Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
Although I thought this was a good talk and I liked the fellow presenting it to me it seems fairly clear that little or no progress has been made in this area over the last decade or so. In the early 1990s I wrote somewhat similar simulations where agents had their own neural networks whose architecture was specified by a genetic algorithm, but just like the speaker I came up against similar problems. As the guy says it should be in principle possible to go all the way from simple types of creatures up to more complex ones, like humans. In practice though what tends to happen is that the complexity of the neural nets reaches a plateau from which little subsequent progress occurs. Even after allowing the system to run for tens of thousands of generations not much of interest happens. I think the main problem here is the low complexity of the environment and the agents themselves. In a real biological system there are all kinds of niches which can be exploited in a variety of ways, but in polyworld (and other similar simulations) it's all very homogeneous. Real biological creatures are coalitions of millions of cells, each of which is a chemical factory containing an abundance of nano machinery, each of which is a possible site for evolutionary change. The sensory systems of real creatures are also far richer than simply being able to detect three colours (even molluscs can do better than this), and this is obviously a limiting factor upon the development of greater intelligence. On 15/11/2007, Jef Allbright [EMAIL PROTECTED] wrote: This may be of interest to the group. http://video.google.com/videoplay?docid=-112735133685472483 This presentation is about a potential shortcut to artificial intelligence by trading mind-design for world-design using artificial evolution. Evolutionary algorithms are a pump for turning CPU cycles into brain designs. With exponentially increasing CPU cycles while our understanding of intelligence is almost a flat-line, the evolutionary route to AI is a centerpiece of most Kurzweilian singularity scenarios. This talk introduces the Polyworld artificial life simulator as well as results from our ongoing attempt to evolve artificial intelligence and further the Singularity. Polyworld is the brain child of Apple Computer Distinguished Scientist Larry Yaeger, who remains the primary developer of Polyworld: http://www.beanblossom.in.us/larryy/P... Speaker: Virgil Griffith Virgil Griffith is a first year graduate student in Computation and Neural Systems at the California Institute of Technology. On weekdays he studies evolution, computational neuroscience, and artificial life. He did computer security work until his first year of university when his work got him sued for sedition and espionage. He then decided that security was probably not safest field to be in and he turned his life to science. (less) Added: November 13, 2007 - Jef - 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=8660244id_secret=65298881-4c0739
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
Yes, resulted behaviors are not impressive, I did similar thing with essentially 1 hidden layer perceptron on 2D square grid in high school and got something that looked not much simpler (weak creatures cycling around gathering, fat carnivores in the center hunting them, few superfat parasites among carnivores vampiring off them). I think such environment needs system of cues that are useful (for survival) to be known during creature's lifetime, which are changing and getting more complex across generations. This way there would be an incentive to develop memory and nontrivial decision making from observations. As it is, it's not clear how even trained human would do given those limited perceptions. On 11/15/07, Bob Mottram [EMAIL PROTECTED] wrote: Although I thought this was a good talk and I liked the fellow presenting it to me it seems fairly clear that little or no progress has been made in this area over the last decade or so. In the early 1990s I wrote somewhat similar simulations where agents had their own neural networks whose architecture was specified by a genetic algorithm, but just like the speaker I came up against similar problems. As the guy says it should be in principle possible to go all the way from simple types of creatures up to more complex ones, like humans. In practice though what tends to happen is that the complexity of the neural nets reaches a plateau from which little subsequent progress occurs. Even after allowing the system to run for tens of thousands of generations not much of interest happens. I think the main problem here is the low complexity of the environment and the agents themselves. In a real biological system there are all kinds of niches which can be exploited in a variety of ways, but in polyworld (and other similar simulations) it's all very homogeneous. Real biological creatures are coalitions of millions of cells, each of which is a chemical factory containing an abundance of nano machinery, each of which is a possible site for evolutionary change. The sensory systems of real creatures are also far richer than simply being able to detect three colours (even molluscs can do better than this), and this is obviously a limiting factor upon the development of greater intelligence. On 15/11/2007, Jef Allbright [EMAIL PROTECTED] wrote: This may be of interest to the group. http://video.google.com/videoplay?docid=-112735133685472483 This presentation is about a potential shortcut to artificial intelligence by trading mind-design for world-design using artificial evolution. Evolutionary algorithms are a pump for turning CPU cycles into brain designs. With exponentially increasing CPU cycles while our understanding of intelligence is almost a flat-line, the evolutionary route to AI is a centerpiece of most Kurzweilian singularity scenarios. This talk introduces the Polyworld artificial life simulator as well as results from our ongoing attempt to evolve artificial intelligence and further the Singularity. Polyworld is the brain child of Apple Computer Distinguished Scientist Larry Yaeger, who remains the primary developer of Polyworld: http://www.beanblossom.in.us/larryy/P... Speaker: Virgil Griffith Virgil Griffith is a first year graduate student in Computation and Neural Systems at the California Institute of Technology. On weekdays he studies evolution, computational neuroscience, and artificial life. He did computer security work until his first year of university when his work got him sued for sedition and espionage. He then decided that security was probably not safest field to be in and he turned his life to science. (less) Added: November 13, 2007 - Jef - 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/?; -- 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=8660244id_secret=65300965-1c5cc1
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
On Thursday 15 November 2007 02:30, Bob Mottram wrote: I think the main problem here is the low complexity of the environment Complex programs can only be written in an environment capable of bearing that complexity: http://sl4.org/archive/0710/16880.html - Bryan - 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=8660244id_secret=65328088-2120d3
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
On Nov 15, 2007 2:16 PM, Benjamin Goertzel [EMAIL PROTECTED] wrote: I remember playing with PolyWorld 10 years ago or so Yeah. I've only had time to watch the first 20 minutes of that talk but my reaction so far is disappointment: it's just exactly the same as it was a decade ago? Modern hardware should be able to do better. (Correct me if advances are presented in the later part of the talk.) Overall, I came away from my flirtation with Alife with the impression that it was doomed due to the lack of a viable artificial chemistry (chemistry arguably being the source of the richness of real biology). The closest I've ever seen to artificial chemistry is an experiment I did some years ago in evolving Go-playing programs; I didn't get anything that used strategy as humans or even hand-written programs understand it, but it had the paper-scissors-stone _feel_ of biochemistry, which makes sense in hindsight: Go is rich enough to support something on that level. Though I think physics - of the ordinary everyday variety - is the biggest missing element if one is trying to get animal-type intelligence out of a Polyworld-type environment. Use modern graphics hardware, give the simulated critters a 512x512 or somesuch camera view, make simulated bodies with a decently large number of degrees of freedom and contact sensors, a brain specified by general computation and big enough to do something with all those inputs and outputs, and I think you could get something a lot further on the road to an artificial lizard than has been produced thus far. And then I decided Alife was not gonna be a shortcut and turned wholly to AI insetad ;-) Same here :) - 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=8660244id_secret=65342185-3e2eea
Re: Essay - example of how the CSP bites [WAS Re: [agi] What best evidence for fast AI?]
Mike Tintner wrote: Sounds a little confusing. Sounds like you plan to evolve a system through testing thousands of candidate mechanisms. So one way or another you too are taking a view - even if it's an evolutionary, I'm not taking a view view - on, and making a lot of asssumptions about -how systems evolve -the known architecture of human cognition. No, I think because of the paucity of information I gave you have misunderstood slightly. Everything I mentioned was in the context of an extremely detailed framework that tries to include all of the knowledge we have so far gleaned by studying human cognition using the methods of cognitive science. So I am not making assumptions about the architecture of human cognition I am using every scrap of experimental data I can. You can say that this is still assuming that the framework is correct, but that is nothing compared to the usual assumptions made in AI, where the programmer just picks up a grab bag of assorted ideas that are floating around in the literature (none of them part of a coherent theory of cognition) and starts hacking. And just because I talk of thousands of candidate mechanisms, that does not mean that there is evolution involved: it just means that even with a complete framework for human cognition to start from there are still so many questions about the low-level to high-level linkage that a vast number of mechanisms have to be explored. about which science has extremely patchy and confused knowledge. I don't see how any system-builder can avoid taking a view of some kind on such matters, yet you seem to be criticising Ben for so doing. Ben does not start from a complete framework for human cognition, nor does he feel compelled to stick close to the human model, and my criticisms (at least in this instance) are not really about whether or not he has such a framework, but about a problem that I can see on his horizon. I was hoping that you also had some view on how a system 's symbols should be grounded, especially since you mention Harnad, who does make vague gestures towards the brain's levels of grounding. But you don't indicate any such view. On the contrary, I explained exactly how they would be grounded: if the system is allowed to build its own symbols *without* me also inserting ungrounded (i.e. interpreted, programmer-constructed) symbols and messing the system up by forcing it to use both sorts of symbols, then ipso fact it is grounded. It is easy to build a grounded system. The trick is to make it both grounded and intelligent at the same time. I have one strategy for ensuring that it turns out intelligent, and Ben has another my problem with Ben's strategy is that I believe his attempt to ensure that the system is intelligent ends up compromising the groundedness of the system. Sounds like you too, pace MW, are hoping for a number of miracles - IOW creative ideas - to emerge, and make your system work. I don't understand where I implied this. You have to remember that I am doing this within a particular strategy (outlined in my CSP paper). When you see me exploring 'thousands' of candidate mechanisms to see how one parameter plays a role, this is not waiting for a miracle, it is a vital part of the strategy. A strategy that, I claim, is the only viable one. Anyway, you have to give Ben credit for putting a lot of his stuff principles out there on the line. I think anyone who wants to mount a full-scale assault on him ( why not?) should be prepared to reciprocate. Nice try, but there are limits to what I can do to expose the details. I have not yet worked out how much I should release and how much to withhold (I confess, I nearly decided to go completely public a month or so ago, but then changed my mind after seeing the dismally poor response that even one of the ideas provoked). Maybe in the near future I will write a summary account. In the mean time, yes, it is a little unfair of me to criticise other projects. But not that unfair. When a scientist sees a big problem with a theory, do you suppose they wait until they have a completely worked out alternative before discussing the fact that there is a problem with the theory that other people may be praising? That is not the way of science. 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=8660244id_secret=65349870-56ef76
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
About PolyWorld and Alife in general... I remember playing with PolyWorld 10 years ago or so And, I had a grad student at Uni. of Western Australia build a similar system, back in my Perth days... (it was called SEE, for Simple Evolving Ecology. We never published anything on it, as I left Australia in the middle of the research...) But after fiddling with stuff like this a while, it becomes clear that, just as each GOFAI or machine learning program can be pushed so far and no further; similarly each Alife program can be pushed so far and no further... One of the most fascinatng busts in that area was Tom Ray's attempt to induce robust virtual evolution of multicellular life. I forget the name of his project but he was doing it at ATR in Japan. It was a follow-up to his excellently successful Tierra program, which was the first to demonstrate biology-like reproduction in artificial organisms Anyway Tom's attempt and many others to to beyond the complexity threshold observed in Alife programs did not pan out... Overall, I came away from my flirtation with Alife with the impression that it was doomed due to the lack of a viable artificial chemistry (chemistry arguably being the source of the richness of real biology). So, there was some cool work on artificial chemistry of a sort, done by Walter Fontana and many others, which I don't remember very well... The deep question I came away with was: What exactly are the **abstract properties** of the periodic table of elements that allows it to give rise to chemical compounds and ensuing biological structures with so much complexity? And then I decided Alife was not gonna be a shortcut and turned wholly to AI insetad ;-) Thing is, I'm sure Alife can work, but the computational requirements have gotta be way way bigger than for AI. And conceptually, it doesn't seem like Alife is really a shortcut -- because puzzling out the requirements that artificial chemistry needs to have, in order to support robust artificial biology, seems just as hard or harder than building a simulated brain or a non-brain-based AGI. After all it's not like we know how real chemistry gives rise to real biology yet --- the dynamics underlying protein-folding remain ill-understood, etc. etc. So I find this a deep and fascinating area of research (the borderline btw artificial chemistry and artificial biology, more so than Alife proper), but I doubt it's a shortcut to AGI ... though it would be cool to be proven wrong ;-) -- Ben G On Nov 15, 2007 3:30 AM, Bob Mottram [EMAIL PROTECTED] wrote: Although I thought this was a good talk and I liked the fellow presenting it to me it seems fairly clear that little or no progress has been made in this area over the last decade or so. In the early 1990s I wrote somewhat similar simulations where agents had their own neural networks whose architecture was specified by a genetic algorithm, but just like the speaker I came up against similar problems. As the guy says it should be in principle possible to go all the way from simple types of creatures up to more complex ones, like humans. In practice though what tends to happen is that the complexity of the neural nets reaches a plateau from which little subsequent progress occurs. Even after allowing the system to run for tens of thousands of generations not much of interest happens. I think the main problem here is the low complexity of the environment and the agents themselves. In a real biological system there are all kinds of niches which can be exploited in a variety of ways, but in polyworld (and other similar simulations) it's all very homogeneous. Real biological creatures are coalitions of millions of cells, each of which is a chemical factory containing an abundance of nano machinery, each of which is a possible site for evolutionary change. The sensory systems of real creatures are also far richer than simply being able to detect three colours (even molluscs can do better than this), and this is obviously a limiting factor upon the development of greater intelligence. On 15/11/2007, Jef Allbright [EMAIL PROTECTED] wrote: This may be of interest to the group. http://video.google.com/videoplay?docid=-112735133685472483 This presentation is about a potential shortcut to artificial intelligence by trading mind-design for world-design using artificial evolution. Evolutionary algorithms are a pump for turning CPU cycles into brain designs. With exponentially increasing CPU cycles while our understanding of intelligence is almost a flat-line, the evolutionary route to AI is a centerpiece of most Kurzweilian singularity scenarios. This talk introduces the Polyworld artificial life simulator as well as results from our ongoing attempt to evolve artificial intelligence and further the Singularity. Polyworld is the brain child of Apple Computer Distinguished Scientist Larry Yaeger, who remains the primary
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
I think that linguistic interaction with human beings is going to be what lifts Second Life proto-AGI's beyond the glass ceiling... Our first SL agents won't have language generation or language learning capability, but I think that introducing it is really essential, esp. given the limitations of SL as a purely physical environment... ben On Nov 15, 2007 1:38 PM, Bob Mottram [EMAIL PROTECTED] wrote: Which raises the question of whether the same complexity glass ceiling will be encountered when running AGI controlled agents within Second Life. SL is probably more complex than polyworld, although that could be debatable depending upon your definition of complexity. One factor which would raise the bar would be the additional baggage being introduced into the virtual world from the first life of human participants. On 15/11/2007, Bryan Bishop [EMAIL PROTECTED] wrote: On Thursday 15 November 2007 02:30, Bob Mottram wrote: I think the main problem here is the low complexity of the environment Complex programs can only be written in an environment capable of bearing that complexity: http://sl4.org/archive/0710/16880.html - Bryan - 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/?; - 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=8660244id_secret=65511033-66e22b
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
Which raises the question of whether the same complexity glass ceiling will be encountered when running AGI controlled agents within Second Life. SL is probably more complex than polyworld, although that could be debatable depending upon your definition of complexity. One factor which would raise the bar would be the additional baggage being introduced into the virtual world from the first life of human participants. On 15/11/2007, Bryan Bishop [EMAIL PROTECTED] wrote: On Thursday 15 November 2007 02:30, Bob Mottram wrote: I think the main problem here is the low complexity of the environment Complex programs can only be written in an environment capable of bearing that complexity: http://sl4.org/archive/0710/16880.html - Bryan - 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=8660244id_secret=65507313-967074
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
On Nov 15, 2007 8:57 PM, Bryan Bishop [EMAIL PROTECTED] wrote: On Thursday 15 November 2007 08:16, Benjamin Goertzel wrote: non-brain-based AGI. After all it's not like we know how real chemistry gives rise to real biology yet --- the dynamics underlying protein-folding remain ill-understood, etc. etc. Can anybody elaborate on the actual problems remaining (beyond etc. etc.-- which is appropriate from Ben who is most notably not a biochemist/chemist/bioinformatician)? Hey -- That is a funny comment -- I've published a dozen bioinformatics papers in the last 5 years, and am CEO / Chief Scientist of a bioinformatics company (Biomind LLC, www.biomind.com) I am no chemist but I'm pretty much an expert on analyzing microarray and SNP data, and various other corners of bioinformatics, having introduced some funky new techniques into the field. In fact my most popular research paper is not on AGI but rather on Chronic Fatigue Syndrome -- it was the first-ever paper giving evidence for a (weak) genetic basis for CFS. -- Ben G - 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=8660244id_secret=65715822-29017b
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
No worries!! just wanted to clarify... To address your question more usefully: There is soo much evidence that chemistry is subtly important for biology in ways that are poorly understood. In neuroscience for instance the chemistry of synaptic transmission btw neurons is still weakly understood, so we still don't know exactly how poor a model the formal neuron used in computer science is As a single example you have both ionotropic and metabotropic glutamate receptors along neurons ... whose synaptic transmission properties depend on ambient chemistry in the intracellular medium in ways no one understands really.. etc. etc. etc. ;-) ben On Nov 15, 2007 10:07 PM, Bryan Bishop [EMAIL PROTECTED] wrote: On Thursday 15 November 2007 20:02, Benjamin Goertzel wrote: On Nov 15, 2007 8:57 PM, Bryan Bishop [EMAIL PROTECTED] wrote: Can anybody elaborate on the actual problems remaining (beyond etc. etc.-- which is appropriate from Ben who is most notably not a biochemist/chemist/bioinformatician)? Hey -- That is a funny comment Oh my. This is a big, big mistake on my part. I am sorry. Please accept my apologies .. and the knowledge that my parenthetical comment no longer applies. - Bryan - 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=8660244id_secret=65726263-86dc00
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
On Thursday 15 November 2007 20:02, Benjamin Goertzel wrote: On Nov 15, 2007 8:57 PM, Bryan Bishop [EMAIL PROTECTED] wrote: Can anybody elaborate on the actual problems remaining (beyond etc. etc.-- which is appropriate from Ben who is most notably not a biochemist/chemist/bioinformatician)? Hey -- That is a funny comment Oh my. This is a big, big mistake on my part. I am sorry. Please accept my apologies .. and the knowledge that my parenthetical comment no longer applies. - Bryan - 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=8660244id_secret=65722472-5cdf65
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
On Thursday 15 November 2007 21:19, Benjamin Goertzel wrote: so we still don't know exactly how poor a model the formal neuron used in computer science is Speaking of which: isn't this the age-old simple math function involving an integral or two and a summation over the inputs? I remember seeing this many years ago (before I knew its importance) on ai-junkie or maybe from Jeff Hawkins' On Intelligence. Way back when. And clearly I haven't been keeping track of the literature on neuronal modeling, but I would hope that there are other models out there by now. I need to read more journals. - Bryan - 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=8660244id_secret=65727925-e235c5
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
Here's an impressive movie: http://video.google.com/videoplay?docid=-2874207418572601262 Henry Markram, EPFL/BlueBrain: The Emergence of Intelligence in the Neocortical Microcircuit On 11/16/07, Bryan Bishop [EMAIL PROTECTED] wrote: On Thursday 15 November 2007 21:19, Benjamin Goertzel wrote: so we still don't know exactly how poor a model the formal neuron used in computer science is Speaking of which: isn't this the age-old simple math function involving an integral or two and a summation over the inputs? I remember seeing this many years ago (before I knew its importance) on ai-junkie or maybe from Jeff Hawkins' On Intelligence. Way back when. And clearly I haven't been keeping track of the literature on neuronal modeling, but I would hope that there are other models out there by now. I need to read more journals. - Bryan - 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/?; -- 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=8660244id_secret=65733872-1e3d2e
