simply record - the actual writing of programs, and then IMO fail to keep going. You could trace their keyboard presses back to the cerebellum and motor cortex, yes, this is true, but this isn't going to be like tracing the programmer pathway in a brain. You might just end up tracing the entire brain [which is another project that I fully support, of course]. You can imagine this as the signals being traced back to their origins back to the spine and the CNS like the cerebellum and motor cortex, and then from the somatosensory cortex that gave them the feedback for debugger error output (parse error, rawr), etc. You could even spice up the experimental scenario by tracking different strategies and their executions in response to bugs, sure. Ask them to use the keyboard for everything - (how much do you guys use the keyboard vs say paper or other things?) - and you can automatically record key-presses. Right. Hasn't anyone done this in any shape or form? It might sound as if it would produce terribly complicated results, but my guess is that they would be fascinating just to look at (and compare technique) as well as analyse. I don't think it's sufficient to keep it as analyses, here's why: http://heybryan.org/humancortex.html Basically, wouldn't it be interesting to have an online/real-time/run-time system for keeping track of your brain as you program? This would allow for neurofeedback and some other possibilities. - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] A NewMetaphor for Intelligence - the Computer/Organiser
On Friday 05 September 2008, William Pearson wrote: 2008/9/5 Mike Tintner [EMAIL PROTECTED]: By contrast, all deterministic/programmed machines and computers are guaranteed to complete any task they begin. If only such could be guaranteed! We would never have system hangs, dead locks. Even if it could be made so, computer systems would not always want to do so. Have you every had a programmed computer system say to you. This program is not responding, do you wish to terminate it. There is no reason in principle why the decision to terminate the program couldn't be made automatically. These errors are computed. Do what I mean, not what I say is a common phrase thrown around in programming circles. The errors are not because that suddenly the ALU decided to not be present, and the errors are not because it suddenly lost its status as a Turing machine (although if you drove a rock through it, this is quite likely). Rather this is because you failed to write a good kernel. And yes, the decision to terminate programs can be made automatically, and I sometimes choose scripts on my clusters to kill things that haven't been responding for a certain amount of time, but usually I prefer to investigate it by hand since it's so rare. Very different kinds of machines to us. Very different paradigm. (No?) We commonly talk about single program systems because they are generally interesting, and can be analysed simply. My discussion on self-modifying systems ignored the interrupt driven multi-tasking nature of the system I want to build, because that makes analysis a lot more hard. I will still be building an interrupt driven, multi tasking system. That's an interesting proposal, but I'm wondering about something. Suppose you have a cluster of processors, and they are all communicating with each other in some way to divide up tasks and compute away. Now, given the ability to send interrupts from one another, and given the linear nature of each individual unit, is it really multitasking? At some point it has to integrate all of the results together at a single node for writing at a single address on the hdd (or something) so that the results are in one single place, that or the reading function of the results must do this. Is it really then multi-tasking and parallel? - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] A NewMetaphor for Intelligence - the Computer/Organiser
On Friday 05 September 2008, Mike Tintner wrote: Were your computer like a human mind, it would have been able to say (as you/we all do) - well if that part of the problem is going to be difficult, I'll ignore it or.. I'll just make up an answer... or by God I'll keep trying other ways until I do solve this.. or... .. or ... Computers, currently, aren't free thinkers. I'm pretty sure that compiler optimizers, that go in and look at your loops and other computational elements of a program, are able to make assessments like that. Of course, they'll just leave it as it is instead of completely ignoring parts of your program that you wish to compile, but it does seem similar. I recently came across an evolutionary optimizer for compilers to test parameters to gcc to try to figure the best way to compile a program on a certain architecture (to learn all of the gcc parameters yourself seems impossible sometimes, you see). Perhaps there's some evolved laziness in the human brain that could be modeled with gcc easily enough. - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] A NewMetaphor for Intelligence - the Computer/Organiser
On Friday 05 September 2008, Mike Tintner wrote: fundamental programming problem, right?) A creative free machine, like a human, really can follow any of what may be a vast range of routes - and you really can't predict what it will do or, at a basic level, be surprised by it. What do you say to the brain simulation projects? There is a biophysical basis to the brain and it's being discovered and hammered out. You can, in fact, predict the results of the eye-blink rabbit experiments (I'm working with a lab on this - the simulations return results faster than the real neurons do in the lab. You can imagine how this is useful for hypothesis testing purposes.). - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] A NewMetaphor for Intelligence - the Computer/Organiser
On Saturday 06 September 2008, William Pearson wrote: I'm very interested in computers that self-maintain, that is reduce (or eliminate) the need for a human to be in the loop or know much about the internal workings of the computer. However it doesn't need a vastly different computing paradigm it just needs a different way of thinking about the systems. E.g. how can you design a system that does not need a human around to fix mistakes, upgrade it or maintain it in general. Yes, these systems are interesting. I can easily imagine a system that generates systems that have low human maintenance costs. But suppose that the system that you make generates a system (with that low hu maint cost), and this 2nd-gen system does it again and again. This is the problem of clanking replicators too -- you need to have some way to correct divergence and for errors of replication; and not only that, but as you go into new environments there are new things that have to be taken into account for maintenance. Bacteria solve this problem with having many billions of cells per culture and then having enough genetic variability to somehow scrounge up a partial solution within time -- so that once you get to the Nth-generation you're not screwed entirely if some change occurs in the environment. There was a recent experiment in the news that has been going for 20 years, the Michigan man who had bacterial selection experiments in bottles for the past 20 years only to find that they evolved an ability to metabolize something they didn't metabolize before. That's an example of being able to work in new environments, and there's a lot of cost to it (dead bacteria, many generations, etc.) that silicon projects can't quite do simply because of resource/cost constraints if you use traditional approaches. What would an alternative approach look like? One where you don't need dead silicon projects, and one where you have enough instances of programs that you're able to find a solution with your genetic algorithm in enough time? The increasing availability of RAM and hdd space might be enough to let us bruteforce it, but the embodiment of bacteria in the problem domains is something that more memory strategies don't quite address. Thoughts? - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] A NewMetaphor for Intelligence - the Computer/Organiser
On Saturday 06 September 2008, Mike Tintner wrote: Our unreliabilty is the negative flip-side of our positive ability to stop an activity at any point, incl. the beginning and completely change tack/ course or whole approach, incl. the task itself, and even completely contradict ourself. But this is starting to get into an odd-mix of folk psychology. I was reading an excellent paper the other day that says this very plainly, written by Gerhard Werner: The Siren Call of metaphor: Subverting the proper task of Neuroscience http://www.ece.utexas.edu/~werner/siren_call.pdf The case of Neuro-Psychological vs. Naturalistic Neuroscience. For grounding the argument, let us look at the case of ‘deciding to’ [34] in studies of conditioned motor behavior in monkeys, on which there is a rich harvest of imaginative experimental work on scholarly reviews available. I write this in profound respect for the investigators who conduct this work with immense ingenuity and sophistication. However, I question the soundness of the conceptual framework on which such experiments are predicated, observations are interpreted, and conclusions are formulated. I contend that current practices tend to disregard genuine issues in Neurophysiology with its own definitions of what legitimate propositions and criteria of valid statements in this discipline are. Here is the typical experimental protocol: the experimenter uses some measure of neural activity of his/her choice (usual neural spike discharges), recorded from a neural structure (selected by him/her on some criterion, and determines relations to behavior that he/she created as link between two events: an antecedent stimulus ( chosen by him/her) and a consequent, arbitrary behavior, induced by the training protocol [49]. So far, the experimenter has done all the ‘deciding’, except leaving it up to the monkey to assign a “value” to complying with the experimental protocol. Different investigators summarize their experimental objective in various ways (in the interest of brevity, I slightly paraphrase, though being careful to preserving the original sense): to characterize neural computations representing the formation of perceptual decision [12]; to investigate the neural basis of a decision process [37]; to examine the coupling of neural processes of stimulus selection with response preparation [34], reflecting connections between motor system and cognitive processes [38] ; to assess neural activity indicating probabilistic reward anticipation [22,27]. In Shadlen and Newsome’s [37] evocative analogy “it is a jury’s deliberation in which sensory signals are the evidence in open court, and motor signals the jury’s verdict”. Helpful as metaphors and analogies can be as interim steps for making sense of the observation in familiar terms, they also import the conceptual burden of their source domain and lead us to attribute to the animal a decision and choice making capacity along principles for which Psychology has developed evidential and conceptual accounts in humans under entirely different conditions, and based on different observational facts. Nevertheless, armed with the metaphors of choice and decision, we assert that the observed neural activity is a “correlate” [19] of a decision to emit the observed behavior. As the preceding citations indicate, the observed neural activity is variously attributed to perceptual discrimination between competing (or conflicting) stimuli, to motor planning, or to reward anticipation; the implication being that the neural activity stands for (“represents”) one or the other of these psychological categories. So, Mike, when you write like: Our unreliabilty is the negative flip-side of our positive ability to stop an activity at any point, incl. the beginning and completely change tack/ course or whole approach, incl. the task itself, and even completely contradict ourself. It makes me wonder how you can assert the existence of a neurophysical basis of the existence of 'task', in terms of the *brain*, not in terms of our folk psychology and collective cultural background that has given us these names to these things. It's hard to talk about the brain from the biology-up, yes, that's true, but it's also very rewarding in that we don't make top-down misunderstandings. - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] A NewMetaphor for Intelligence - the Computer/Organiser
On Friday 05 September 2008, Terren Suydam wrote: So, Mike, is free will: 1) an illusion based on some kind of unpredictable, complex but *deterministic* interaction of physical components 2) the result of probabilistic physics - a *non-deterministic* interaction described by something like quantum mechanics 3) the expression of our god-given spirit, or some other non-physical mover of physical things I've already mentioned an alternative on this mailing list that you haven't included in your question, would you consider it? http://heybryan.org/free_will.html ^ Just so that I don't have to keep on rewriting it over and over again. - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] Recursive self-change: some definitions
On Thursday 04 September 2008, Mike Tintner wrote: Bryan, How do you know the brain has a code? Why can't it be entirely impression-istic - a system for literally forming, storing and associating sensory impressions (including abstracted, simplified, hierarchical impressions of other impressions)? 1). FWIW some comments from a cortically knowledgeable robotics friend: The issue mentioned below is a major factor for die-hard card-carrying Turing-istas, and to me is also their greatest stumbling-block. You called it a code, but I see computation basically involves setting up a model or description of something, but many people think this is actually synonomous with the real-thing. It's not, but many people are in denial about this. All models involves tons of simplifying assumptions. EG, XXX is adamant that the visual cortex performs sparse-coded [whatever that means] wavelet transforms, and not edge-detection. To me, a wavelet transform is just one possible - and extremely simplistic (meaning subject to myriad assumptions) - mathematical description of how some cells in the VC appear to operate. No, this is just a confusion of terminologies. I most certainly was not talking about 'code' in the sense of sparse-coded wavelet transform. I'm talking about code in the sense of source code. Sorry. - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] open models, closed models, priors
On Thursday 04 September 2008, Matt Mahoney wrote: Yes you do. Every time you make a decision, you are assigning a higher probability of a good outcome to your choice than to the alternative. You'll have to prove to me that I make decisions, whatever that means. - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
RE: Language modeling (was Re: [agi] draft for comment)
--- On Sun, 9/7/08, John G. Rose [EMAIL PROTECTED] wrote: From: John G. Rose [EMAIL PROTECTED] Subject: RE: Language modeling (was Re: [agi] draft for comment) To: agi@v2.listbox.com Date: Sunday, September 7, 2008, 9:15 AM From: Matt Mahoney [mailto:[EMAIL PROTECTED] --- On Sat, 9/6/08, John G. Rose [EMAIL PROTECTED] wrote: Compression in itself has the overriding goal of reducing storage bits. Not the way I use it. The goal is to predict what the environment will do next. Lossless compression is a way of measuring how well we are doing. Predicting the environment in order to determine which data to pack where, thus achieving higher compression ratio. Or compression as an integral part of prediction? Some types of prediction are inherently compressed I suppose. Predicting the environment to maximize reward. Hutter proved that universal intelligence is a compression problem. The optimal behavior of an AIXI agent is to guess the shortest program consistent with observation so far. That's algorithmic compression. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: AI isn't cheap (was Re: Real vs. simulated environments (was Re: [agi] draft for comment.. P.S.))
Matt, On 9/6/08, Matt Mahoney [EMAIL PROTECTED] wrote: Steve, where are you getting your cost estimate for AGI? 1. I believe that there is some VERY fertile but untilled ground, which if it is half as good as it looks, could yield AGI a LOT cheaper than other higher estimates. Of course if I am wrong, I would probably accept your numbers. 2. I believe that AGI will take VERY different (cheaper and more valuable) forms than do other members on this forum. Each of the above effects are worth several orders of magnitude in effort. Is it a gut feeling, or something like the common management practice of I can afford $X so it will cost $X? Given sufficient motivation and flexibility as to time and deliverable, this isn't a bad way to go. You can have it good, quick, and cheap - choose any two. If you drop cheap and choose the other two, then you can get there. My estimate of $10^15 is based on the value of the world economy, US $66 trillion per year and increasing 5% annually over the next 30 years, which is how long it will take for the internet to grow to the computational power of 10^10 human brains (at 10^15 bits and 10^16 OPS each) at the current rate of growth, doubling every couple of years. Even if you disagree with these numbers by a factor of 1000, it only moves the time to AGI by a few years, so the cost estimate hardly changes. Here, you are doing the same thing that you were accusing me of - working from available resources. I believe that present AGI efforts are SO misdirected (and actively interfered with) that unless something changes, humanity will never ever produce a true AGI. However, with those changes, I believe that really *useful* computational intelligence is just a few years away, though an AGI, that everyone here on this forum would wholeheartedly agree without reservation is an AGI, is probably 30 years away, as you guesstimated. However, my time guesstimate is based on the time needed to do some really basic research that remains undone. And even if the hardware is free, you still have to program or teach about 10^16 to 10^17 bits of knowledge, assuming 10^9 bits of knowledge per brain [1] and 1% to 10% of this is not known by anyone else. Software and training costs are not affected by Moore's law. Even if we assume human level language understanding and perfect sharing of knowledge, the training cost will be 1% to 10% of your working life to train the AGI to do your job. You are starting to appreciate what I have said on this forum several times, that if you do a realistic estimate of the software maintenance costs of an AGI population, that it approaches the efforts of the entire human race. In short, there is nothing to be gained by having AGIs, because while they may do our menial work, we will be working full time just to keep them running. Also, we have made *some* progress toward AGI since 1965, but it is mainly a better understanding of why it is so hard, I agree with this statement, though I arrived at this conclusion a little differently than you. I will now make some critical comments that indirectly support this same conclusion... e.g. - We know that general intelligence is not computable [2] or provable [3]. There is no neat theory. In recent decades, new forms of logic have emerged (e.g. Game Theory) that are NOT incremental improvements over general intelligence. Our evolutionary biological and social development has NOT allowed for this possibility, providing a HUGE opportunity for machine intelligence. The AGI goals of those here indirectly seek to dispense with this advantage while apparently seeking to construct new artificial mouths to feed. Hence, I see your statement to simply be further proof that AGI is a rather questionable goal compared with a more rigorous form of machine intelligence. - From Cyc, we know that coding common sense is more than a 20 year effort. Lenat doesn't know how much more, but guesses it is maybe between 0.1% and 10% finished. Of course, the BIG value is in UNcommon sense. For example, someone saying I had no choice but to... is really saying that they have constrained their thinking, probably to the point of failing to even recognize the existence of the optimal course of action. Common sense would be to accept the truth of their statement. UNcommon sense would be to recognize the error contained in their statement. - Google is the closest we have to AI after a half trillion dollar effort. You really should see my Dr. Eliza demo. Steve Richfield 1. Landauer, Tom (1986), How much do people remember? Some estimates of the quantity of learned information in long term memory, Cognitive Science (10) pp. 477-493. 2. Hutter, Marcus (2003), A Gentle Introduction to The Universal Algorithmic Agent {AIXI}, in *Artificial General Intelligence*, B. Goertzel and C. Pennachin eds., Springer.
Re: [agi] open models, closed models, priors
--- On Sun, 9/7/08, Bryan Bishop [EMAIL PROTECTED] wrote: On Thursday 04 September 2008, Matt Mahoney wrote: Yes you do. Every time you make a decision, you are assigning a higher probability of a good outcome to your choice than to the alternative. You'll have to prove to me that I make decisions, whatever that means. Depends on what you mean by I. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
[agi] Re: AI isn't cheap
--- On Sun, 9/7/08, Steve Richfield [EMAIL PROTECTED] wrote: 1. I believe that there is some VERY fertile but untilled ground, which if it is half as good as it looks, could yield AGI a LOT cheaper than other higher estimates. Of course if I am wrong, I would probably accept your numbers. 2. I believe that AGI will take VERY different (cheaper and more valuable) forms than do other members on this forum. Each of the above effects are worth several orders of magnitude in effort. You are just speculating. The fact is that thousands of very intelligent people have been trying to solve AI for the last 50 years, and most of them shared your optimism. Perhaps it would be more fruitful to estimate the cost of automating the global economy. I explained my estimate of 10^25 bits of memory, 10^26 OPS, 10^17 bits of software and 10^15 dollars. You really should see my Dr. Eliza demo. Perhaps you missed my comments in April. http://www.listbox.com/member/archive/303/2008/04/search/ZWxpemE/sort/time_rev/page/2/entry/5:53/20080414221142:407C652C-0A91-11DD-B3D2-6D4E66D9244B/ In any case, what does Dr. Eliza do that hasn't been done 30 years ago? -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] open models, closed models, priors
On Sunday 07 September 2008, Matt Mahoney wrote: Depends on what you mean by I. You started it - your first message had that dependency on identity. :-) - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] A NewMetaphor for Intelligence - the Computer/Organiser
Hey Bryan, To me, this is indistinguishable from the 1st option I laid out. Deterministic but impossible to predict. Terren --- On Sun, 9/7/08, Bryan Bishop [EMAIL PROTECTED] wrote: From: Bryan Bishop [EMAIL PROTECTED] Subject: Re: [agi] A NewMetaphor for Intelligence - the Computer/Organiser To: agi@v2.listbox.com Date: Sunday, September 7, 2008, 11:44 AM On Friday 05 September 2008, Terren Suydam wrote: So, Mike, is free will: 1) an illusion based on some kind of unpredictable, complex but *deterministic* interaction of physical components 2) the result of probabilistic physics - a *non-deterministic* interaction described by something like quantum mechanics 3) the expression of our god-given spirit, or some other non-physical mover of physical things I've already mentioned an alternative on this mailing list that you haven't included in your question, would you consider it? http://heybryan.org/free_will.html ^ Just so that I don't have to keep on rewriting it over and over again. - Bryan http://heybryan.org/ Engineers: http://heybryan.org/exp.html irc.freenode.net #hplusroadmap --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] open models, closed models, priors
--- On Sun, 9/7/08, Bryan Bishop [EMAIL PROTECTED] wrote: Depends on what you mean by I. You started it - your first message had that dependency on identity. :-) OK then. You decided to reply to my email, vs. not replying. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] draft for comment
Pei:As I said before, you give symbol a very narrow meaning, and insist that it is the only way to use it. In the current discussion, symbols are not 'X', 'Y', 'Z', but 'table', 'time', 'intelligence'. BTW, what images you associate with the latter two? Since you prefer to use person as example, let me try the same. All of my experience about 'Mike Tintner' is symbolic, nothing visual, but it still makes you real enough to me... I'm sorry if it sounds rude Pei, You attribute to symbols far too broad powers that they simply don't have - and demonstrably, scientifically, don't have. For example, you think that your experience of Mike Tintner - the rude guy - is entirely symbolic. Yes, all your experience of me has been mediated entirely via language/symbols -these posts. But by far the most important parts of it have actually been images. Ridiculous, huh? Look at this sentence: If you want to hear about it, you'll probably want to know where I was born, and what a lousy childhood I had, and how my parents were occupied before they had me, and all the David Copperfield crap, but if you want to know the truth, I don't really want to get into it. In 60 words, one of the great opening sentences of a novel, Salinger has created a whole character. How? He did it by creating a voice. He did it by what is called prosody (and also diction). No current AGI method has the least idea of how to process that prosody. But your brain does. Pei doesn't. But his/your brain does. And your experience of MT has been heavily based similarly on processing the *sound* images - the voice behind my words. Hence your I'm sorry if it *sounds* rude.. Words, even written words, aren't just symbols, they are sounds. And your brain hears those sounds and from their music can tell many, many things, including the emotions of the speaker, and whether they're being angry or ironic or rude. Now, if you had had more of a literary/arts education, you would probably be alive to that dimension. But, as it is, you've missed it, and you're missing all kinds of dimensions of how symbols work. Similarly, if you had more of a visual education, and also more of a psychological developmental background, you wouldn't find time and intelligence so daunting to visualise. You would realise that it takes a great deal of time and preparatory sensory/imaginative to build up abstract concepts You would realise that it takes time for an infant to come to use that word, and still more for a child to understand the word intelligence. I doubt that any child will understand time before they've seen a watch or clock, and that's what they will probably visualise time as, first. Your capacity to abstract time still further, will have come from having become gradually acquainted with a whole range of time-measuring devices, and seeing the word time and associating that with many other kinds of measurement especially in relation to maths. and science. Similarly, a person's concept of intelligence will come from seeing and hearing people solving problems in different ways - quickly and slowly, for example.. It will be deeply grounded in sensory images and experience. All the most abstract maths and logic that you may think totally abstract are similarly and necessarily grounded. Ben, in parallel to you, didn't realise that the decimal numeral system is digital, based on the hand, and so, a little less obviously, is the roman numeral system. Numbers and logic have to be built up out of experience. [You might profit BTW by looking at Barsalou, [many of his papers online], to see how the mind modally simulates concepts - with lots of experimental evidence] I, as you know, am very ignorant about computers; but you are also very ignorant about all kinds of dimensions of how symbols work, and intelligence generally, that are absolutely essential for AGI. You can continue to look down on me, or you can open your mind, recognize that general intelligence can only be achieved by a confluence of disciplines way beyond the reach of any single individual, and see that maybe useful exchanges can take place. --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] draft for comment
Mike, If you think your AGI know-how is superior to the know-how of those who already built testable thinking machines then why don't you try to build one yourself? Maybe you would learn more that way than when spending significant amount of time trying to sort out great incompatibilities between your views and views of the other AGI researchers. If you don't have resources to build the system then, perhaps, you could just put together some architecture doc (including your definitions of important terms) for your as-simple-as-possible AGI. The talk could then be more specific/interesting/fruitful for everyone involved. Sorry if I'm missing something. I'm reading this list only occasionally. But when I get to your posts, I often see things very differently and I know I'm not alone. I guess, if you try to view things from developers perspective + if you systematically move forward improving a particular AGI design, your views would change drastically. Just my opinion.. Regards, Jiri Jelinek --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
[agi] Philosophy of General Intelligence
Jiri: Mike, If you think your AGI know-how is superior to the know-how of those who already built testable thinking machines then why don't you try to build one yourself? Jiri, I don't think I know much at all about machines or software never claim to. I think I know certain, only certain, things about the psychological and philosophical aspects of general intelligence - esp. BTW about the things you guys almost never discuss, the kinds of problems that a general intelligence must solve. You may think that your objections to me are entirely personal about my manner. I suggest that there is also a v. deep difference of philosophy involved here. I believe that GI really is about *general* intelligence - a GI, and the only serious example we have is human, is, crucially, and must be, able to cross domains - ANY domain. That means the whole of our culture and society. It means every kind of representation, not just mathematical and logical and linguistic, but everything - visual, aural, solid, models, embodied etc etc. There is a vast range. That means also every subject domain - artistic, historical, scientific, philosophical, technological, politics, business etc. Yes, you have to start somewhere, but there should be no limit to how you progress. And the subject of general intelligence is tberefore, in no way, just the property of a small community of programmers, or roboticists - it's the property of all the sciences, incl. neuroscience, psychology, semiology, developmental psychology, AND the arts and philosophy etc. etc. And it can only be a collaborative effort. Some robotics disciplines, I believe, do think somewhat along those lines and align themselves with certain sciences. Some AI-ers also align themselves broadly with scientists and philosophers. By definition, too, general intelligence should embrace every kind of problem that humans have to deal with - again artistic, practical, technological, political, marketing etc. etc. The idea that general intelligence really could be anything else but truly general is, I suggest, if you really think about it, absurd. It's like preaching universal brotherhood, and a global society, and then practising severe racism. But that's exactly what's happening in current AGI. You're actually practising a highly specialised approach to AGI - only certain kinds of representation, only certain kinds of problems are considered - basically the ones you were taught and are comfortable with - a very, very narrow range - (to a great extent in line with the v. narrow definition of intelligence involved in the IQ test). When I raised other kinds of problems, Pei considered it not constructive. When I recently suggested an in fact brilliant game for producing creative metaphors, DZ considered it childish, because it was visual and imaginative, and you guys don't do those things, or barely. (Far from being childish, that game produced a rich series of visual/verbal metaphors, where AGI has produced nothing). If you aren't prepared to use your imagination and recognize the other half of the brain, you are, frankly, completely buggered as far as AGI is concerned. In over 2000 years, logic and mathematics haven't produced a single metaphor or analogy or crossed any domains. They're not meant to, that's expressly forbidden. But the arts produce metaphors and analogies on a daily basis by the thousands. The grand irony here is that creativity really is - from a strictly technical pov - largely what our culture has always said it is - imaginative/artistic and not rational.. (Many rational thinkers are creative - but by using their imagination). AGI will in fact only work if sciences and arts align. Here, then is basically why I think you're getting upset over and over by me. I'm saying in many different ways, general intelligence really should be general, and embrace the whole of culture and intelligence, not just the very narrow sections you guys espouse. And yes, I think you should be delighted to defer to, and learn from outsiders, (if they deserve it), just as I'm delighted to learn from you. But you're not - you resent outsiders like me telling you about your subject. I think you should also be prepared to admit your ignorance - and most of you, frankly, don't have much of a clue about imaginative/visual/artistic intelligence and vast swathes of problemsolving, ( just as I have don't have much of a clue re your technology and many kinds of problemsolving...etc). But there is v. little willingness to admit ignorance, or to acknowledge the value of other disciplines. IN the final analysis, I suggest, that's just sheer cultural prejudice. It doesn't belong in the new millennium when the defining paradigm is global (and general) as opposed to the local (and specialist) mentality of the old one - recognizing the value and interdependence of ALL parts of society and culture. And it doesn't
[agi] Bootris
--- snip --- [1220390007] receive [EMAIL PROTECTED] bootris, invoke mathematica [1220390013] told #love cool hand luke is like a comic heroic jesus [1220390034] receive [EMAIL PROTECTED] bootris, solve russell's paradox [1220390035] told #love invoke mathematica [1220390066] receive [EMAIL PROTECTED] he's invoking mathematica [1220390089] receive [EMAIL PROTECTED] he's invoking mathematica. bootris, solve russell's paradox [1220390090] told #love solve russell's paradox [1220390096] receive [EMAIL PROTECTED] he's invoking mathematica. bootris, solve russell's paradox. bootris, yes [1220390097] told #love Or make her laugh then tell her shes not good for when you say that like its going to learn islenska. --- snip --- Honestly it wasn't trivial getting to this stage --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] Philosophy of General Intelligence
Hi Mike, Good summary. I think your point of view is valuable in the sense of helping engineers in AGI to see what they may be missing. And your call for technical AI folks to take up the mantle of more artistic modes of intelligence is also important. But it's empty, for you've demonstrated no willingness to cross over to engage in technical arguments beyond a certain, quite limited, depth. Admitting your ignorance is one thing, and it's laudable, but it only goes so far. I think if you're serious about getting folks (like Pei Wang) to take you seriously, then you need to also demonstrate your willingness to get your hands dirty and do some programming, or in some other way abolish your ignorance about technical subjects - exactly what you're asking others to do. Otherwise, you have to admit the folly of trying to compel any such folks to move from their hard-earned perspectives, if you're not willing to do that yourself. Terren --- On Sun, 9/7/08, Mike Tintner [EMAIL PROTECTED] wrote: From: Mike Tintner [EMAIL PROTECTED] Subject: [agi] Philosophy of General Intelligence To: agi@v2.listbox.com Date: Sunday, September 7, 2008, 6:26 PM Jiri: Mike, If you think your AGI know-how is superior to the know-how of those who already built testable thinking machines then why don't you try to build one yourself? Jiri, I don't think I know much at all about machines or software never claim to. I think I know certain, only certain, things about the psychological and philosophical aspects of general intelligence - esp. BTW about the things you guys almost never discuss, the kinds of problems that a general intelligence must solve. You may think that your objections to me are entirely personal about my manner. I suggest that there is also a v. deep difference of philosophy involved here. I believe that GI really is about *general* intelligence - a GI, and the only serious example we have is human, is, crucially, and must be, able to cross domains - ANY domain. That means the whole of our culture and society. It means every kind of representation, not just mathematical and logical and linguistic, but everything - visual, aural, solid, models, embodied etc etc. There is a vast range. That means also every subject domain - artistic, historical, scientific, philosophical, technological, politics, business etc. Yes, you have to start somewhere, but there should be no limit to how you progress. And the subject of general intelligence is tberefore, in no way, just the property of a small community of programmers, or roboticists - it's the property of all the sciences, incl. neuroscience, psychology, semiology, developmental psychology, AND the arts and philosophy etc. etc. And it can only be a collaborative effort. Some robotics disciplines, I believe, do think somewhat along those lines and align themselves with certain sciences. Some AI-ers also align themselves broadly with scientists and philosophers. By definition, too, general intelligence should embrace every kind of problem that humans have to deal with - again artistic, practical, technological, political, marketing etc. etc. The idea that general intelligence really could be anything else but truly general is, I suggest, if you really think about it, absurd. It's like preaching universal brotherhood, and a global society, and then practising severe racism. But that's exactly what's happening in current AGI. You're actually practising a highly specialised approach to AGI - only certain kinds of representation, only certain kinds of problems are considered - basically the ones you were taught and are comfortable with - a very, very narrow range - (to a great extent in line with the v. narrow definition of intelligence involved in the IQ test). When I raised other kinds of problems, Pei considered it not constructive. When I recently suggested an in fact brilliant game for producing creative metaphors, DZ considered it childish, because it was visual and imaginative, and you guys don't do those things, or barely. (Far from being childish, that game produced a rich series of visual/verbal metaphors, where AGI has produced nothing). If you aren't prepared to use your imagination and recognize the other half of the brain, you are, frankly, completely buggered as far as AGI is concerned. In over 2000 years, logic and mathematics haven't produced a single metaphor or analogy or crossed any domains. They're not meant to, that's expressly forbidden. But the arts produce metaphors and analogies on a daily basis by the thousands. The grand irony here is that creativity really is - from a strictly technical pov - largely what our culture has always said it is - imaginative/artistic and not rational.. (Many rational thinkers are creative - but by using their imagination). AGI will in
[agi] Re: Bootris
One thing I think is kind of notable is that the bot puts everything it says, including phrases that are invented or mutated, into a personality database or list of possible favourite phrases, then takes six-axis mood assessments of follow-ups to its interjections, uses them to modify a mean score for the phrase, and prunes or clones it accordingly. This list can be searched a lot faster than the list of every unique phrase the bot has seen, and should statistically come to contain mostly phrases that make people like it. However, at 1GHz ConceptNet's mood assessment method is prohibitively slow... I haven't moved on to the context sensitivity and common-sense stuff that's in there. The natural-language module (ConceptNetNLTools) contains everything I'm using and seems to take over 100M in RAM alone. ConceptNetDB though seems to be worth opening up next. By using irclib with ConceptNet (both for Python) I can let the bot accrue a potentially unlimited database of up-to-date phrases, indexed by chronology and unique parts of speech, and from them extrapolate salient replies. Since the process is novelty-seeking, I think you'd reach a point where the training corpus ceases to expand except for current events and new terms. Whether this would take 4G or 40G of RAM I can't say yet, but the process obviously is not fast. The bot's heartbeat is incoming messages on the channels it's on, and it doesn't posess faculties for reflection or induction. By mimicking humans and watching the moods of people around it to assess its success and modify its behaviour, it ought to be able to pass as human without having most of the internal processes that characterize one... I don't know if there's a lesson here. Eric B On 9/7/08, Eric Burton [EMAIL PROTECTED] wrote: --- snip --- [1220390007] receive [EMAIL PROTECTED] bootris, invoke mathematica [1220390013] told #love cool hand luke is like a comic heroic jesus [1220390034] receive [EMAIL PROTECTED] bootris, solve russell's paradox [1220390035] told #love invoke mathematica [1220390066] receive [EMAIL PROTECTED] he's invoking mathematica [1220390089] receive [EMAIL PROTECTED] he's invoking mathematica. bootris, solve russell's paradox [1220390090] told #love solve russell's paradox [1220390096] receive [EMAIL PROTECTED] he's invoking mathematica. bootris, solve russell's paradox. bootris, yes [1220390097] told #love Or make her laugh then tell her shes not good for when you say that like its going to learn islenska. --- snip --- Honestly it wasn't trivial getting to this stage --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
[agi] Re: Bootris
Oh, thanks for helping me get this off my chest, everyone. If I ever finish the thing I'm definitely going to freshmeat it. I think this kind of bot, which is really quite trainable, and creative to boot -- it falls back to a markov chainer -- could be a shoe-in for naturalistic NPC dialogue in games. Just disable learning new phrases but keep some level of mood assessment and phrase mutation and it should functionally never become annoying. Obviously lacking real cognitive processes means that Bootris is not a general intelligence, but as an interactive curiousity who craves human acceptance/language data, he is a fair way to accrue a large corpus of online conversation for later mining and transforms. I will give an example of one use he's suited to today. With a cleaned-out markov cloud I took the bot to an IRC net populated by international botnet jockeys and their scanning/spamming bots. Within a minute or two the bot was making interjections to a dozen channels of two distinct natures... colour-coded replies like those from the bots, and commands to run scans of his own. Very disruptive! I almost put the code on sourceforge right away when I saw that happen, but it really was not finished. Ok, that's all. On 9/7/08, Eric Burton [EMAIL PROTECTED] wrote: One thing I think is kind of notable is that the bot puts everything it says, including phrases that are invented or mutated, into a personality database or list of possible favourite phrases, then takes six-axis mood assessments of follow-ups to its interjections, uses them to modify a mean score for the phrase, and prunes or clones it accordingly. This list can be searched a lot faster than the list of every unique phrase the bot has seen, and should statistically come to contain mostly phrases that make people like it. However, at 1GHz ConceptNet's mood assessment method is prohibitively slow... I haven't moved on to the context sensitivity and common-sense stuff that's in there. The natural-language module (ConceptNetNLTools) contains everything I'm using and seems to take over 100M in RAM alone. ConceptNetDB though seems to be worth opening up next. By using irclib with ConceptNet (both for Python) I can let the bot accrue a potentially unlimited database of up-to-date phrases, indexed by chronology and unique parts of speech, and from them extrapolate salient replies. Since the process is novelty-seeking, I think you'd reach a point where the training corpus ceases to expand except for current events and new terms. Whether this would take 4G or 40G of RAM I can't say yet, but the process obviously is not fast. The bot's heartbeat is incoming messages on the channels it's on, and it doesn't posess faculties for reflection or induction. By mimicking humans and watching the moods of people around it to assess its success and modify its behaviour, it ought to be able to pass as human without having most of the internal processes that characterize one... I don't know if there's a lesson here. Eric B On 9/7/08, Eric Burton [EMAIL PROTECTED] wrote: --- snip --- [1220390007] receive [EMAIL PROTECTED] bootris, invoke mathematica [1220390013] told #love cool hand luke is like a comic heroic jesus [1220390034] receive [EMAIL PROTECTED] bootris, solve russell's paradox [1220390035] told #love invoke mathematica [1220390066] receive [EMAIL PROTECTED] he's invoking mathematica [1220390089] receive [EMAIL PROTECTED] he's invoking mathematica. bootris, solve russell's paradox [1220390090] told #love solve russell's paradox [1220390096] receive [EMAIL PROTECTED] he's invoking mathematica. bootris, solve russell's paradox. bootris, yes [1220390097] told #love Or make her laugh then tell her shes not good for when you say that like its going to learn islenska. --- snip --- Honestly it wasn't trivial getting to this stage --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
[agi] Re: Bootris
(see: irc.racrew.us) On 9/7/08, Eric Burton [EMAIL PROTECTED] wrote: Oh, thanks for helping me get this off my chest, everyone. If I ever finish the thing I'm definitely going to freshmeat it. I think this kind of bot, which is really quite trainable, and creative to boot -- it falls back to a markov chainer -- could be a shoe-in for naturalistic NPC dialogue in games. Just disable learning new phrases but keep some level of mood assessment and phrase mutation and it should functionally never become annoying. Obviously lacking real cognitive processes means that Bootris is not a general intelligence, but as an interactive curiousity who craves human acceptance/language data, he is a fair way to accrue a large corpus of online conversation for later mining and transforms. I will give an example of one use he's suited to today. With a cleaned-out markov cloud I took the bot to an IRC net populated by international botnet jockeys and their scanning/spamming bots. Within a minute or two the bot was making interjections to a dozen channels of two distinct natures... colour-coded replies like those from the bots, and commands to run scans of his own. Very disruptive! I almost put the code on sourceforge right away when I saw that happen, but it really was not finished. Ok, that's all. On 9/7/08, Eric Burton [EMAIL PROTECTED] wrote: One thing I think is kind of notable is that the bot puts everything it says, including phrases that are invented or mutated, into a personality database or list of possible favourite phrases, then takes six-axis mood assessments of follow-ups to its interjections, uses them to modify a mean score for the phrase, and prunes or clones it accordingly. This list can be searched a lot faster than the list of every unique phrase the bot has seen, and should statistically come to contain mostly phrases that make people like it. However, at 1GHz ConceptNet's mood assessment method is prohibitively slow... I haven't moved on to the context sensitivity and common-sense stuff that's in there. The natural-language module (ConceptNetNLTools) contains everything I'm using and seems to take over 100M in RAM alone. ConceptNetDB though seems to be worth opening up next. By using irclib with ConceptNet (both for Python) I can let the bot accrue a potentially unlimited database of up-to-date phrases, indexed by chronology and unique parts of speech, and from them extrapolate salient replies. Since the process is novelty-seeking, I think you'd reach a point where the training corpus ceases to expand except for current events and new terms. Whether this would take 4G or 40G of RAM I can't say yet, but the process obviously is not fast. The bot's heartbeat is incoming messages on the channels it's on, and it doesn't posess faculties for reflection or induction. By mimicking humans and watching the moods of people around it to assess its success and modify its behaviour, it ought to be able to pass as human without having most of the internal processes that characterize one... I don't know if there's a lesson here. Eric B On 9/7/08, Eric Burton [EMAIL PROTECTED] wrote: --- snip --- [1220390007] receive [EMAIL PROTECTED] bootris, invoke mathematica [1220390013] told #love cool hand luke is like a comic heroic jesus [1220390034] receive [EMAIL PROTECTED] bootris, solve russell's paradox [1220390035] told #love invoke mathematica [1220390066] receive [EMAIL PROTECTED] he's invoking mathematica [1220390089] receive [EMAIL PROTECTED] he's invoking mathematica. bootris, solve russell's paradox [1220390090] told #love solve russell's paradox [1220390096] receive [EMAIL PROTECTED] he's invoking mathematica. bootris, solve russell's paradox. bootris, yes [1220390097] told #love Or make her laugh then tell her shes not good for when you say that like its going to learn islenska. --- snip --- Honestly it wasn't trivial getting to this stage --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] Philosophy of General Intelligence
Terren, You may be right - in the sense that I would have to just butt out of certain conversations, to go away educate myself. There's just one thing here though - and again this is a central philosophical difference this time concerning the creative process. Can you tell me which kind of programming is necessary for which end-problem[s] that general intelligence must solve? Which kind of programming, IOW, can you *guarantee* me will definitely not be a waste of my time (other than by way of general education) ? Which kind are you *sure* will help solve which unsolved problem of AGI? P.S. OTOH the idea that in the kind of general community I'm espousing, (and is beginning to crop up in other areas), everyone must be proficient in everyone else's speciality is actually a non-starter, Terren. It defeats the object of the division of labour central to all parts of the economy. If you had to spend as much time thinking about those end-problems as I have, I suggest you'd have to drop everything. Let's just share expertise instead? Terren: Good summary. I think your point of view is valuable in the sense of helping engineers in AGI to see what they may be missing. And your call for technical AI folks to take up the mantle of more artistic modes of intelligence is also important. But it's empty, for you've demonstrated no willingness to cross over to engage in technical arguments beyond a certain, quite limited, depth. Admitting your ignorance is one thing, and it's laudable, but it only goes so far. I think if you're serious about getting folks (like Pei Wang) to take you seriously, then you need to also demonstrate your willingness to get your hands dirty and do some programming, or in some other way abolish your ignorance about technical subjects - exactly what you're asking others to do. Otherwise, you have to admit the folly of trying to compel any such folks to move from their hard-earned perspectives, if you're not willing to do that yourself. Terren --- On Sun, 9/7/08, Mike Tintner [EMAIL PROTECTED] wrote: From: Mike Tintner [EMAIL PROTECTED] Subject: [agi] Philosophy of General Intelligence To: agi@v2.listbox.com Date: Sunday, September 7, 2008, 6:26 PM Jiri: Mike, If you think your AGI know-how is superior to the know-how of those who already built testable thinking machines then why don't you try to build one yourself? Jiri, I don't think I know much at all about machines or software never claim to. I think I know certain, only certain, things about the psychological and philosophical aspects of general intelligence - esp. BTW about the things you guys almost never discuss, the kinds of problems that a general intelligence must solve. You may think that your objections to me are entirely personal about my manner. I suggest that there is also a v. deep difference of philosophy involved here. I believe that GI really is about *general* intelligence - a GI, and the only serious example we have is human, is, crucially, and must be, able to cross domains - ANY domain. That means the whole of our culture and society. It means every kind of representation, not just mathematical and logical and linguistic, but everything - visual, aural, solid, models, embodied etc etc. There is a vast range. That means also every subject domain - artistic, historical, scientific, philosophical, technological, politics, business etc. Yes, you have to start somewhere, but there should be no limit to how you progress. And the subject of general intelligence is tberefore, in no way, just the property of a small community of programmers, or roboticists - it's the property of all the sciences, incl. neuroscience, psychology, semiology, developmental psychology, AND the arts and philosophy etc. etc. And it can only be a collaborative effort. Some robotics disciplines, I believe, do think somewhat along those lines and align themselves with certain sciences. Some AI-ers also align themselves broadly with scientists and philosophers. By definition, too, general intelligence should embrace every kind of problem that humans have to deal with - again artistic, practical, technological, political, marketing etc. etc. The idea that general intelligence really could be anything else but truly general is, I suggest, if you really think about it, absurd. It's like preaching universal brotherhood, and a global society, and then practising severe racism. But that's exactly what's happening in current AGI. You're actually practising a highly specialised approach to AGI - only certain kinds of representation, only certain kinds of problems are considered - basically the ones you were taught and are comfortable with - a very, very narrow range - (to a great extent in line with the v. narrow definition of intelligence involved in the IQ test). When I raised other kinds of problems, Pei considered it not constructive. When I recently suggested an in fact brilliant game for producing
[agi] Does prior knowledge/learning cause GAs to converge too fast on sub-optimal solutions?
Hi, I have a general question for those (such as Novamente) working on AGI systems that use genetic algorithms as part of their search strategy. A GA researcher recently explained to me some of his experiments in embedding prior knowledge into systems. For example, when attempting to automate the discovery of models of a mechanical system, they tried adding some textbook models to the set of genetic operators. The results weren't good - the prior knowledge worked too well, causing the GA to converge too fast onto the prior knowledge. so fast that there wasn't time for the GA to build up sufficient diversity and quality in other solutions that might have helped get out of the local maxima. The message seemed to be that prior knowledge is too powerful - it can 'blind' a search - and that if you must use it, you'd have to very very aggressively artificially deflate the fitness of instances that use prior knowledge (and this is tricky to get right). This struck me as relevant to GA-based AGIs that continually build on and improve a knowledge-base. Once an AGI learns very simple initial models of the world, if it then tries to evolve deeper knowledge about more difficult problems (but, in the context of its prior learning), then its initial models may prove to be too good: forcing the GA to converge on poor local maxima that represent only minor variations on the initial models it learnt in its earliest days. Does this issue actually crop up in GA-based AGI work? If so, how did you get around it? If not, would you have any comments about what makes AGI special so that this doesn't happen? -Ben --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] Does prior knowledge/learning cause GAs to converge too fast on sub-optimal solutions?
I'd just keep a long list of high scorers for regression and occasionally reset the high score to zero. You can add random specimens to the population as well... On 9/7/08, Benjamin Johnston [EMAIL PROTECTED] wrote: Hi, I have a general question for those (such as Novamente) working on AGI systems that use genetic algorithms as part of their search strategy. A GA researcher recently explained to me some of his experiments in embedding prior knowledge into systems. For example, when attempting to automate the discovery of models of a mechanical system, they tried adding some textbook models to the set of genetic operators. The results weren't good - the prior knowledge worked too well, causing the GA to converge too fast onto the prior knowledge. so fast that there wasn't time for the GA to build up sufficient diversity and quality in other solutions that might have helped get out of the local maxima. The message seemed to be that prior knowledge is too powerful - it can 'blind' a search - and that if you must use it, you'd have to very very aggressively artificially deflate the fitness of instances that use prior knowledge (and this is tricky to get right). This struck me as relevant to GA-based AGIs that continually build on and improve a knowledge-base. Once an AGI learns very simple initial models of the world, if it then tries to evolve deeper knowledge about more difficult problems (but, in the context of its prior learning), then its initial models may prove to be too good: forcing the GA to converge on poor local maxima that represent only minor variations on the initial models it learnt in its earliest days. Does this issue actually crop up in GA-based AGI work? If so, how did you get around it? If not, would you have any comments about what makes AGI special so that this doesn't happen? -Ben --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
Re: [agi] Philosophy of General Intelligence
Hi Mike, It's not so much the *kind* of programming that I or anyone else could recommend, it's just the general skill of programming - getting used to thinking in terms of, how exactly do I solve this problem - what model or procedure do I create? How do you specify something so completely and precisely that a mindless machine can execute it? It's not just that, it's also understanding how the written specification (the program) translates into actions at the processor level. That's important too. Obviously having these skills and knowledge is not the answer to creating AGI - if it was, it'd have been solved decades ago. But without understanding how computers work, and how we make them work for us, it is too easy to fall into the trap of mistaking a computer's operation in terms of some kind of homunculus, or that it has a will of its own, or some other kind of anthropic confusion. If you don't understand how to program a computer, you will be tempted to say that a chess program that can beat Gary Kasparov is intelligent. Your repeated appeals to creating programs that can decide for themselves without specifying what they do underscores your technical weakness, because programs are nothing but exact specifications. You make good points about what General Intelligence entails, but if you had a solid grasp of the technical aspects of computing, you could develop your philosophy so much further. Matt Mahoney's suggestion of trying to create an Artificial Artist is a great example of a direction that is closed to you until you learn the things I'm talking about. Terren in response to your PS: I'm not suggesting everyone be proficient at everything, although such folks are extremely valuable... why not become one? Anyway, sharing expertise is all well and good but in order to do so, you have to give ground to the experts - something I haven't seen you do. You seem (to me) to be quite attached to your viewpoint, even regarding topics that you admit ignorance to. Am I wrong? --- On Sun, 9/7/08, Mike Tintner [EMAIL PROTECTED] wrote: Can you tell me which kind of programming is necessary for which end-problem[s] that general intelligence must solve? Which kind of programming, IOW, can you *guarantee* me will definitely not be a waste of my time (other than by way of general education) ? Which kind are you *sure* will help solve which unsolved problem of AGI? P.S. OTOH the idea that in the kind of general community I'm espousing, (and is beginning to crop up in other areas), everyone must be proficient in everyone else's speciality is actually a non-starter, Terren. It defeats the object of the division of labour central to all parts of the economy. If you had to spend as much time thinking about those end-problems as I have, I suggest you'd have to drop everything. Let's just share expertise instead? Terren: Good summary. I think your point of view is valuable in the sense of helping engineers in AGI to see what they may be missing. And your call for technical AI folks to take up the mantle of more artistic modes of intelligence is also important. But it's empty, for you've demonstrated no willingness to cross over to engage in technical arguments beyond a certain, quite limited, depth. Admitting your ignorance is one thing, and it's laudable, but it only goes so far. I think if you're serious about getting folks (like Pei Wang) to take you seriously, then you need to also demonstrate your willingness to get your hands dirty and do some programming, or in some other way abolish your ignorance about technical subjects - exactly what you're asking others to do. Otherwise, you have to admit the folly of trying to compel any such folks to move from their hard-earned perspectives, if you're not willing to do that yourself. Terren --- On Sun, 9/7/08, Mike Tintner [EMAIL PROTECTED] wrote: From: Mike Tintner [EMAIL PROTECTED] Subject: [agi] Philosophy of General Intelligence To: agi@v2.listbox.com Date: Sunday, September 7, 2008, 6:26 PM Jiri: Mike, If you think your AGI know-how is superior to the know-how of those who already built testable thinking machines then why don't you try to build one yourself? Jiri, I don't think I know much at all about machines or software never claim to. I think I know certain, only certain, things about the psychological and philosophical aspects of general intelligence - esp. BTW about the things you guys almost never discuss, the kinds of problems that a general intelligence must solve. You may think that your objections to me are entirely personal about my manner. I suggest that there is also a v. deep difference of philosophy involved here. I believe that GI really is about *general* intelligence - a GI, and the only serious example we have is human, is, crucially,
Re: [agi] Philosophy of General Intelligence
Mike, every kind of representation, not just mathematical and logical and linguistic, but everything - visual, aural, solid, models, embodied etc etc. There is a vast range. That means also every subject domain - artistic, historical, scientific, philosophical, technological, politics, business etc Developers need to find a way how to represent data we get through our senses, but that does not necessarily mean that for example audio data need to be perceived as audio in order to be useful for general problem solving. the subject of general intelligence is tberefore, in no way, just the property of a small community of programmers, or roboticists - it's the property of all the sciences, incl. neuroscience, psychology, semiology, developmental psychology, AND the arts and philosophy etc. etc. And it can only be a collaborative effort. For teaching AGI - it's good to get experts from *many* domains. For design development - experts from few domains are IMO good enough. The idea that general intelligence really could be anything else but truly general is, I suggest, if you really think about it, absurd. It's like preaching universal brotherhood, and a global society, and then practising severe racism. View GI as powerful problem solving or so and move on. How well the system solves problems - that's what counts (not how it's labeled endless arguing about GI definitions). You're actually practising a highly specialised approach to AGI - only certain kinds of representation, only certain kinds of problems are considered - basically the ones you were taught and are comfortable with - a very, very narrow range As I mentioned before, the representation needs to reflect what we get through senses and the practical approaches need to be based on available technology. If you think you have breakthrough ideas then be specific. When I recently suggested an in fact brilliant game for producing creative metaphors, DZ considered it childish, I did not read that one.. cannot comment on it [now].. In over 2000 years, logic and mathematics haven't produced a single metaphor or analogy or crossed any domains. false But the arts produce metaphors and analogies on a daily basis by the thousands. That certainly can be coded. general intelligence really should be general, and embrace the whole of culture and intelligence, not just the very narrow sections you guys espouse. Many of us here are thinking hard about how to develop non-narrow AI. most of you, frankly, don't have much of a clue about imaginative/visual/artistic intelligence I understand that a particular problem solving may require nD model(s), but can you please give me an example of a problem solved by artistic intelligence that could not be solved by non-artistic intelligence? and then just possibly you won't find me quite so upsetting I'm not upset. In these days, I'm getting here to relax ;-). Regards, Jiri --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com