Re: [agi] AI Buzz in Mainstream
On Thu, 17 Feb 2005, JW Johnston wrote: Bob Cowell's At Random column in this month's IEEE Computer magazine was about his renewed excitement in AI given Jeff Hawkins' book and work: http://www.computer.org/computer/homepage/0105/random/index.htm Then today, found a similar article in Computer World: http://www.computerworld.com/softwaretopics/software/story/0,10801,99691,00. html?source=NLT_PMnid=99691. Talks about Hawkins' work as well as that of Robert Hecht-Nileson and Stan Franklin. The AI winter ending? I'm always glad to see public enthusiasm for Ai, Neuroscience, et al. But unfortunately, the age-old problems of AI continue to haunt us. The homonculus is alive and well, even after all these years, and warnings. Every time some amateur AI philosopher, such as this Jeff Hawkins, throws his hat into the ring, I'm reminded of this pernicious problem. I haven't read the book, but from this fairly substantial review article, I get the sense that his model of the brain consists of a pattern matcher under the control of intention. There's the homonculous all over again. And Bob, the column author, loves it. I guess I can't blame Bob, the homonculous has been so pernicious precisely because we tend to anthropomorphosize complex processes. But the author of the book, if accurately portrayed by this review, deserves a ruler across the fingers for falling into the swamp of the homonculus, despite plenty of posted warnings. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] AI Buzz in Mainstream
Brad, I read Hawkins' book, and while I don't agree with his ideas about AI, I don't think he falls prey to any simple homunculus fallacy.. Some of my thoughts on his book are at: http://www.goertzel.org/dynapsyc/2004/ProbabilisticVisionProcessing.htm (BTW, my site seems to be down today but it will be back up shortly) Well consider my grumblings retracted then. It's certainly gratifying to see the enthusiasm generated on his http://onintelligence.org site. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Cell
On Fri, 11 Feb 2005, Eugen Leitl wrote: Just want to be clear Eugen, when you talk about evolutionary simulations, you are talking about simulating the physical world, down to a cellular and perhaps even molecular level? -B --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Cell
On Wed, 9 Feb 2005, Martin Striz wrote: --- Brad Wyble [EMAIL PROTECTED] wrote: Hardware advancements are necessary, but I think you guys spend alot of time chasing white elephants. AGI's are not going to magically appear just because hardware gets fast enough to run them, a myth that is strongly implied by some of the singularity sites I've read. Really? Someone may just artificially evolve them (it happened once already on wetware), and evolution in silico could move 10, nay 20, orders of magnitude faster. No never. Evolution in silico will never move faster than real matter interacting. But yes it's true, there are stupidly insane emounts of CPU power that would give us AI instantly (although it would be so alien to us that we'd have no idea how to communicate with it). However nothing that we'll get in the next 100 century will be so vast. You'd need a computer many times the size of the earth to generate AI through evolution in a reasonable time frame. Martin Striz __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Cell
There are several major stepping stones with hardware speed. One, is when you have enough for a nontrivial AI (price tag can be quite astronomic). Second, enough in an *affordable* installation. Third, enough crunch to map the parameter space/design by evolutionary algorithms. Fourth, the previous item in an affordable (arbitrarily put, 50-100 k$) package. Arguably, we're approaching the region where a very large, very expensive installation could, in theory, support a nontrivial AI. Yes, *in theory*, but you still have to engineer it. That's the hard part. Maybe I'm overstating my case to make a point, but it's a point that dearly needs to be made: the control architecture is everything. Let's do a very crude thought experiment, and for the moment not consider evolving AI, because the hardware requirements for that are a bit silly. So imagine it like this, you've got your 10^6 CPU's and you want to make an AI. You have to devote some percentage of those CPU's to thinking (ie analyzing and representing information) and the remainder to restricting that thinking to some useful task. No one would argue, I hope, that it's useful to blindly analyze all available information. The part that's directing your resources is the control architechture and it requires meticulous engineering and difficult design decisions. What percentage do you allocate? 5%? 20%? The more you spend, the more efficiently the remaining CPU power is spent. There's got to be a point at which you achieve a maximum efficiency for your blob of silicon. The brain is thoroughly riddled with such control architechture, starting at the retina and moving back, it's a constant process of throwing out information and compressing what's left into a more compact form. That's really all your brain is doing from the moment a photon hits your eye, determining whether or not you should ignore that photon. And it is a Very Hard problem. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Cell
The brain is thoroughly riddled with such control architechture, starting at the retina and moving back, it's a constant process of throwing out information and compressing what's left into a more compact form. That's really all your brain is doing from the moment a photon hits your eye, determining whether or not you should ignore that photon. And it is a Very Hard problem. Yes, but it's a solved problem. Biology is rife with useful blueprints to seed your system with. The substrate is different, though, so some things are harder and others are easier, so you have to coevolve both. This is where you need to sink moles of crunch. I don't think you and I will ever see eye to eye here, because we have different conceptions in our heads of how big this parameter space is. Instead, I'll just say in parting that, like you, I used to think AGI was practically a done deal. I figured we were 20 years out. 7 years in Neuroscience boot-camp changed that for good. I think anyone who's truly serious about AI should spend some time studying at least one system of the brain. And I mean really drill down into the primary literature, don't just settle for the stuff on the surface which paints nice rosy pictures. Delve down to network anatomy, let your mind be blown by the precision and complexity of the connectivity patterns. Then delve down to cellular anatomy, come to understand how tightly compact and well engineered our 300 billion CPUs are. Layers and layers of feedback regulation interwoven with an exquisite perfection, both within cells and between cells. What we don't know yet is truly staggering. I guarantee this research will permanently expand your mind. Your idea of what a Hard problem is will ratchet up a few notches, and you will never again look upon any significant slice of the AGI pie as something simple enough that it can can be done by GA running on a few kg of molecular switches. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Cell
I'd like to start off by saying that I have officially made the transition into old crank. It's a shame it's happened so early in my life, but it had to happen sometime. So take my comments in that context. If I've ever had a defined role on this list, it's in trying to keep the pies from flying into the sky. Evolution is limited by mutation rates and generation times. Mammals need from 1 to 15 years before they reach reproductive age. Generation That time is not useless or wasted. Their brains are acquiring information, molding themselves. I don't think you can just skip it. times are long and evolution is slow. A computer could eventually simulate 10^9 (or 10^20, or whatever) generations per second, and multiple mutation rates (to find optimal evolutionary methodologies). It can already do as many operations per second, it just needs to be able to do them for billions of agents. 10^ 9 generations per second? This rate depends(inversely) on the complexity of your organism. And while fitness functions for simple ant AI's are (relatvely) simple to write and evaluate, when you start talking about human level AI, you need a very thorugh competition, involving much scoial interaction. This takes *time* whether simulated time or realtime, it will add up. A simple model of interaction between AI's will give you simple AI's. We didn't start getting really smart until we could exchange meaningful ideas. But yes it's true, there are stupidly insane emounts of CPU power that would give us AI instantly (although it would be so alien to us that we'd have no idea how to communicate with it). However nothing that we'll get in the next 100 century will be so vast. You'd need a computer many times the size of the earth to generate AI through evolution in a reasonable time frame. That's not a question that I'm equipped to answer, but my educated opinion is that when we can do 10^20 flops, it'll happen. Of course, rationally designed AI could happen under far, far less computing power, if we know how to do it. I'd be careful throwing around guesses like that. You're dealing with so many layers of unknown. Before the accusation comes, I'm not saying these problems are unsolvable. I'm just saying that (barring planetoid computers) sufficient hardware is a tiny fraction of the problem. But I'm hearing a disconcerting level of optimism here that if we just wait long enough, it'll happen on all of our desktops with off-the shelf AI building kits. Let me defuse another criticism of my perspective, I'm not saying we need to copy the brain. However, the brain is an excellent lesson of how Hard this problem is and should certainly be embraced as such. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Cell
I'm confused, all you want are Ants? Or did you mean AGI in ant-bodies? Social insects are a good model, actually. Yes, all I want is a framework flexible and efficient enough to produce social insect level on intelligence on hardware of the next decades. If you can come that far, the rest is relatively trivial, especially if you have continous accretion of data from wet and computational neuroscience. I'm going to have to stop on this note. You and I live in different worlds. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Cell
Hardware advancements are necessary, but I think you guys spend alot of time chasing white elephants. AGI's are not going to magically appear just because hardware gets fast enough to run them, a myth that is strongly implied by some of the singularity sites I've read. The hardware is a moot point. If a time traveler from the year 2022 were to arrive tomorrow and give us self-powered uber CPU fabrication plants, we'd be barely a mouse fart closer to AGI. Spend your time learning how to use what we have now, that's what evolution did, starting from the primitive processing capabilities of single celled organisms. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] What are qualia...
On Sat, 22 Jan 2005, Philip Sutton wrote: Once complex brained / complecly motivated creatures start using qualia they could play into lifepatterns so profoundly that even obscure trends in the use of qualia for aesthetic purposes could actually effect reproductive prospects. For example, male peacocks have large tails that look nice - clearly qualia are playing a role in the differentiation process that decides which peacocks will be more or less successful in breeding. Cheers, Philip This is not at all true. I could design a neural network, or perhaps even symbolic computer program that can evaluate the attractivenes of a peacock tail and tune it to behave in a similar fashion as that tiny portion of a real peacock's brain. Does this crude simulation contain qualia? -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] What are qualia...
Yes, that's consistent with my line of thinking. Qualia are intensity of patterns ... in human brains these are mostly neural patterns ... and what we *call* qualia are qualia that are patterns closely associated with the part of the brain that deals with calling ... -- Ben I'd like to make a motion that this discussion topic be slated for disposal in the Yucatan storage facility, preferrably in drums that read WARNING: UNRESOLVABLE PHILOSOPHY Do Not Discuss for at least 30,000 years. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] A theorem of change and persistence????
On Sun, 19 Dec 2004, Ben Goertzel wrote: Hmmm... Philip, I like your line of thinking, but I'm pretty reluctant to extend human logic into the wildly transhuman future... Ben, this isn't so much about logic as it is about thermodynamics and it's going to be a very long time indeed before we can get around that one. Phil's idea comes down to stating that the entity will need to exert energy to counteract local entropy in order to remain a coherent being. I'd agree and state a trivial extension: The larger the sphere (in physical or some other space) of entropy that the entity is counteracting by expending energy, the greater it's chance of survival. Consider humanity, let's assume we'll survive as a species long as the earth remains intact. We're still vulnerable from asteroids (admittedly a miniscule chance). If we extend our sphere of control of entropy into space (by building gizmos and whatsits to protect the earth), we further increase our chance of deep time survival. We've made a bubble of entropy control around the earth. I'd also put forth this one: it's more energy efficient to ensure deep time longevity by reproduction than by protection. There's a book called the Robot's Rebellion which espouses the view that humans are a deep-time survival mechanism for our DNA. I haven't read it yet, but it sounds right on target for this topic. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] A theorem of change and persistence????
The Robot's Rebellion : Finding Meaning in the Age of Darwin by Keith E. Stanovich University of Chicago Press (May 15, 2004) ISBN: 0226770893 Cheers, Philip I'm glad you looked this up and posted it, as there are two books titled The Robot's Rebellion, the other being a very controversial attack on organized religion. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] 25000 rat brain cells in a dish can beat you at Flight Simulator
This research represents a major series of techincal triumphs, but the lay press versions of the story are somewhat misleading. There is no real learning going on, at least in the sense of synaptic modification. This is not really a brain system which exhibits information processing, reward punishment, self organization, or any of the other classical ideas of brain learning. According to the author's comments at: http://slashdot.org/comments.pl?sid=126880threshold=-1commentsort=0tid=191mode=threadcid=10614281 it's basically a system of adaptation to input and if the network were disconnected from the interface for some amount of minutes, it would return to a baseline state (ie have forgotten) Think of the neurons as springs, they adapt in a very characteristic way to input (as a spring's force increases with stretching), but on a time scale of minutes. By engineering the interface carefully, they end up with a system that depends on this pattern of adaptation, and therefore the system appears to learn, as the airplane is initially unstable but becomes stable. It is not, however, performing any kind of long term synaptic modification, or cellular reorganization. Very cool stuff techniques though. This is the same group that built a wheeled robot that learned to navigate in a similar way. Their expertise in keeping a brain culture alive for a long time and then using it is stunning, to say the least. http://www.neuro.gatech.edu/groups/potter/index.html -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Model simplification and the kitchen sink
Intelligence is not necessary to create intelligence. Case in point: us. The evolutionary process is a simple algorithm. In the very text that you quoted, I didn't say intelligence was necessary, I said a resource pool far larger than that of the entity being designed/deconstructed is necessary. And evolution certainly had that. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Model simplification and the kitchen sink
On Sun, 24 Oct 2004, Ben Goertzel wrote: Well, I don't think that building an AI is in principle too hard for a single mind to handle Understanding the brain may well be, because the brain has so damn many parts with their individual complex dynamics -- an AI doesn't need to be as complicated as the brain, though. (As a rough analogy, look how much harder it is to understand a bird wing than an airplane wing...). Airplane wings were easy for one person to understand back when they were simple things. But now that airplane wings have been adapted to generate more lift under different conditions, and also possess other vital functions (carrying fuel, air intakes, hydraulics), they are increasing in complexity dramatically. So shall it be with AGI, as they scale up in complexity to handle real world challenges, they will spiral out of the size that can be comprehended by a single person. I predict that this point at which AGI design exceeds a human's understanding will come long before they are capable of generative self modification. Hence, we will need teams of people before we get to that point. -Brad We're trying to build an AI, not via one person's efforts only, but via the combined efforts of a small team. I'm betting this is enough. I don't understand all of the Novamente codebase in detail -- no one person does -- but our small team, collectively, does. -- Ben G -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of J.Andrew Rogers Sent: Sunday, October 24, 2004 11:19 PM To: [EMAIL PROTECTED] Subject: Re: [agi] Model simplification and the kitchen sink On Oct 24, 2004, at 2:14 PM, Brad Wyble wrote: Another point to this discussion is that the problems of AI and cognitive science are unsolvable by a single person. 1 brain can't understand itself, but perhaps 10,000 brains can understand or design 1 brain. This does not follow. You can build arbitrarily complex machines with a very tiny finite control function and plenty of tape. The complexity of AI as an algorithm and design space is not in the same class as the complexity of an instance of human-level AI, even though the latter is just the former given some state space to play with. It is highly improbable that the core control function of intelligence cannot be understood by one person, or at least I see no evidence in theory to support this conjecture. Intelligence appears to be a pretty simple thing, even in theory; most of the nominal complexity can be attributed to people who don't really understand it (IMNSHO) or who require the addition of some complexity to solve a practical design problem. What you are saying is kind of like saying that no one can comprehend pi because no one can recite all the digits. j. andrew rogers --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Model simplification and the kitchen sink
Yes, but if any kid can buy a system with Avogadro number of switches, and large corporations 10^6 of that, no reason why we can't breed AI starting from an educated guess (a spiking network of automata controlling virtual co-evolving critters). That future is some 30-50 years remote. I think you're a few orders of magnitude off, but I made basically the same point here: http://www.mail-archive.com/[EMAIL PROTECTED]/msg01509.html But we're getting off track, my point at the start of this was that simple theories allow efficient communication, and therefore are essential for rapid progress in an effort like this, in which people are trying to design or understand something that is holistic. i.e. something that cannot be decomposed into mathematically discrete chunks the way that the physical sciences often can. AI and cognitive science both fall squarely into that category. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Model simplification and the kitchen sink
Engineering massively emergent systems is not something we're familiar with. But it doesn't mean it can't be done. You know the fitness function, let the system design itself. I'm not saying it can't be done. I'm saying it can't be done by one person. I'm saying the discipline requires the interaction of many scientists. Molecular neuroscience allows you map molecular events in their impact on structure and function. We're at the beginning here, but there are a lot simple parameters there (as well as terribly complex ones) for tweaking. While this hasn't been done, inflating the neocortex should result in a smarter critter. And that's a trivial parameter. I know you are all probably getting sick of me talking about how all of this is complicated, but it really is. Hearing that inflating the cortex is a trivial parameter grates on me terribly. Inflating the cortex would require, apart from a huge increase in metabolism and female pelvic diameter, lots of control structures so that this new cortex does something useful. Any new functional modules in the brain have to be carefully balanced against the structural and functional characteristics of existing ones or there are a great variety of mental illnesses that can result. We are already skirting the envelope of cognitive dysfunction, as evidenced by the number of people with psychological problems. You might, for example, have a system that is now incapable of resonating at frequencies that the brain expects to work at, and now you've broken consciousness. Or perhaps this new super genius is constantly wracked by crippling grand-mal seizures, because you've failed to apply the correct amount of neuromodulatory control to your new cortical real estate. Also, you've just increased the required amount of blood flow by perhaps double resulting perhaps in a massive increase in strokes due to larger and more numerous arteries. These are very real problems that evolution is constantly grappling with. You might hear about the rare cases of encephalitic babies with very oddly shaped brains that are fully functional members of society and think that human brains are heavily robust. What you don't hear about are the much larger number of stillborn children with brains that didn't work properly. Trivial parameter indeed. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Model simplification and the kitchen sink
On Mon, 25 Oct 2004, Ben Goertzel wrote: Brad wrote: I know you are all probably getting sick of me talking about how all of this is complicated, but it really is. Hearing that inflating the cortex is a trivial parameter grates on me terribly. Brad, I agree with you re human brains, but of course there is a big difference between engineered AI systems and human brains in this regard. Agreed, that whole email was concerning his comments about augmenting human brains, and was a bit off track for this list. My apologies. We can engineer our AI systems specifically so that adding more processor and memory WILL simply increase their intelligence, without a lot of complications and hassles Novamente is designed this way. You are correct about physical/architechtural hassles, but I'm not so sure about mental hassles. I can imagine functional complications that result from having too much unstructured room to play in. As a simple analogy, I recall that Principal Component Analysis can fail to give you useful results if you allow it to use too many dimensions. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Model simplification and the kitchen sink
The Godel statement represents itself, completely, via diagonalization. Unfortunately I'm not equipped to discuss Godel in depth. All I can do is argue by simple analogy, that is, it takes N1 neurons in the brain to mentally represent the idea of a neuron. Therefore the brain cannot represent itself. Now it's true that the representation of the system can be reduced in complexity by good theories, but given our severe limitations in working memory and abstract spatial conceptualization both of which are necessary to understand a complex system, we are orders of magnitude from having the capacity to understand even a reduced version of ourselves. In other words, any representation of the brain that is so reduced as to be singularly understandable by a single human will be so abstract as to have surrendered much of its explanatory power. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Unlimited intelligence.
On Thu, 21 Oct 2004, deering wrote: True intelligence must be aware of the widest possible context and derive super-goals based on direct observation of that context, and then generate subgoals for subcontexts. Anything with preprogrammed goals is limited intelligence. You have pre-programmed goals. And you are certainly not aware of the widest possible context, you'd need a brain several orders of magnitude larger than the universe you're trying to be aware of in order to posses that remarkable ability. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Ben vs. the AI academics...
On Sun, 24 Oct 2004, Ben Goertzel wrote: One idea proposed by Minsky at that conference is something I disagree with pretty radically. He says that until we understand human-level intelligence, we should make our theories of mind as complex as possible, rather than simplifying them -- for fear of leaving something out! This reminds me of some of the mistakes we made at Webmind Inc. I believe our approach to AI there was fundamentally sound, yet the theory underlying it (not the philosophy of mind, but the intermediate level computational-cog-sci theory) was too complex which led to a software system that was too large and complex and hard to maintain and tune. Contra Minsky and Webmind, in Novamente I've sought to create the simplest possible design that accounts for all the diverse phenomena of mind on an emergent level. Minsky is really trying to jam every aspect of the mind into his design on the explicit level. Can you provide a quote from Minsky about this? That's certainly an interesting position to take. The entire field of cognitive psychology is intent on reducing the complexity of its own function so that it can be understood by itself. On the other hand, Minsky's point is probably more one of evolutionary progress across the entire field, we should try many avenues and select those that work best, rather than getting locked into narrow visions of how the brain works as has happened repeatedly throughout the history of Psychology. Re: Deb, his stuff is clearly an amazing accomplishment, although I think that his success is more of a technical than a deeply theoretical flavor. On a more general note, I wouldn't expect to impress the AI community with just your theories and ideas. There are many AI frameworks out there, and it takes too much effort to understand new ones that come along until they do something amazing. So you'll need a truly impressive demo to make a splash. Until you do that, every AI conference you go to will be like this one. Deb's learned this lesson and learned it well :) -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Ben vs. the AI academics...
Hi Brad, really excited about Novamente as an AGI system, we'll need splashy demos. They will come in time, don't worry ;-) We have specifically chosen to Looking forward to it as ever :) I can understand your frustration with this state of affairs. Getting people to buy into your theoretical framework requires a major time investment on their part. This is why my own works stays within the bounds of conventional experimental and psychological research. I speak the same langauge as everyone else, and so it's easy to cross pollenate ideas. Of course, this is also why SOAR and similar architectures have such appeal despite their limitations. Because the SOAR community is speaking the same language to one another, it's possible (in theory) for the whole of them to make faster progress than if they each had their own pet architechture. This synergy is very real, but may be outweighed by SOAR's limitations. And, I hope my comments didn't seem to be dissing Deb Roy's work. It's really good stuff, and was among the more interesting stuff at this conference, for sure. Not at all, I think we're in general agreement about the value of his work. Now, I understand well that the human brain is a mess with a lot of complexity, a lot of different parts doing diverse things. However, what I think Minsky's architecture does is to explicitly embed, in his AI design, a diversity of phenomena that are better thought of as being emergent. My argument with him then comes down to a series of detailed arguments as to whether this or that particular cognitive phenomenon a) is explicitly encoded or emergent in human cognitive neuroscience b) is better explicitly encoded, or coaxed to emerge, from an AI system In each case, it's a judgment call, and some cases are better understood based on current AI or neuroscience knowledge than others. But I think Minsky has a consistent, very strong bias toward explicit encoding. This is the same kind of bias underlying Cyc and a lot of GOFAI. Whether something is explicit or emergent depends only on your perspective of what counts as explicit. I'll assume you mean anatomically explicit in some way (where anatomical refers to features of both neurophysiology and box/arrow design). With this assumption, I think b follows from a. Evolution has always looked for the efficient solution, so if evolution has explicitly encoded these behaviors, it's likely the best way to do it, at least as far as we'll be able to determine with our stupid human brains :) There's certainly a huge preponderance of evidence that our brains have leaned towards specific anatomically explicit solutions to problems in the domains that we can examine easily (near the motor and sensory areas). Of course, in many cases these anatomically explicity solutions are emergent from developmental processes, but I still think they should be considered explicit. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Ben vs. the AI academics...
So much for getting work done today :) I noticed at this conference that different researchers were using basic words like knowledge and representation and learning and evolution in very different ways -- which makes communication tricky! Don't get me started on Working Memory. In an AI context, it means whether something exists explicitly in the source code, rather than coming about dynamically as an indirect result of the sourcecode, in the bit-patterns in RAM created by the executable running... A fair definition. Agreed. And I think that sensorimotor stuff is more likely to be explicit rather than emergent in the brain And that, in coding an AI system, it's hopeless to try to make too much of cognition explicit rather than emergent -- but the same statement probably doesn't hold for perception action... If that were the case, would you not expect to see more variance in high level behaviors? Instead we tend to see the same types of behavior expressed, the only difference between people being the relative amount of expression of these tendencies. But I guess that's an arguable point, whether these observed tendencies among a population of people are actually there, or are only a product of the theories used to classify them. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Model simplification and the kitchen sink
Another point to this discussion is that the problems of AI and cognitive science are unsolvable by a single person. 1 brain can't understand itself, but perhaps 10,000 brains can understand or design 1 brain. Therefore, these sciences depend on the interaction of communities of scientists in a way that the physical sciences do not. And for this interaction to succeed, you need simplicity of theory above all else, because the individual agents in the discipline need to be able to communicate efficiently with one another, and that's the biggest bottleneck in the scientific process. So unless one lone researcher can solve the problem in isolation, and it is a mathematical fact that this is impossible, their long years of toil will be in vain unless the ideas can be communicated. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI research consortium
On Mon, 28 Jun 2004, J. Andrew Rogers wrote: There is most certainly not an infinite range of solutions, and there is an extremely narrow range of economically viable solutions. There are certainly an infinite range of solutions in AI, even for a specific problem, let alone for a space of many problems. What's an optimal economic solution for application X is a waste of resources for application Y. I think the burden of proof is on you to demonstrate how this range is extremely narrow. Even within this mailing list there are vastly divergent opinions on the basic fundamentals of the architechture that might do the job. We can't even settle the simple question of whether it's better pursuing biological analogues or starting from scratch. I would state that your entire characterization of technological history is incorrect. There is very little diversity in engineering in practice, and I don't expect AGI to be any different. Most of the variation between implementations of a given core technology is window dressing or attributable to variations in specification. We're perched on the edge of a semantic debate concerning what's fundamentally different. But I'll content myself with the concrete example of PC CPU's They do the same job, but they work differently inside, different people have designed them and most importantly for the purposes of this discussion: different companies get money when I purchase them. And the only reason CPU's function similarly is because they need to run the same software at an instruction level. AGI will be under no such constraint. You can parameterize an implementation for variation, but you can't parameterize the underlying mathematics and physics that allows a technology to exist in the first place. AGI is a math problem with huge Except that this isn't about physics, it's about information processing. And given a certain amount of information processing resources (as limited by physics yes), there are a great many ways to use them to perform a task. economic consequences. Economics will dictate the implementation parameters of the mathematics, and will strongly bound the architecture of successful implementations. There won't be diversity, there will be slightly different flavors of the same solution. Only in the very first iteration. The solution space we explore will bifurcate rapidly. Any design competition will be decisive and short-lived. There will be no second place. You are mis-modeling how AGI technology will interact with the marketplace by treating it as a conventional production-bound widget when this will be a case of something very different. There is only one barrier to entry into the AGI market, but that barrier is precisely such that the first-mover will have a decisive advantage. AI will be bound by the same economic models as everything else, but you have to use valid data and parameters or you'll end up with completely broken expectations. AI technology is almost ideally pathological from a market competition perspective. It's true, I'm fitting AGI into the conventional mold of product development and economics. I do this because I think AGI development is going to be an iterative process with solutions that are incremental improvements, and with several companies close on each other's heels. There won't be a clear winner, because the fitness landscape of AGI is poorly defined. I'd like to hear how you define AI being pathological from a market competition perspective? Is it because AI's will improve themselvers, and therefore the early winners will accelerate their own development process? --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI research consortium
Great stuff Andrew. I should have specified extremely narrow for implementations in our universe as we generally understand it. This is an old discussion, so I'm not going to rehash it. The enemy of implementation is *tractability*, not will this work in theory if I throw astronomical quantities of computing resources at the problem. This is a crucial point that still seems to escape many people. It is fairly evident from what we know of the mathematics that this is the case. You have a wide diversity of choices in theoretical designs, but only because while dabbling in theory you can ignore engineering tractability in practice. My perspective comes from reverse engineering the brain. In doing this, I am exposed to the many compromises and innovations evolution was forced to make. But these are very specific solutions to very specific problems, such as how to recognize the minute differences between thousands of faces, or learning to understand speech, remember events that occur over 80 years. And the specific solutions of how we cognitively deal with complex decision making tasks is a problem we're just beginning to scratch at. The more one studies the specifics of the solutions the brain uses, the more one realizes the incredible varieties of strategies that could have been used, yes economically, and yes in our universe. Evolution was forced into certain compromises and choices not because they were the only solutions to the problem but because evolution is heavily bound by its own historical precedent. If it helps you, I'll restate it this way: Almost every major AI effort is a theoretically correct design. Extremely few, if any, are reasonably tractable AI designs in our universe. The problem has never been about theoretical correctness, at least not if one has a reasonable understanding of the underlying problem. Yes, I understood your point. I do have a reasonable understanding of the underlying problem. The difference is simply one of perspective. I see limitless solutions, you see a narrow track. Finding someone with a contrary opinion on any topic is easy. That they have an opinion at all does not make their opinion meaningful or valid ipso facto. I think there is far more agreement on the basics than you seem to think. Not being able to discern between the idle chatter of people who don't grok the underlying subject matter and people with a serious clue will give you that impression though. There are a lot of airplane enthusiasts who do not understand the basic physical design principles of powered flight. For any topic that attracts fanboys and cranks, it often takes a great deal of genuine expertise to separate the cranks and fanboys from people with genuine expertise. I've been involved with the hard AI, the weak AI and the neurophys approaches to these problems at an academic level. There is very little agreement there. All CPUs are technologically identical for all intents and purposes, and commodities. The differences revolve almost entirely around the manufacturing costs and efficiencies. This is a terrible example actually, as it is an industry where competitors differentiate themselves by manufacturing economics. So, if you had a company that made its money sorting sets of data in a competitive market, you are saying that the selection of sorting algorithm is irrelevant to your ability to survive in the marketplace? I'm not a big believer in self-improving AI, at least not in the sense that it often seems many others use that term. The core design will have to be pretty close from the beginning for an implementation to even be plausible in practice. The primary difference is that AI will be a business where the economics is bound almost entirely by knowledge rather than more traditional manufacturing cost structures. Even nominally pure information products, such as software, are bound by the manufacturing cost of the code, though the overall process starts to show symptoms of knowledge-bound economics. So by knowledge you mean essentially empirical facts, that 1 + 2 = 3. The hard part in AGI is not finding that knowledge, but in developing an agent that can distill that knowledge. That's what we do, as people. The problem with knowledge, is that it has to be learned. One can't simply buy knowledge, and the whole point of intelligence is to That's certainly not true. I can buy a chemistry textbook full of essentially mathematical knowledge of how atoms interact. That knowledge doesn't have to be learned either, I can use it straight from the book. I think you were thinking of something else when you discuss knowledge, and if so, you'll need to clarify it. manufacture knowledge from data. Economics and economic advantage is all about knowledge -- that is the real value of intelligence after all. Any machine intelligence will quickly be in a position of clear economic advantage in the marketplace which will allow it to maximize its
RE: [agi] AGI's and emotions
On Wed, 25 Feb 2004, Ben Goertzel wrote: Emotions ARE thoughts but they differ from most thoughts in the extent to which they involve the primordial brain AND the non-neural physiology of the body as well. This non-brain-centricity means that emotions are more out of 'our' control than most thoughts, where 'our' refers to the modeling center of the brain that we associate with the feeling of 'free will.' -- Ben G I would agree with this. Emotions seem to arise from parts of the brain that your central executive has minimal control over. They can be suppressed and manipulated with effort but they are distinct from the character of thoughts originating in other parts of the brain. It's probably a mistake to characterize emotions as a unitary phenomenon though. Different emotions have different functions, and likely originate from different structures. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] AGI's and emotions
I guess we call emotions 'feelings' because we feel them - ie. we can feel the effect they trigger in our whole body, detected via our internal monitoring of physical body condition. Given this, unless AGIs are also programmed for thoughts or goal satisfactions to trigger 'physical' and/or other forms of systemic reaction, I suppose their emotions will have a lot less 'feeling' depth to them than humans and other biological species experience. That's not the entirety of the difference between emotions and other types of thoughts. A reasoning entity can detect that their thoughts are under the influence of an emotion. For example, consider being in a road rage situation, which I'm sure we can all relate to. You know full well that your reaction of anger towards someone who's unwittingly committed a minor offense to you is wildly irrational and yet you can't help but feel a flash of extreme animosity towards someone else (or maybe your steering wheel :)). The fact that you know it's an emotional reaction doesn't prevent you from feeling its effects on your thoughts, it just lets you handle it without acting on it. So any entity capable of remembering their thought processes would be able to detect the influence of an emotion (at least the human variety) on the current flow of their thoughts even without body-state markers. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Within-cell computation in biological neural systems??
Nonlinear dendritic integration can be accurately captured by the comparmental model which divides dendrites into small sections with ion channels and other internal reaction mechanisms. This is the most accurate level of modeling. It may be possible to simplify this model with machine learning techniques and without significant loss in accuracy. I am well aware of compartmental modelling and have done it myself. But this type of model only accounts for the physical size/character of a dendrite, ignoring, in principle, a whole raft of complex molecular dynamics of what might be occuring inside it. Such molecular dynamics will sure contribute to the nonlinear aspects of a dendrite. Just as an example, a new type of neuron has recently been discovered that can hold a steady state of firing in isolation, apply current, rate increases and remains stable at a new threshold. It's dynamically settable, which blows away all standard Integrate Fire models. I don't know the exact mechanisms that give rise to that type of neurons, but the comparmental model should be able to cover this. What is needed is a large-scale database of neuronal characteristics (automation). Yes, one can create a model of a neuron that does this, it's already been done. It's far from a standard model though. My point, however, was that there is an entire world of complexity within the cell that will be relevant to its role in a neural network (as opposed to simply metabolic) that we are just beginning to understanding. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Within-cell computation in biological neural systems??
The jury is very much out Phillip. Eliezer goes too far in saying it's a myth perpetuated by computer scientists. They use the simplest representations they know to exist in their models for purposes of parsimony. It's hard to fault them for being rigorous in this respect. But neurons are surely far more complex than this. The majority of computation may well occur within the nonlinear bursting dynamics of dendrites. Just as an example, a new type of neuron has recently been discovered that can hold a steady state of firing in isolation, apply current, rate increases and remains stable at a new threshold. It's dynamically settable, which blows away all standard Integrate Fire models. We're just scratching the surface of an enormous iceberg. If you're trying to build some useful index of brain power based on number of neurons (or even synapses), give up and wait 30 years at least. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Real world effects on society after development of AGI????
Ben Wrote: 1) AI is a tool and we're the user, or 2) AI is our successor and we retire, or 3) The Friendliness scenario, if it's really feasible. This collapse of a huge spectrum of possibilities into three human-society-based categories isn't all that convincing to me... Yes, a list like this should always include 4) Something else --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Real world effects on society after development of AGI????
On Tue, 13 Jan 2004, deering wrote: Brad, I completely agree with you that the computer/human crossover point is meaningless and all the marbles are in the software engineering not the hardware capability. I didn't emphasize this point in my argument because I considered it a side issue and I was trying to keep the email from being any longer than necessary. But even when someone figures out how to write the software of the mind, you still need the machine to run it on. I believe in the creative ability of the whole AGI research ecosystem to be able to deliver the software when the hardware is available. I believe that the human mind is capable of solving this design/engineering problem, and will solve it at the earliest opportunity presented by hardware availability. You seem to contradict yourself, saying first that the hardware crossover point is meaningless, then implying that we'll solve the design problem at the first opportunity. I won't reiterate my stance again, you know what it is :) Regarding nanotechnology development, I think we are approaching nano-assembly capability much faster than you seem to be aware. Check out the nanotech news http://nanotech-now.com/ Being able to make these bits n bobs in the lab is a different problem than having autonomous little nanorobots doing it. You then have problems of power distribution, intelligent coordination, heat dissipation. It's quite a ways off in my opinion. Again, I'm not sure whether this or AGI will come first, they are both 'H' hard. Regarding science, Yes, turtles all the way down. Probably. But atoms are so handy. Everything of any usefulness is made of atoms. To go below atoms to quarks and start manipulating them and making stuff other than the 92 currently stable atoms has such severe theoretical obstacles that I can't imagine solving them all. Granted, I may be lacking imagination, or maybe I just know too much about quarks to ignore all the practical problems. Quarks are not particles. You can't just pull them apart and start sticking them together any way you want. Quarks are quantified characteristics of the particles they make up. We have an existence proof that you can make neat stuff out of atoms. Atoms are stable. Quarks are more than unstable, they don't even have a separate existence. I realize that my whole argument has one great big gaping hole, We don't know what we don't know. Okay, but what I do know about quarks leads me to believe that we are not going to have quark technology. On a more general vein, we have known for some time that areas of scientific research are shutting down. Mechanics is finished. Optics is finished. Chemistry is finished. Geology is basically finished. We can't predict earthquakes but that's not because we don't know what is going on. Metrology we understand be can't calculate, not science's fault. Oceanography, ecology, biology, all that is left to figure out is the molecular biology and they are done. Physics goes on, and on, and on, but to no practical effect beyond QED and that is all about electrons and photons and how they interact with atoms, well roughly. Perhaps some of them have evolved into different kinds of science that you no longer recognize as such. That's not the same thing as shutting down. I don't expect this clarification to change your mind. I think we are going to have to agree to disagree and wait and see. Yes indeedy. :) See you after the Singularity. Ah ah, but the Singularity says you can't make that prediction :) -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Real world effects on society after development of AGI????
On Mon, 12 Jan 2004, deering wrote: Brad, you are correct. The definition of the Singularity cited by Verner Vinge is the creation of greater-than-human intelligence. And his quite logical contention is that if this entity is more intelligent than us, we can't possibly predict what it will think or do, hence the incomprehensibility. Many people subscribe to this statement as if it were scripture, not me. A few years later, Ray Kurzweil noticed that the advancement of knowledge in molecular biotechnology and miniaturization of electrical and mechanical systems were graphing tracks that closely matched the graphs for the advancement of computational capacity. It appears from the graph data that computational capacity of desktop computers will surpass human brains at I keep making this point as often as it has to be made: surpassing the computational capacity of the brain is not even close to sufficient to develop AGI. The hard part, the real limitation, is the engineering of the type that Ben's doing. Software engineering will be our biggest hurdle for decades after we cross the brain CPU barrier. about the same time as miniaturization reaches positional molecular assembly and knowledge of molecular biotechnology reach completeness. If you think about it, the fact that these three areas of technological advancement are tracking together toward specific goals is not surprising. They are all very closely tied to each other. The advance of miniaturization of electrical and mechanical systems are producing the tools for the investigation of living organisms at the molecular level. It is also producing the hardware for the advancement of computational capacity. The more powerful computers are providing the control systems for the automation of molecular biotechnology speeding up the assimilation of knowledge. Computers and nanotechnology are progressing lockstep; scientists need more powerful computers to advance nanotechnology, computers need more miniaturized circuits to become more powerful. And molecular biotechnology is dependent on both the advancement of computational capacity and the advancement of nanotech tools. So is born the concept of the three technology Singularity. A good point, however nanomanufacturing has some special challenges, as does mind design. One is likely much harder than the other, I just don't know which. 1. Intelligence will top out at levels of great efficiency, accuracy, and speed; and the best types of thought processes will be similar to ways of thinking used by our best geniuses, a mode of thought that is not beyond our comprehension, just merely beyond our perfect execution. It will be beyond your comprehension. I don't know about you, but I cannot comprehend the way hardcore theoretical mathematicians think about equations. I feel like a cat staring at its master. In the same way, you spend most of your day thinking about topics that are utterly incomprehensible to people 100 years old. So it will be with your children and theirs. 2. Physical technology will reach a limit at the complete control of the positioning of atoms in fabrication, maintenance, and functioning, including molecular scaled robots and machinery. You sound like physicists 100 years ago who thought that the proton, neurtron and electron were the end of the road. Why is it so hard to imagine that we can put quantum particles to use? Or that there are not layer upon layer of sub-quantum particles? I think It's turtles all the way down, and so far History has proved me right. 3. Science will reach a limit with the completion of cataloging and understanding of all molecular processes in living organisms. This is largely irrelevant. Science is so much more than cataloging living organisms. It is completely open-ended. When I say limits, I don't mean that we will stop innovating, merely that we will have all of the basic knowledge and capability we are ever going to have, and what is left is art. Sure we will still be inventing better mouse traps, but not whole new areas of science. Sorry. I can't see how you've demonstrated this. Given these limits, the Singularity becomes very comprehensible. We know what basic capabilities we will have. We can plan how we want to use them. We get to decide what principles our society will be based on, and how we will implement them. I couldn't disagree more. Okay, you can start laughing now. Just shaking my head in wonder is all :) --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Dr. Turing, I presume?
I see your point, but I'm not so sure you're correct. If you're devoting resources specifically to getting some attention, you may indeed speed up the process. I wish you luck. However even if you do get such attention, it will still take quite a while for the repercussions to percolate through society. Mike seemed to be implying a technological rapture with very rapid changes at all levels of society. I think that people at all levels will be slow to react while a small percentage of early adopters who grab hold and start creating a market. This belief is based on historical precedent. -Brad I think about it this way: * Sometimes bullshit get huge amounts of media attention and money. * Sometimes really *demonstrably* valuable things get pathetically little media attention and money * Sometimes really demonstrably valuable things get huge amounts of media attention and money Assuming Novababy really eventuates like I hope/believe it will, I intend to ensure that Novamente AGI falls into the latter category. I don't think its so impossible to achieve this, it just requires approaching the task of fundraising and publicity-seeking with some energy and inventiveness. I think I have a good idea of what achieving this requires. For instance, I have a good friend who lives here in DC who is a very successful PR agent and would be quite helpful on the media aspect of this (one of his jobs was doing PR for the Republic of Sealand, which was totally obscure before he started working with them, and wound up on the cover of Wired and in every major paper... and is a heck of a lot less generally interesting than Novamente...). And I know a few people in the US gov't research funding establishment, who personally like AGI, but who can't authorize AGI funding due to internal-politics constraints. It wouldn't take such a big nudge for the research-funding establishment to give them the go-ahead to follow their intuitions and fund AGI. I think that raising funds and serious positive publicity for a scientifically successful baby AGI project is a *hard* problem, but nowhere near as hard as making the baby AI in the first place. Confident as I am in Novamente, it's the making the baby AI work problem that worries me more, not the how to publicize and monetize AGI once the baby AI works problem!! -- Ben G --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Dr. Turing, I presume?
On Sat, 10 Jan 2004, deering wrote: Ben, you are absolutely correct. It was my intention to exaggerate the situation a bit without actually crossing the line. But I don't think it is much of an exaggeration to say that a 'baby' Novamente even with limited hardware and speed is a tremendous event in the history of life on Earth. A phase change starts with one molecule. As computers are Yes, but what effect will it immediately have? How long after the development of the transistor that the average person's life was significantly changed? This baby novamente will be one of many blips on the radar. The public is constantly innundated with reports of revolutions in AI and they have become jaded by such sensationalistic reporting. So that if/when Ben succeeds, how is anyone to know that they're looking at a real baby AI, and not some slight enhancement of the AIBO? They won't. Only you, I and maybe 998 other other people would understand the significance and these 1000 only because we're well versed with Ben's activities. Any AGI will take a decade to make itself known and to rise above the signal/noise ratio of scientific media. becoming more powerful and nanotech capabilities reach closer to the ultimate goal of molecular positional assembly the world will cross a threshold similar to supercooled water where one triggering event will set off a chain reaction causing a phase change to ice throughout the entire mass. Okay, I'm exaggerating again, but not much. The money men know it is coming. But they have been burned so many times before in the A.I. category that they are not willing to touch the stove again, unless someone can show them something that works. It doesn't have to be a finished product, just something that demonstrates a new capability. Your 'baby' Novamente or Peter's proof-of-concept example or James Rogers' who-knows-super-secret-whatits will trigger a phase change in funding for AGI. The practical applications are unlimited. The profit potential is unlimited. That's why the money men threw away so much twenty years ago on projects that didn't have a ghost of a chance and got burned. I'm not saying that your 'baby' Novemente will change the whole world overnight all by itself. But any working example of AGI, no matter how limited, will trigger a complicated chain reaction in the economy and mindset of the world. The initial example, whatever it is, may turn out to be a flawed design of limited usefulness (I wouldn't want to see scaled-up jumbo 'Wright Flyers' populating airport terminals) but it will not matter. Just look at the funding that GOOGLE has attracted with some cleverly written but dumb (non-AGI) rules. You, me and all of us are a collection of cleverly written but dumb rules :) --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] The emergence of probabilistic inference from hebbian learning in neural nets
This is exactly backward, and which makes using it as an unqualified presumption a little odd. Fetching an object from true RAM is substantially more expensive than executing an instruction in the CPU, and the gap has only gotten worse with time. That wasn't my point, which you may have missed. The point is that with our current technology track it's far cheaper to double your memory than to double your CPU speed. I'm not referring to the amount of memory bits processed by the CPU, but the total number of pigeonholes available. These are not one and the same. Therefore you can make gains in representational power by boosting the amount of RAM, and having each bit of memory be a more precise representation. You can afford to have, for example, a neuron encoding blue sofas and a neuron encoding red sofas. While a more restricted RAM approach would need to rely on a distributed representation, one with only sofa neurons and color neurons. (apologies for the poor example, but I'm in a hurry) Your points are correct, but refer to the bottleneck of getting information from RAM to the CPU, not on the total amount of RAM available. Back to the problem of the human brain, a big part of the problem in the silicon case is that the memory is too far from the processing which adds hard latency to the system. The human brain has the opposite problem, the processing is done in the same place as the memory it operates on (great for latency), but the operational speed of the processing architecture is fundamentally very slow. The reason the brain seems so fast compared to silicon for many tasks is that the brain can support a spectacular number of effective memory accesses per second that silicon can't touch. Both technologies have their advantages and disadvantages. The brain's memory capacity (in terms of number of addressable bits) cannot be increased easily while a computer's can be. I merely suggest that this fundamental difference is something to consider if one is intent on implementing AGI in a Neumann architechture. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] The emergence of probabilistic inference from hebbian learning in neural nets
Guess I'm too used to more biophysical models in which that approach won't work. In the models I've used (which I understand aren't relevant to your approach) you can't afford to ignore a neuron or its synapses because they are under threshold. Interesting dynamics are occurring even when the neuron isn't firing. You could ignore some neurons that are at rest and hadn't received any direct or modulatory input for some time, but then you'd need some fancy optimizations to ensure you're not missing anything. But in the situation you're referring to with a more abstract (and therefore more useful to AGI) implementation, these details are irrelevant. I just wanted to chime in and ramble a bit :) Very glad to hear things are going well with Novamente. Hope the holidays treat all of you well. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Evolution and complexity (Reply to Brad Wyble)
On Wed, 8 Oct 2003, Majboroda O.M. 16.03.2001 wrote: CPU cicles is not analogue of energy. Transformation of energy is not necessarily accompanied by transformation of the information. But transformation of the information is always occured due to energy. It's a fine analogy actually. It's not perfect, analogies never are. I understand and I appreciate your idea. I might tell it last year too. [snip] Lots of big words in there, but unless you believe that there was a creator, or that for some reason computers can't simulate physical laws complex enough to evoke a nice fitness landscape (ie quantum randomness is necessary for evolution), nothing that you've said countermands my point that we can, in principle, generate Ai through sheer brute force when our computers get fast enough (ie planet sized nano computers). And, as I said, we can use our wits to shortcut the evolutionary process by a few(hundred) orders of magnitude, which is essentially the goal of AI. Seems pretty cut and dried. I think you're thinking too hard. Evolution is conceptually really simple. Take closed system, add energy, bake for 4 billion years, get complexity. -Brad Wyble --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Discovering the Capacity of Human Memory
Good point Shane, I didn't even pay attention to the ludicrous size of the number, so keen was I to get my rant out. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Discovering the Capacity of Human Memory
It's also disconcerting that something like this can make it through the review process. Transdisciplinary is oftentimes a pseudonym for combining half-baked and ill-formed ideas from multiple domains into an incoherent mess. This paper is an excellent example. (bad math + bad neuroscience != good paper) --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
RE: [agi] Early AGI training - multiple communications channels /multi-tasking
You're right, it's possible, but don't underestimate the problems of having multiple interaction channels. It's not a almost a freebie (to heavily paraphrase your and Phil's comments). Multiple streams of interaction across broadly varying contexts would require some forms of independent memory for them, or you'll get cause and effect all jumbled up as an event from stream A comes directly after stream B. If the memory system doesn't understand this distinction, the search space to figure out correlations and causations will be extremely huge. Our brains can support multiple simultaneous streams of interaction without a major performance hit as long as the streams are more or less orthogonal (e.g. walk while talking). But there are quite good reasons why we can't follow two conversations at the same time. Now yes an AI can handle multiple streams, but you are going to pay for it somehow, either with multiple independent memory systems for each stream which must later be integrated, or by a hugely increased processing cost for analyzing and consolidating a single memory system. My advice is to learn to crawl before you try running. -Brad Yeah -- the design supports multiple interaction channels This idea comes up naturally when one thinks about having an AI * chat with different folks all around the Web, at the same time * look through sensors located at different places in the world * control robots undersea as well as in outer space, etc Why shouldn't one mind do all these things, after all? Why should cognition and memory need to be restricted to a single locale in space, a single integrated set of sensors/actuators? I guess that the subjective experiences AI's with multiple interaction channels may be quite diverse. Maybe there will be a threshold of interactivity between the interaction channels and their cognitive-support units, at which there is a phase transition from multiple subjective consciousnesses to single subjective consciousness. This is yet another way that digital intelligence will likely be very, very different from human intelligence. -- Ben -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of Philip Sutton Sent: Tuesday, September 02, 2003 9:07 PM To: [EMAIL PROTECTED] Subject: [agi] Early AGI training - multiple communications channels / multi-tasking Hi Ben, It just occurred to me that very early in a Novamente's training you might want to give it more than one set of coordinated communication channels so that the Novamente can learn to communicate with more than one external intelligence at a time. My guess is that this would lead to a a multilobed consciousness - where each communication channel (2 way, possibly multiple senses) would have it's own mini-consciousness and the Novamente would have a metaconsciousness that knits all its mental parts together as a whole self. I don't think we should assume a single communications channel mode for Novamentes just because that's how we think of biological minds communicating. Maybe it's a bit like teaching a person to play the piano with two hands??? Or how people learn to use whole-body motor skills for sport. But with a sharper and higher level of independent consciouness attached to each communication channel/conversation. We learn to play with each hand and with both hands together. Cheers, Philip --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Web Consciousness and self consciousness
Just a word of advice, you'd get more and better feedback if your .htm didn't crash IE. If you've got some wierd html in there, tone it down a bit. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
[agi] sick of AI flim-flam
A tiny rant about bogus AI. I was depressed to find this site: http://www.intelagent.org/ and another by the same snakeoil salesman (Sol Endelman) http://hardwear.org/personal/PC/ See that pencil drawing of the wearable computer? Lifted straight from the MIT Mithril project website. I'm going to fire off a letter to Mithril and point it out, but the point is this whole webpage, if you bother to read it, is practically gibberish. And then you go back to hardwear.org to see the citing outrageous prices for stuff like trousers and socks, let alone the $60,000 actuator array. On going back to intelagent.org, it seems clear that this guy's purpose is to create authentic looking companies and swindle the bejesus out of gullible VC's. He throws around alot of buzzwords and technical sounding garbage, which might sound good to a naive investor. This guy is worse than the spammers. People like this cripple the ability of legitimate AI companies to get funding by scaring off investors. Disgusting. I only hope Sol is on this list and can read this, but I doubt it. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Perl AI Weblog
The open source concept to AI, which is essential what you are doing here, is a very interesting one. However, the open source success stories have always involved lots of tiny achievable goals surrounding one mammoth success (the functional kernel). i.e. there were many stepping stones which served to organize efforts. This approach doesn't seem to have a series of achievable goals that will direct efforts. And if I may offer some constructive criticism of clarity, the text of this email is very clear, but that of the webpage is much harder to follow. If you wish people to take this seriously, make an effort to make it very clear exactly what you are hoping for them to do. Some questions I was unable to answer in 5 minutes of browsing your site: How do these minds compete? On what/whose servers will they run? What input is the AI system given? By what means will they be evaluated? Why Perl? What (who's)code does the main Alife loop connect with for the submodules? You use the word port as if programmers are merely translating an engine from one codebase to another, but that doesn't seem to be the case? What did you mean by port exactly? And finally, the claim that AI has been solved in your webpage title is a bit offputting. I envy you your enthusiasm with this project though. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Fw: Do This! Its hysterical!! It works!!
I use elm so I couldn't tell, was there a virus riding on that? Just curious. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Robert Hecht-Nielsen's stuff
Well the short gist of this guy's spiel is that Lenat is on the right track. The key is to accumulate terabytes of stupid, temporally forward associations between elements. A little background check reveals that this guy isn't a complete nutcase. He's got some publications (but not many), and a real lab position. However, his claims are a bit too grandiose and he smacks a bit of a snake oil salesman at the end when he's fielding the questions, especially the one about the inability of his theory to handle the tightly regimented sequence of commands necessary to execute motor programs. He sidesteps that one in a particularly obfuscatory fashion. Nor is his model very interesting in its application. Neuroscience data stands counter to his basic claim that the cortex is just a big sheet of associatiors, there are many genetically described connection patterns. His claim that we set up relatively immutable patterns early in life have only been shown to be true for the visual cortex as far as I know. AI isn't a failure because everyone involved is an idiot and keeps missing the obvious point that this genius has stumbled upon. AI is a failure because AI is hard. I give it a C-. It's long on words and full of idealistic grandeur, but short on substance when you really boil it down. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Educating an AI in a simulated world
It's an interesting idea, to raise Novababy knowing that it can adopt different bodies at will. Clearly this will lead to a rather different psychology than we see among humans --- making the in-advance design of educational environments particularly tricky!! First of all, please read Diaspora by Greg Egan. As a SF author, he excels in his informed approach to AI design, philosophy, and neuroscience. This book touches on this topic(AI's designed for multiple bodies) very directly. This VR training room initially seemed like a great idea to me, but on consideration, I'm not so certain it's worth the trouble. First of all, you are reducing the complexity of the environment by orders of magnitude. One could argue that it is a baby's physical interation with the world is the cornerstone on which all future intelligence resides. Now you've made pains to point out that you're not trying to recreate people, but intelligence. However, a Novamente grounded in a different reality will be difficult for people to interact with. So here are two possible issues: the VR world might actually slow down the intellectual growth of the Nova Baby. And even when intelligent, it will be more alien from us than it needs to be. A second point about this plan is that you are creating extra work for yourself both in designing a VR training paradigm, and then in bridging the gap from VR to the real world, which would be no picnic. So there are some possible negatives, the positives you've already listed. If this course is decided upon, consider giving the Novamente an ability to sense objects in their native format (sprites in 3d coordinates). If your intent is to simplify the world, don't add in the fuzz of the artificial visual input, which is often flawed (e.g. clipping errors). Give the Novababy access to the underlying framework of the world or it will be eternally confused as it tries to figure out why it can walk through trees, or why Mr. Smith's left toes are inside its own foot. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Intelligence enhancement
What wasn't made very clear in that article is that the sole function of TMS is shutting down specific areas of the brain for a short while. So it's not that he's improving a given piece of brain tissue, he's shutting off certain areas which changes the balance of power in the mind, and allows creativity to become temporarily dominant in people that are usually very left-brained. I'm prepared to believe these results, and I think it's fascinating work. However, I'm not going to be first in line to let someone do TMS on me. Nothing that temporarily shuts down areas of the brain can be good for them. I'm imagining huge swaths of ions blasting back and forth through neural membranes as the metabolic processes try to maintain homeostasis, accumulating cell damage or death in the process. Maybe if this same thing were achieved through cooling (although I can't imagine how it could be done non-invasively), or with pharmacology, I would feel better about it. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Dog-Level Intelligence
It might be easier to build a human intelligence than a dog intelligence simply because we don't have a dog's perspective and we can't ask them to reflect on it. Don't be quick to assume it would be easier just because they are less intelligent. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Hard Wired Switch
An AGI system will turn against us probably if humans turn against it first. It's like raising a child, if you beat the child every day, they are not going to grow up very friendly. If you raise a child to co-operate and co-exist with its environment, what possible motivation is there for it to grow up to be hostile? cheers, Simon It's not question of hostility so much as deciding that we are an obstruction in its goal path. There are plenty of lines of reasoning that could lead an AGI to deciding that its existence would be easier without us getting in the way. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Playing with fire
Extra credit: I've just read the Crichton novel PREY. Totally transparent movie-scipt but a perfect text book on how to screw up really badly. Basically the formula is 'let the military finance it'. The general public will see this inevitable movie and we we will be drawn towards the moral battle we are creating. In early times it was the 'tribe over the hill' we feared. Communication has killed that. Now we have the 'tribe from another planet' and the 'tribe from the future' to fear and our fears play out just as powerfully as any time in out history. Note: I'm not arguing for or against AI here, just bringing to light some personal observations This particular situation is different than the others you describe(tribe over the hill). To accept the dangers of AI, one must first swallow racial pride and admit that we are not the top-dogs in the universe. Few people are willing to do this, even among well-educated, science minded engineers. I just tested this topic on my group of internet friends in a private forum with 20 some people. I was unable to convince a single person that this danger is real with a day's worth of intensive back and forth discussion. They assumed the typical we can just control it mentality that has always been prevalent. Notice that even in gloomy bad-AI stories such as Terminator and the Matrix, the humans always win in the end. This is what the mainstream will believe becauses they want to believe it. In other words, I don't think the public is going to care one-iota about the dangers of AI. They'd prefer to focus their energy on banning truly harmless technologies, such as cloning. People fear clones because as far as they are concerned, clones are people too, so we're dealing with an equal, and can lose. But AI's are just machines, they can be out-smarted or out-evolved as far as the average person is concerned. The upside is that AI researchers won't have to fight to keep their research legal. The downside of this is that we're more likely to destroy ourselves. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Playing with fire
One thing I should add: It's the same hubris I mentioned in my previous message that prompted us to send out satellites effectively bearing our home address and basic physiology on a plaque in the hope that aliens would find it and come to us. Even NASA scientists seem to have no fear of anything non-human. From a species-survival perspective, we'd be better off contacting alien races on our terms, rather than inviting them to come by for a visit. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] seed AI vs Cyc, where does Novamente fall?
Yep. Novamente contains particular mechanisms for converting between declarative and procedural knowledge... something that is learned procedurally can become declarative and vice versa. In fact, if all goes according to plan (a big if of course ;) Novamente should *eventually* be much better at this than the human brain. I'm glad that you choose to incorporate elements of human cognitive theory into Novamente, even if you are not intent on building a brain. Such commonalities will make the design of NM far more intuitive and accessible to designer and lay-person alike. For instance, humans are not very good at making procedural knowledge declarative -- it takes a rare human to be able to explain and understand how they do something they know how to do well. There is a real algorithmic difficulty here, but even so, I think a lot of the difficulty that humans have in doing this is unnecessary, i.e. a consequence of the particular way the brain is structured rather than a consequence of the (admittedly large) difficulty of the problem involved. I disagree that we have a problem converting procedural to declarative for all domains. As an example, I can retrieve a phone number from procedural memory with 1 retrieval operation (watch my fingers dial it). Admittedly this system isn't as slick as one that would work purely internally, it requires performance of the task, but it works. Grammar is tougher, I can test any given rule by using it out in a sentence and seeing how it sounds. But extrapolating all of the rules I use is a tricky problem, in fact it's one we haven't completely finished solving (the rules of English grammar are similar, but not identical to the rules our brains want to use). And then communicating how to swing a golf club is another matter, but I think the limitation there lies in a lack of communication. Our brains have no good way of transmitting or interpreting such fine grained information. And to be fair to our brains, transcribing a motor memory of how to move 10,000 muscles in a very precise sequence into declarative knowledge is an extremely challenging problem. Particularly because that sequence isn't static, but requires feedback from joint sensors. The information isn't just the sequence of neural impulses, it's the substance of the entire network. That said, Novamente would be far better at it than we. With the ability to understand it's own code, NM could just rattle off the relevant parameters into declarative memory. Making this declarative knowledge useful would require understanding how it functions though. That would be the tricky part. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] seed AI vs Cyc, where does Novamente fall?
Indeed, making the declarative knowledge derived from rattling off parameters describing procedures useful is a HARD problem... but at least Novamente can get the data, which as you have greed, would seem to give AI systems an in-principle advantage over humans in this area... It's hard to overestimate the intelligence-enhancement potential of a more fluid process of interconversion btw declarative and procedural knowledge Yes, getting this data is what the entire field of neurophys is about. Being able to extract it without using surgery, electrodes, amplifiers, and gajillions of manhours would be outstanding. A lack of data is the primary thing holding neuroscience back and to a large degree, the depth of cognitive theory over time mirrors the quality of the acquisition and analysis tools. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] seed AI vs Cyc, where does Novamente fall?
That was exactly my impression when I last looked seriously into neuroscience (1995-96). I wanted to understand cognitive dynamics, and I hoped that tech like PET and fMRI would do the trick. But nothing existing giving the combination of temporal and spatial acuity that you'd need to even make a start on the problem I had a PhD student (Graham Zemunik) Just FYI, MEG's (Magnetoencephalography) is a good step in providing temporal precision, but is still a long way from discerning individual neurons. It can basically give us EEG measurements from deep inside the brain without using electrodes(which obviously opens alot of doors for human experimentation) who tried to make a detailed model of the cognitive dynamics in a cockroach's brain -- and even that was pretty dicey because the data found by different researchers was often inconsistent. From what you're I'm sure you know this, but for the benefit of others: Insect brains are much easier to study because the neurons are explicitly laid down by the genetic code. They are identifiable neuron by neuron and are roughly identical from insect to insect (within the same species). The fact that even these networks aren't quite yet understood is a shining example of how far we have to go in understanding the human brain. describing, some headway is finally being made on modeling cognitive dynamics in parts of the rat's brain, and that's a great thing. I've enjoyed following Walter Freeman's work on olfaction in rabbits, but, I've also noticed the pattern of bold hypotheses and partial retractions in his work over time, which is due to the fact that the data is not quite rich enough to support the kind of theorizing he wants to do. I support fringe theorists like Freeman as long as they stay in touch with the community and don't sail off to parts unknown. (Edelman tends to do this). Progress takes all types, the careful, methodical data collectors, and the people on the front lines pushing the theories to extents that the data barely supports. Fortunately, neuro-analysis technologies are advancing really fast just like computer chips. In another 10-30 years we will have the data to understand our brains, and the computers and algorithms to crunch this data. (And we may have AI's to do the work for us, who knows ;) Here's hoping. Although I fear they probably said similar things 10-30 years ago. Only nanotech can get us the type of noninvasive, detailed data that we need. The type of electrodes we currently use are never going to suffice. Lucky for us that the brain uses electrically recordable signals from a structure that is so easily accessible. We'd be in dire straits if the brain used entirely chemical mechanisms and was located in an abdominal sack. Thank you evolution for making our jobs as easy as they are :) --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] seed AI vs Cyc, where does Novamente fall?
I actually have a big MEG datafile on my hard drive, which I haven't gotten around to playing with. It consists of about 120 time series, each with about 100,000 points in it. It represents the magnetic field of someone's brain, measured through 120 sensors on their skull, while they sit in a chair and perform an experiment of clicking a button when they see a line appear on a screen. (Pretty exciting, huh?) I got the data from my friend Barak Pearlmutter at UNM, who has spent a few years working on signal processing tools (using blind source signal separation methods) designed to clean up the raw data (basically subtracting off for noise caused by repeated reflection of magnetic fields off the inside of the skull). It's actually a very complicated data analysis. You basically have a spherical surface of data (the sensors), and you are trying to reconstruct the sources and sinks in 3d that created the 2d data you are observing. The problem is underconstrained, because many 3d data sets could produce the same 2d data set, but you try to build in some anatomical assumptions (ie: we know the hippocampus is probably a powerful source/sink, so pin that thumbtack there) to constrain the possible results. As you can imagine it's very weak spatially, but far more precise temporally than PET or FMRI, which can only measure blood flow changes occuring 1 second or more after the source activity. I think combined MEG/FMRI(or was it PET/FMRI) is going to be able to get the best of both worlds. Either way, there are plenty of technological obstacles. I guess that MEG can be used, over time, for stuff subtler than clicking buttons when lines appear. But using it to track the dynamics of thoughts seems a long way off Basically, one needs a lot more than 120 sensors!! ... and then one needs to hope the signal processing code scales well (it probably can be made to do so) Well you can use far more complex behavioral tasks than that even with existing MEG technology(have people navigate a maze, do math, word problems, etc). But in order to get a footing with the new MEG technology, they need to start at the basics so that they can map MEG responses with known EEG signatures available from work that's already been done. The first decade of any new neurophys technique is characterized by a whirlwind of very basic, boring results (usually that create pretty pictures generating funding). Only after the tech has matured do you even begin to hit the cool stuff. I'll bet AI's will be required to analyze the data sets will be getting in the next 20 years. -brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] swarm intellience
The limitation in multi-agent systems is usually the degree of interaction they can have. The bandwidth between ants, for example, is fairly low even when they are in direct contact, let alone 1 inch apart. This limitation keeps their behavior relatively simple, simple relative to what you might expect for the large neural mass involved. Also, swarms only scale to a limited degree. An anthill 1 mile high is not going to possess much more smarts than a 3 inch anthill. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] swarm intellience
But hopefully the bandwidth of communication is compensated by the power of parallel processing. So long as communication between ants or processing nodes is not completely blocked, some sort of intelligence should self-organize, then its just a matter of time. As programmers or engineers we can manipulate those communication channels ... Interesting stuff today, I wish I wasn't so insanely busy with this thesis. I don't think swarms are a good way to approach AGI design because of the design principles. An infinite number of AGI designs are possible, but some are going to be drastically easier to create and cheaper to run than others. The advantage of the swarm is to be robust in the face of severe damage. Wipe out 3/4 of it, and the remainder will function perfectly well and eventually regrow. Even a single element is a functional entity (although not necessarily self-sufficient). The brain certainly does not share this ability beyond preserved functionality in the face of a hemispheric lesion. Orient your lesion perpendicular to the axis of symmetry and you have a vegetative organism. The advantage the brain gets for being so tightly integrated is greatly enhanced functionality. Because we(in general) are not planning to design AGI's to survive in the face of 75% damage, swarms seem to be an inefficient approach. Although obviously there are domains in which this sort of resilience is a huge advantage, such as in the deployment of a network of robotic sensory drones in a hostile battlefield. I don't think they will ever have AGI like behavior however. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] seed AI vs Cyc, where does Novamente fall?
Just to pick a point, Eliezer defines Seed AI as Artificial Intelligence designed for self-understanding, self-modification, and recursive self-enhancement. I do not agree with you that pure Seed AI is a know-nothing baby. I was perhaps a bit extreme in my word choice, but I do not believe that the axis I mentioned is orthogonal to the question of hand-wiring of a knowledge base. Certainly some concepts and tinkering must be included in a seed-AI, but I think that the representations a seed-AI will develop are one and the same as the knowledge that Cyc will require to be hand-made. This isn't really a semantic argument as it might seem at first glance, rather it's a question of the degree to which knowledge is separable from the internal machinery that a seed-AI will construct in its growth. But that's just my opinion, it's been known to change before :) -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] more interesting stuff
Cliff wrote: It's not a firm conclusion, but I'm basing it on information / complexity theory. This relates, in certain ways, to ideas about entropy -- and energy is negentropy. I.e. without the sun's input we would be nothing. I'm not convinced of this idea on an intuitive basis, but rather on a mathematical basis -- that is, the mathematical idea that complexity cannot be freely produced. You cannot get truly random numbers out of a fixed process would be another way to state this. No, but you can get pseudorandomness from energy. I would say that we ourselves are, at best, pseudorandom as well. True randomness is a tricky thing to get. I also wonder if AGI's could really be trained in a simulated micro- environment...perhaps the real universe's randomness is *necessary* for development of intelligence. How? Your ideas are interesting, but I don't see the need for them. We understand the core principles of evolution, we see it in action and it makes intuitive sense. We also know that raw energy input can lead to an increase in evolution. We have all the necessary and sufficient conditions for evolution to have worked, there's not really a missing piece to fill in. Consider the following thought experiment: a computer able to simulate the earth down to an atomic level (let's put aside the possibility that quantum phenomena influence events on the scale of earth-life). This system has 1 source of input, a constant stream of energy. The machine is simple, runs on a turing machine. Do you doubt that this machine could recreate evolution in its simulation? If you do not, then we're all done here, as this machine is completely isolated and receives no complexity input. If you do doubt it, what is the missing piece that you are trying to fill? Well, that's what I'm wondering about. Does co-evolution increase the total complexity (in a mathematical, Kolmogorov sense) or does it *mirror* or *absorb* complexity. Lately, I suspect the latter -- you cannot really increase it, you can only reflect it. Energy-complexity in a closed system. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] more interesting stuff
Kevin said: I would say that complex information about anything can be conveyed in ways outside of your current thinking, but if you ask me to prove it, I cannot. There is evidence of it in things like the ERP experiment which show the existence of a possible substrate that we have not yet measured or verified... Which experiment? I'd like to hear about it. As far as I know, there is yet to be found any conclusive evidence of a cognitively releveant substrate in the brain that we have not measured or verified. Nothing in neuroscience data, that I am aware of, cannot be explained by cellular interactions at the atomic+ scale. Or did you not refer to Event-Related Potentials? The ERP acronym has multiple connotations. Question: the big bang occured in a closed system, yet the information for every phenomena we witness was the result of that occurance. How was that information stored? How did it get promulgated? Information did not have to be stored for interesting things to develop. I don't think, for example, that you would find the text for the Constitution of the United States hiding somewhere in the big bang proto-particle. Simple systems can give rise to complex series of information. One can posit that quantum randomness (Shroedinger's cat) can imply the universe is non-predictable from its start state, but that doesn't mean quantum phenomenae are transmitting complex and significant information across some unverified substrate. It just means that quantum randomness can occasionally push things one or another, as in the butterfly wing analogy, in which a small change at one time can drastically alter the future. But I fear we are now getting pretty far from the AGI list's manifesto. _brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] really cool
They are not mapping to IP addresses, probably geography as Ben suggests. I went to the search window and intercepted searches done by other people. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
[agi] seed AI vs Cyc, where does Novamente fall?
Ok, let's get rolling then. Ben, here's a question. To what extent are parts of Novamente hand built ala Cyc? I can easily imagine a dimension here. At one end is Cyc, which is carefully and meticulously constructed by people. The design work is two fold, creating the structure within which to store information, and then the inputting of the information. At the other end of the dimension is pure Seed AI. A know-nothing baby. You put it together, hit the power button and let it grow(admittedly it would require tuning as it matured). Where would Novamente fall along this dimension? What domains of knowledge do you expect to have to explicitly create by hand? -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
[agi] the Place system of the rodent hippocampus
I whipped this up this afternoon in case any of your are interested. I tried to gear it towards functionally relevant features. Enjoy Reference document: The Hippocampal navigational system by Brad Wyble A primer of neurophysiological correlates of spatial navigation in the rodent hippocampus. Why AI enthusiasts should care: The place system is a unique way to study what the rat is thinking and how it uses information to compute. Place cells represent a particular way that the rodent brain analyzes spatial location in a way that is cognitively accessible to us. The behavior of place cells are relatively homogenous across the population. Contrast this to recordings from frontal cortex in which cellular activity is extremely varied. Frontal cortical cells are doing very interesting things with respect to behavior, but they are very different from one another which makes it practically impossible to draw conclusions. The structural and functional simplicity of the hippocampus makes it a gateway to understanding the brain, a strong foothold for our first significant steps. It is largely a happy accident that the place system is so easy to study. The hippocampus is arranged in horizontal sheet of very dense cell near the surface of the skull (in rodents at least) which means that it is possible to get yields of 200+ cells *simulteanously* within one rat using current technology, a feat achievable in no other brain area by at least an order of magnitude. This high cell yield allows us to study the behavior of the entire system in the same way the Nielson system studies the television viewing habits of the entire country using data from a tiny fraction of homes. Place Cell The place cell was described in the 70's by a group led by John O'Keefe (O'Keefe, 1976) and is the foundation for our understanding of the rodent hippocampal navigational system. A place is a hippocampal neuron(pyramidal, complex spike cell of CA1/CA3/Dentate) that will fire reliably and selectively within a small region of space. The firing pattern for that region of space (usually about 10cm in diameter but varies with the size/shape of the environment) is roughly a 2-dimensional bell-curve if the cell spikes are compiled into a histogram with respect to 2-d location.This region of space is called a place field and is defined with respect to a specific neuron in a specific environment. The particular configuration of place fields across all place cells for a given environment is called a place map. One neuron can have multiple place fields within a single environment, but this is rare. Environment: The concept of what constitutes an enviornment is vital to this discussion. For experimental purposes, a rat is introduced into a chamber it has never seen before. A seemingly arbitrary place map develops over 10-15 minutes. This same rat placed in a different environment will immediately develop an entirely different and *uncorrelated* place map for the new environment. There is alot of complexity in figuring out what constitutes a different environment. Alterations in the geometric shape (square- circle) almost always generate a new map. Variations in visual cues will sometimes cause a remapping, sometimes not. Rats remap in all or none fashion. That is to say, as visual cues are altered, the map will stay constant until some arbitrary threshold in passed, at which point the entire population will remap. Map alterations do not cause gradual shifts in the field. Extreme alterations in behavior can cause a remapping. If a rat is trained to do two different tasks(targetted vs random foraging) in the same environment, it will usually develop two different place maps and switches between them based on the task. Sensory Cue control. In environments for which multiple orientations are available (square, cylinder), the place map will align itself with the most obvious visual cues. If a cylinder has a cue card on the wall, and the card is shifted, the place map will follow the card. The place map is largely immune to the removal of cues. If the lights are turned off and no visual cues are available, the rat will continue to use the same place map. It uses a combination of vestibular and kinesthetic cues to integrate its motion, and keep mental track of its position (as evidenced by the preserved functionality of the place map and behavior). It can use olfactory(excrement) and tactile cues to correct for drift error in the path integration process. This is demonstrated by using environments that allow for no olfactory cues by wiping the environment with alcohol and turning off the lights. The place map is stable with respect to the cylinder walls, but drifts in orientation over time because there ! are no olfactory cues to control for rotational drift, while contact with the walls controls for radial drift. Generally visual cues
Re: [agi] the Place system of the rodent hippocampus
On the face of it, these place maps are very reminiscent of attractors as found in formal attractor neural networks. When multiple noncorrelated maps are stored in the same collection of neurons, this sounds like multiple attractors being stored in the same formal neural net. Yeap, there's well developed theories about how an autoassociate network like CA3 could support multiple, uncorrelated attractor maps and sustain activity once one of them was activated. The big debate is about how they are formed. About the ability to study 200 neurons at once: With what time granularity can this be done? Do there exist time series of the activity of these 200 Raw data is usually acquired at 50+ Khz, and then the spikes are identified as to which neuron they belong to and are stored in a reduced form (ie spike X of neuron Y occurs at time T) neuron, both during map learning and during map use? Analyzing this 200-dimensional time series would be interesting. (Not that I have time to do it .. but it would be interesting.) We are currently using Novamente to They're working on it. At present such labs are acquiring data faster than they can analyze it. Figuring out how the maps form is a tricky business because you can only sample the formation of a place field when the rat is in it. Consequently the data is very sparse during the formation. They are making progress though. analyze coupled time series in another biological domain (gene expression data). If there is decent time series data, it could be interesting to codevelop a grant application with someone to see what Novamente can find in this data Very interesting idea. The lab with most of this data is the McNaughton lab in Arizona. They are somewhat reluctant to give it out though, because of the money and time investment in collecting it. It would be very cool if Novamente could be applied to it though. On a more philosophical note, I like the idea that the machinery used for place mapping in rats is similar to the machinery used for more abstract sorts of mapping in humans. Indeed, this reflects the point someone made last week on this list, regarding the fact that humans have much better reasoning ability in familiar domains than unfamiliar ones. Maybe one of the reasons is that when we know a domain well we figure out how to map the domain into a physical-environment metaphor so we can use some physical mapping machinery to reason about it. But some familiarity is needed to create the map into the physical-environment metaphor. I think this is what someone suggested last week -- and your essay makes me like the hypothesis even more. That was me. It will be awhile before we have such human data of course, but they are starting to record from human hippocampi (in eplileptic patients). I'm a big fan of using landscape analogies to reason about problems, it tends to work well for me. But I wonder if such abilities are more reliant on visuospatial areas of the cortex. One of the limitations that strikes me is that of dimensionality. I used to spend time while driving on road trips trying to think in 4-dimensions in a similar way that I can visualize 3-d. I just couldn't get it to work well. The best I could come up with was layers of 3-d representations with 1 feature varying. This is an excellent example of how powerful our minds are at certain kinds of computation, but limited outside of our innate domains. I am reminded a bit of some management-consulting ideas developed by my friend Johan Roos, see e.g. http://www.seriousplay.com/images/landscapes.pdf His work explores the notion of knowledge landscapes, and the use of the physical-landscape metaphor in human thinking about business. I'll check it out, thanks. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] the Place system of the rodent hippocampus
Yeap, there's well developed theories about how an autoassociate network like CA3 could support multiple, uncorrelated attractor maps and sustain activity once one of them was activated. The big debate is about how they are formed. The standard way attractors are formed in formal ANN theory is via variants of Hebbian learning. But pure unsupervised Hebbian learning has never worked very well in simulations. In the CS theory of reinforcement learning, a lot of tricks have been used to make Hebbian learning work better (temporal difference learning, for example), but none of these work that awesomely. Using artificial rules, such as hardball winner-take-all and synaptic weight normalization, it's doable to get ANN's to do this. But in an autoassociative network with realistic biophysical properties, controlling activity to prevent runaway synaptic modification is a very large problem. My own grad advisor, Mike Hasselmo has worked on this very problem using pharmacological modulation to suppress synaptic transmission during learning. The fact that epilepsy usually starts within the hippocampus (with its sheets of 100k neurons, all interconnected with excitatory connections) indicates that this is a real problem for the brain as well as models of it. I imagine the hippocampus is really pushing the evolutionary envelope in terms of being prone to epilepsy. Demand for more memory is probably fighting directly against epileptic tendencies in terms of evolutionary fitness. Another problem is feeding that dense sheet of nerves (which is why the hippocampus is one of the first things to suffer damage during anoxia). It's a very specialized area that's pushing the limits of the body's ability to feed it and keep it from siezing up. Do you think the spike-time data contains enough information that it's not necessary to look for patterns in the raw data? They keep the real data too, but it's *huge* (100+ channels of 70khz data, realtime). The raw data is basically an average of the neural activity of the nearby cells. Spikes from neurons within a small radius of the electrode tip stand out and have a certain characteristic shape/amplitude, which is used to identify said cell. Apart from identifying spikes, I'm not sure you'd get much out of the raw data(assuming you are also collecting EEG data realtime at 10khz or so from one electrode in the nearby region). However, nowadays people are starting to worry about complex spikes too (bursts of spikes). Assigning these spikes to their source neuron is much harder because spikes after the first one in a burst are reduced in amplitude. So you need specialized clustering algorithms that are aware of bursts and what they do to a spike amplitude. You need to go back to the raw data to identify such bursts every time you change your detection algorithm. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Goal systems designed to epxlore novelty
Novelty is recognized when a new PredicateNode (representing an observed pattern) is created, and it's assessed that prior to the analysis of the particular data the PredicateNode was just recognized in, the system would have assigned that PredicateNode a much lower truth value. (That is: the system has seen a pattern that it did not expect to see.) So you're saying a newly formed PredicateNode normally has a low truth value, but PN's about novelty tend to have abnormally higher truth values? Or is it: novel Predicatenodes tend to have lower than normal truth values? Novelty is recognized when a map (a set of Atoms that share a coherent activity pattern) is formed, which is dissimilar to any previously existing maps. Are you familiar with the place cell system of the hippocampus as found in rats? I'll give you a brief synopsis in a new subject in case there's any ideas that you find useful. It should be noted that the rules for recognizing novelty are similar to the rules for mentioning learning. However, novelty focuses on the suddenness of changes in truth value, whereas learning focuses on the total amount of changes in truth value. The two are similar conceptually but different quantitatively. Interesting idea, I'm still unclear about the specifics of how truth relates to novelty, but I get general idea. I'll wait for the nicer review article and leave you to your work. Thanks -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] A probabilistic/algorithmic puzzle...
1) Humans use special-case algorithms to solve these problem, a different algorithm for each domain 2) Humans have a generalized mental tool for solving these problems, but this tool can only be invoked when complemented by some domain-specific knowledge My intuitive inclination is that the correct explanation is 2) not 1). But of course, which explanation is correct for humans isn't all that relevant to AI work in the Novamente v Lakoff and Nunez (http://perso.unifr.ch/rafael.nunez/reviews.html) have a theory that we compare lengths in our head to do arithmetic, when we're not using school-learned rules. Our innate mathematical ability is based on visuo-spatial comparisons in their view. This would basically be #2, and to use this capability we need to get familiar enough with the problem that our mind translates the numbers involved into length. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: Re: [agi] A probabilistic/algorithmic puzzle...
Hi Ben, Thanks for the brain teaser! As a sometimes believer in Occam's Razor, I think it makes sense to assume that Xi and Xj are indepenent, unless we know otherwise. This simplifies things, and is the rational thing to do (for some definition of rational ;-). So why not construct a bayes net modeling the distributions, with causal links only where you _know_ two variables are dependent? For reasoning about orphan variables (e.g., you know nothing at all about Xi), just assume the average of all other distributions. If you have P(Xi|Xj), and want P(Xj|Xi), fudge something together with Bayes' rule. This isn't a complete solution, but its how I would start... Is this one of the things you've tried? Cheers, Moshe As Pei Wang said: Intelligence is the ability to work and adapt to the environment with insufficient knowledge and resources. I think this is a core principle of AGI design and that a system that only makes inferences it *knows* are correct would be fairly uninteresting and incapable of performing in the real world. The fact that the information in the P(xi|xj) list is very incomplete is what makes the problem interesting. Or maybe I'm misinterpreting your intent. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] A probabilistic/algorithmic puzzle...
This is also an example of how weird the brain can be from an algorithmic perspective. In designing an AI system, one tends to abstract cognitive processes and create specific processes based on these abstractions. (And this is true in NN type AI architectures, not just logicist ones.) But evolution is a hacker sometimes: often, rather than abstracting, it reuses stuff that was created for another purpose, providing hacky mappings to enable the reuse. This is terrible software engineering practice, but evolution has a lot of computational resources to work with, and it does create a lot of buggy things ;) The study of historical constraints on evolution's design's principle is fascinating. I took a class with this guy: http://www.mcz.harvard.edu/Departments/Fish/kfl.htm, and he focusses on very interesting problems within systems that would seem to be very boring (the evolution of jaw structures in cichlids). For example, consider hemoglobin, the current means of transmitting oxygen in the body. There might be a better way to do it, in fact, it's almost certain that there is. But evolution would have a very hard time finding it, because we're already heavily invested in the hemoglobin tract. The same thing applies to the brain of course, evolution has invested alot of effort into developing sensory and motor facilities. Logic and reason are crude hacks, tacked on top of a system designed to do nothing of the sort. It's like figuring out how to attach a swimming pool to the space shuttle. (and miracle of miracles, it somehow works, albeit crudely). Small wonder that we are so terribly bad at logic. http://plus.maths.org/issue20/reviews/book1/ Interestingly, there are some primitive parts of our brain that are better at logic and are more rational than our executive function. Animals (and humans) in a classical conditioning paradigm are *excellent* at performing simple behaviors in a way that maximizes reward. We can determine the proper ratio of performance on a two lever task without even being consciously aware of the contingencies. Rats can do this too. In fact, sometimes our advanced forebrain gets in the way of our more primitive structures trying to do what they do best. This is probably why people gamble and play the lottery. I would guess that the payoff matrices for all forms of casino gambling are too subtle and complicated for our primitive rationality agents to comprehend, and so the stupid forebrain gets to have its way. -brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/a href=/faq.html#spam[EMAIL PROTECTED]/a
Re: Re: Re: [agi] A probabilistic/algorithmic puzzle...
Brad wrote: I think this is a core principle of AGI design and that a system that only makes inferences it *knows* are correct would be fairly uninteresting and incapable of performing in the real world. The fact that the information in the P(xi|xj) list is very incomplete is what makes the problem interesting. Or maybe I'm misinterpreting your intent. I agree perfectly with your core principle, and my proposal was not to only make inferences that you know are correct. I think you may be misinterpreting: lets say that we know P(Xi), and want to guess at P(Xi|Xj). We have insufficient knowledge, so we need to make some assumptions to approximate P(Xi|Xj). I argue that under these circumstances, the best assumption to make is that Xi and Xj are independent, (ie, P(Xi|Xj)=P(Xi)). Does this clarify things? You are basically saying, for each unknown P(Xi|Xj), assume it equals P(Xi). I think this conservative approach, while well grounded in rationality, doesn't really allow for the existence of useful and interesting inference. An AGI has to tolerate, and work with, large degrees of uncertainty. This includes assuming dependencies without sufficient evidence. I can say that in the biological sciences, one has to do this constantly. What separates the good scientists from the not-so-good is an ability to keep track of many low-confidence assumptions simultaneously, shake them up and see what theories fall out that violate the fewest of them. -Brad Moshe -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] A probabilistic/algorithmic puzzle...
But anyway, using the weighted-averaging rule dynamically and iteratively can lead to problems in some cases. Maybe the mechanism you suggest -- a nonlinear average of some sort -- would have better behavior, I'll think about it. The part of the idea that guaranteed an eventual equilibrium was to add decay to the variables that can trigger internal probability adjustments(in my case, what I called them truth). Eventually the system will stop self-modifying when the energy(truth) runs out. The only way to add more truth to the system would be to acquire new information via adding goal nodes for that purpose. You could say that the internal conistency checker feeds on the truth energy introduced into the system by the completion of data-acquisition goals(which are capable of incrementing truth values). This should guarantee the prevention of infinite self-modification loops. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
[agi] Goal systems designed to epxlore novelty
The AIXI would just contruct some nano-bots to modify the reward-button so that it's stuck in the down position, plus some defenses to prevent the reward mechanism from being further modified. It might need to trick humans initially into allowing it the ability to construct such nano-bots, but it's certainly a lot easier in the long run to do this than to benefit humans for all eternity. And not only is it easier, but this way he gets the maximum rewards per time unit, which he would not be able to get any other way. No real evaluator will ever give maximum rewards since it will always want to leave room for improvement. Fine, but if it does this, it is not anything harmful to humans. And, in the period BEFORE the AIXI figured out how to construct nanobots (or coerce teach humans how to do so), it might do some useful stuff for humans. So then we'd have an AIXI that was friendly for a while, and then basically disappeared into a shell. Then we could build a new AIXI and start over ;-) This is an interesting aspect to the problem. Evolution has designed a fairly robust reward system, one that encourages us to achieve interesting through our lives and acquire knowledge in an interesting way. Yet even it is vulnerable to short-cutting the reward system as seen in addictive behaviors. Ben, I'm guessing you've thought alot about how to structure the reward/goal system of Novamente. I'd love to hear more about it. It seems that designing a system that forces itself to expand its knowledge base is a fairly non-trivial task. We as entities (demonstrated also in rats) have a certain prediliction for exploring novel situations, environments, objects, ideas, etc. Have you implemented a similar drive for seeking novelty in Novamente? -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] Breaking AIXI-tl
Now, there is no easy way to predict what strategy it will settle on, but build a modest bunker and ask to be left alone surely isn't it. At the very least it needs to become the strongest military power in the world, and stay that way. It might very well decide that exterminating the human race is a safer way of preventing future threats, by ensuring that nothing that could interfere with its operation is ever built. Then it has to make sure no alien civilization ever interferes with the reward button, which is the same problem on a much larger scale. There are lots of approaches it might take to this problem, but most of the obvious ones either wipe out the human race as a side effect or reduce us to the position of ants trying to survive in the AI's defense system. I think this is an appropriate time to paraphrase Kent Brockman: Earth has been taken over 'conquered', if you will by a master race of unfriendly AI's. It's difficult to tell from this vantage point whether they will destroy the captive earth men or merely enslave them. One thing is for certain, there is no stopping them; their nanobots will soon be here. And I, for one, welcome our new computerized overlords. I'd like to remind them that as a trusted agi-list personality, I can be helpful in rounding up Eliezer to...toil in their underground uranium caves http://www.the-ocean.com/simpsons/others/ants2.wav Apologies if this was inapporpriate. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] doubling time watcher.
I would like to contribute new SPEC CINT 2000 results as they are posted to the SPEC benchmark list by semiconductor manufacturers. I expect to post perhaps 10 times per year with this news. This is the source data for my Human Equivalent Computing spreadsheet and regression line. I'm uncomfortable with the phrase Human Equivalent because I think we are very far from understanding what that phrase even means. We don't yet know the relevant computational units of brain function. It's not just spikes, it's not just EEG rhythms. I understand we'll never know for certain, but at the moment, the possibility of guesstimating within even an order of magnitude seems premature. This isn't to say that the regression isn't a bad idea, or irrelevant to AGI design. I just don't like the title. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] doubling time watcher.
Brad writes, Might it not be a more accurate measure to chart mobo+CPU com= bo prices? Maybe. If you wanted to research and post this data I'm sure it would be = helpful to have. Check out www.pricewatch.com. They have a search engine which ranks products by vendors. Using this, you could get lots and lots of data from one source. By averaging mean prices from the top 10 cheapest vendors, you'd wash out wierd one-time price break deals that would pollute your data if you only considered the cheapest. They also have data for complete systems. It's also probable that pricewatch keeps archived data of prices. You might consider emailing them. Finding a smart techie in their NOC who thinks AI is cool and you might get your hands on 5+ years of perfect data on every index of computing power. CPU, hard drives, tape storage, RAM, everything. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] doubling time watcher.
I used the assumptions of Hans Moravec to arrive at Human Equivalent Computer processing power: http://www.frc.ri.cmu.edu/~hpm/ Of course as we get closer to AGI then the error delta becomes smaller. I am comfortable with the name for now and will adjust the metric as more info becomes available. The error delta depends more on neuroscience research than AGI progress. I'm not comfortable with Moravec's calculations, but his approach of estimating based on retinal processing power is better than anything else I've read on it. Retinal neurons aren't quite the same beasts as the enormous pyramidal's that make up much of the brain though. This isn't to say that the regression isn't a bad idea, or irrelevant to AGI design. I just don't like the title. -Brad Oops, I meant to that This isn't to say that the regression *is* a bad idea. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: AGI Complexity (WAS: RE: [agi] doubling time watcher.)
The brain is actually fantasticly simple... It is nothing compared with the core of a linux operating system (kernel+glibc+gcc). Heck, even the underlying PC hardware is more complex in a number of ways than the brain, it seems... The brain is very RISCy... using a relatively simple processing pattern and then repeating it millions of times. Alan, I strongly suggest you increase your familiarity with neuroscience before making such claims in the future. I'm not sure what simplified model of the neuron you are using, but be assured that there are many layers of complexity of function within even a simple neuron, let alone in networks. The coupled resistor/capacitor model is only given as a simplified version in textbooks to make the topic of neural networks digestible to the entry-level student. Dendrites are not simple summators, they have a variety of nonlinear processes including recursive, catalytic chemical reactions and complex second-messenger systems. That's just the tip of the iceberg once you get into pharmacological subsystems, the complexity becomes a bit staggering. If it were fanastically simple, more so than a Linux box, do you think that thousands of scientists working over more than one hundred years would still understand it so poorly, yet it takes a group of 5 people 2 years to crank out a new Linux OS? We know from the biology folks that the human mind contains at least dozens, and probably hundreds of specialized subsystems. In the cortex, I would propose the number is 28 for the left hemisphere, and maybe another 10 or so in the right hemisphere which don't directly overlap with the ones on the left. You realize that the blobs drawn on images of the brain in college level textbooks are simply areas of cell responsivity, and not diagrams of the systems themselves? The cortex is highly differentiated containing probably dozens if not hundreds of systems, not to mention the enormous variety of specialized systems at the subcortical level. The complex soup of the reticular formation is sufficient to turn a sane anatomist into a sobbing wreck with its dozens of specific nerve clusters. Consider the chess problem. The present computer Chess solutions are widely acknowleged to be much less efficient than the ones in the brain. So the complexity that you are trying to argue is necessary for AGI is merely reflective of our currently poor programming methodologies. Chess is a game designed by the mind, so it is no surprise that it is something the mind is good at. It is trivial to design games that computers are vastly superior at, but that does not mean the mind has poor programming methodologies. _Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: AGI Complexity (WAS: RE: [agi] doubling time watcher.)
Well, we invented our own specialized database system (in effect) but not our own network protocol. In each case, it's a tough decision whether to reuse or reimplement. The right choice always comes down to the nasty little details... The biggest Ai waste of time has probably been implementing new programming languages, thinking that if you just had the right language, coding the AI would be SO much easier. Ummm... The thing that gives me the most confidence in you Ben is that you made it to round 2 and you're still swinging. You've personally learned the hard lessons of AGI design and its pitfalls that most of the rest of us can only imagine by analogy. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: AGI Complexity (WAS: RE: [agi] doubling time watcher.)
[META: please turn line-wrap on, for each of these responses my own standards for outgoing mail necessitate that I go through each line and ensure all quotations are properly formatted...] I think we're suffering from emacs issues, I'm using elm. Iff the brain is not unique in its capability to support intelligence then all of this can be replaced by some abstract model with the same basic computational charactaristics but in a very different way. I totally agree. But the genesis of this debate was whether the brain is complicated in a non-trivial way. The fact that it is complicated does not mean it cannot be replicated in a different substrate (and like Ben, I think it would be a misapplication of effort to try). The implementation details are what tells you how the brain functions. I don't care _HOW_ it functions, I care about _WHAT_ a given section accomplishes through its functioning. The nature of neuroscience research doesn't really differentiate between the two at present. In order to understand WHAT a brain part does, we have to understand HOW it, and all structures connected to it function. We need to understand the inputs and the outputs, and that's all HOW. There are people who approach the problem from a purely black-box perspective of course, by giving people memory tests and looking at the pattern of failures. This is extremely interesting work, particularly as regards the types of errors people make while speaking. (http://www.wjh.harvard.edu/~caram/pubs.htm) I don't think it's sufficient, on its own, to figure out the brain without simulteanously looking at the neural data. Given that, it should be relatively streight forward to find a work-alike well, it just isn't. Brains are hard to reverse engineer, and that's basically what you're talking about. Failing that, it is still possible to set up a system akin to Creatures but with a much more powerful engine and wait untill a good'nuff algorithm evolves on its own... It took evolution billions of years with an enormous search space. Obviously we can speed the process. But in the end, you'd end up an equally inscrutable mass of neural tissue. You'd be better off getting yourself a real kid :) rant mode engaged I HATE IVORYTOWERISM!!! IF A BOOK DOESN'T TELL IT LIKE IT IS, IT SHOULD NEVER BE PUBLISHED, EVEN TO LITTLE CHILDREN!! (Especially not to little children.) My comment was in the context of you saying that the brain is fantastically simple and then citing Calvin as a source for your conclusion. I'm saying that books by pop authors are insufficient to draw conclusions from, not that they are useless. His ideas are great, I love his work. The brain does have an innate structure in the form of the topology I mentioned earlier. This topology naturally leads to the development of functional systems. HOWEVER, there is no law in the *cortex* which governs what behaviors it will produce (likes, dislikes etc...) these must be inputed either from the environment or from the subcortical structures. I disagree with this, but I see where you are coming from. We don't know enough about the cortex to say things like this. The reason that subcortical structures seem more concrete to us, is that they are simpler in design and therefore easier to understand than cortical structures. Yes, and I don't think those varriations in layers or even connectivity are at all significant. Ofcourse you want to know which layer is for input and which layer is for feedback but you don't really worry yourself about the measurements which are probably a biproduct of having more neurons in those regions that are heavily connected and not, in themselves, interesting... The extray layers in the occipital lobe are probably nothing more than the equivalent of a math coprocessor in a computer... The addition or deletion of layers is going to drastically change the nature of computations a given bit of cortex performs. I've spent 8 years studying hippocampal anatomy. It is fascinating and highly structured in a way the cortex isn't (or its simplicity allows us to perceive the structure). Vast volumes of data about its anatomy are available and I have read most of it. GIMME GIMME GIMME!!! =P I said I read it, I didn't say I could remember all of it :) I( and the rest of the hippocampal community) am at a loss to tell you how it functions. Do we know what it does? (how its outputs relate to its inputs) Nope. We think it might have to do with spatial navigation in rodents (rats tend to think in terms of 2-D space) and more complex types of memory in higher order critters. Anatomy and neurophysiology seem to suggest it should relate memory to motor actions and behavioral states, but lesion it and animals seem relatively unimpaired in that respect(lesions are a troublesome way to reverse engineer the brain). *throws up
Re: AGI Complexity (WAS: RE: [agi] doubling time watcher.)
Not exactly. It isn't that I think we should give up on AGI, but rather that we should be consciously planning for it to take several decades to get there. We should still tackle the problems in front of us, instead of giving up on real AI work altogether. But we need to get past the idea that every AI project should start from scratch and end up delivering a human-equivalent AGI, because that isn't going to happen. We just aren't that close yet. The way the software industry has solved big challenges in the past is to break them up into sub-problems, figure out which sub-problems can be solved right now, solve them as thoroughly as possible, and offer the resulting solutions as black boxes that can then become inputs into the next round of problem solving. That's what happened with operating systems, and development environments, and database systems. If we want to see real progress in AI, the same thing needs to happen to problems like NLP, computer vision, memory, attention, etc. In as much as I'm a neurophile, I disagree that this is the best approach. AI research has been having a hard time making progress by working on little black boxes and then hooking them together. I think without the context of the whole entity (the top level AGI), it's harder to think about and implement solutions to the black box problems. Evolution certainly didn't work with black boxes. It made functionally complete organisms at each step of the way, and I think AI design can work in the same manner. The progress of bottom-up, whole organism roboticism, ala Rod Brooks, is an impressive example of what can happen when you attack the whole organism simultaneously. The top level thinking is grounded in the structure of the representations used by the lower level stuff that actually interacts with the world. Now this agrees with most of what you are saying, namely that we can't implement a cloud in the sky AGI that thinks in a vacuum. But it disagrees with you in saying that we can't afford to work on these sub-problems without the context of the entire organism. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: AGI Complexity (WAS: RE: [agi] doubling time watcher.)
The nature of neuroscience research doesn't really differentiate between the two at present. In order to understand WHAT a brain part does, we have to understand HOW it, and all structures connected to it function. We need to understand the inputs and the outputs, and that's all HOW. I wouldn't say even that much... The exact format of the IO is not necessary either but only the general Information X Y and Z is carried to here from here. We don't even know what the information is, honestly. Cells fire spikes. Sometimes there are clear behavioral correlates which makes it easy to figure out (place cells), usually not. The spike firing code depends on the function of the underlying structures. We have to know how they represent information to know what information is being transmitted. Understand, by the way, that there are plenty of computational and mathematical specialists working on this, applying plenty of information theoretic approaches. I've seen a very interesting report on the reverse engineering of the hearing system though I am still months away from finishing my first reading of Principles of neuroscience. The primary modalities are the easiest systems to decode, because you can control precisely what the inputs are. Those are the first systems to be decoded. Yes, that is because they don't constitute a computer. I suppose you need a really deep understanding of what computation is to see how the cortex is a computer (and hence has all the same properties of nonpredictability and such...) Well it computes. So... it's a computer, sure. Feel free to tell me more. Does it really? ;) I would suggest that the individual cortical columns represent a fairly consistient set of adaptive logic gates (of considerable complexity). I would further suggest that as the ferrit example showed the computation the cortical region performs depends mostly on where in the logic network the inputs are sent and the outputs taken. In this way you can take just about any cortical region and get it to do just about anything any other region does (except for the extra layers of the occipital lobe) just by hooking it up differently... I don't really have any strong data for or against that hypothesis. We're not sure how brittle columns are, functionally. Simple neural net models tell us though that it's very easy to drastically alter the functional character of a network by changing one parameter. I'll read the ferret example, but I'm guessing that all they found was evidence of striation, which doesn't mean the system is working correctly. However, given the resilience of the brain to changes performed at a young age, it is likely there was some visual perception. Where is the evidence for celular differentiation beyond the 20 or so classes of neurons? I'm not talking just about neuron types, but also about connection patterns of neurons between and within areas. Subregions CA3 and CA1 of the hippocampus are identical from a cellular composition perspective, but their connectivity patterns are so different that noone who studies the system would expect them to do the same thing. Neurophysiological evidence demonstrates that they do in fact differ in their functional characteristics. Absent this evidence, how can you say that a certain structure of cells X, Y, and Z which are arranged in layers 1-6 in cortical region A do something significantly different from those in region B? For starters, an autoassociative network performs differently than a heteroassociative one. Or add noradrenergic modulation(or one of 10+ other neuromodulators), or delete a subclass of GABA cells, or triple the percentage of stellate cells. It is easy to make a neural network behave differently. This is easily demonstrated with models. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] doubling time revisted.
Processing speed is a necessary but far from sufficient criterion of AGI design. The software engineering aspect is going to be the bigger limitation by far. It is common to speak of the brain as x neurons and Y synapses but the truth of it is that there are layers of complexity beneath the synapses. Even more important is the vast heterogeneity between brain regions. Even within cortex regions of similar architechture(and there are many different types of cortex!), the interconnections between regions alone effectively equate to specializied subsystems. If raw horsepower were the limiting factor, evolution could have easily given us massively homogeneous blobs of neural tissue at a cheap engineering/DNA cost. The fact that evolution uses a diverse and heterogeneous neural architecture tells us above all that it is *necessary*. Long term evolution doesn't do things it doesn't have to. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] doubling time revisted.
Hmmm. I think the critical problem is neither processing speed, NOR software engineering per se -- it's having a mind design that's correct in all the details. Or is that what you meant by software engineering? To me, software engineering is about HOW you build it, not about WHAT you build in a mathematical/conceptual sense. That is what I meant, yes. The WHAT that you build. The HOW isn't so much important except in that it's efficient enough to get the job done and doesn't leave the authors lost. I guess my point is that if you take as a thought experiment the idea that on waking up tomorrow and we found all of our cpu's magically running at 10x the speed (memory 10x, etc), we wouldn't be that much closer to an AGI because we're still working on what to do with the power. However, with your experience at webmind you would know better than I how cpu limits constrain AGI design. And cheaper computers let individuals in less wealthy nations get online and start computing, which adds more brainpower to the mix... -- Ben G An excellent, and rarely stated point. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] doubling time revisted.
It is obvious that no one on this list agrees with me. This does not mean = that I am obviously wrong. The division is very simple. My position: the doubling time has been reducing and will continue to do s= o. Their position: the doubling time is constant. It is incredibly unlikely that the doubling time is constant. But whatever the data show, as Ben says, it is impossible that the decrease in doubling time can continue ad infinitum. It will approach various asymptotic limits as defined by technology and market pressures (noone is going to spend billions in RD to make computers that cost $.05) and eventually by the laws of physics themselves as we approach the atomic and quantum scales. We will not have $10, 20 ghz, computers in 3 years. This is not a question of philosophy but only of the data. What does the d= ata show? If we had a stack of COMPUTER SHOPPER magazines for the past tw= enty years the question could be decided in short order. The drop in doubl= ing time starts out very slowly. That is why it is not obvious yet. By th= e time it becomes obvious it will be too late. Mike Deering. www.SingularityActionGroup.com---new website. --- To unsubscribe, change your address, or temporarily deactivate your subscri= ption,=20 please go to http://v2.listbox.com/member/?[EMAIL PROTECTED] --=_NextPart_000_0010_01C2D67C.E8727860 Content-Type: text/html; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable !DOCTYPE HTML PUBLIC -//W3C//DTD HTML 4.0 Transitional//EN HTMLHEAD META http-equiv=3DContent-Type content=3Dtext/html; charset=3Diso-8859-1 META content=3DMSHTML 6.00.2800.1126 name=3DGENERATOR STYLE/STYLE /HEAD BODY bgColor=3D#ff DIVFONT face=3DArialIt is obvious that no one on this list agrees with= =20 me.nbsp; This does not mean that I am obviously wrong.nbsp; The division = is=20 very simple./FONT/DIV DIVFONT face=3DArial/FONTnbsp;/DIV DIVFONT face=3DArialMy position:nbsp; the doubling time has been reduc= ing and=20 will continue to do so./FONT/DIV DIVFONT face=3DArial/FONTnbsp;/DIV DIVFONT face=3DArialTheir position:nbsp; the doubling time is=20 constant./FONT/DIV DIVFONT face=3DArial/FONTnbsp;/DIV DIVFONT face=3DArialThis is not a question of philosophy but only of th= e=20 data.nbsp; What does the data show?nbsp; If we had a stack of COMPUTER=20 SHOPPERnbsp; magazines for the past twenty years the question could be dec= ided=20 in short order.nbsp; The drop in doubling time starts out very slowly.nbs= p;=20 That is why it is not obvious yet.nbsp; By the time it becomes obvious it = will=20 be too late./FONT/DIV DIVFONT face=3DArial/FONTnbsp;/DIV DIVFONT face=3DArial/FONTnbsp;/DIV DIVFONT face=3DArialMike Deering./FONT/DIV DIVFONT face=3DArialA=20 href=3Dhttp://www.SingularityActionGroup.com;www.SingularityActionGroup.c= om/Anbsp;nbsp;nbsp;=20 lt;---new website./FONT/DIV/BODY/HTML --=_NextPart_000_0010_01C2D67C.E8727860-- --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] doubling time revisted.
I know this topic is already beaten to death in previous discussions, but I'll throw out one more point after reading that we may already have the equivalent power of some 3000 minds in raw CPU available worldwide. The aggregate neural mass of the world's population of insects and animals are probably at least an order of magnitude greater than that of humanity(and this using processing units literally identical to our own, no uncomfortable assumptions of computational equivalence are involved). And yet they aren't the ones building spaceships. Putting processing power to good, effective use is a *hard* problem. Also, integrating the power of multiple units is another hard problem. I don't recall the figure, but the vast majority of the brain is interconnective tissue. Networking hardware scales nonlinearly with the number of processing units. Even if you had sole dominion of those millions of desktop units and the perfect AGI software to run on them, the bandwidth bottleneck would make the thing unusable. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] Breaking AIXI-tl
I guess that for AIXI to learn this sort of thing, it would have to be rewarded for understanding AIXI in general, for proving theorems about AIXI, etc. Once it had learned this, it might be able to apply this knowledge in the one-shot PD context But I am not sure. For those of us who have missed a critical message or two in this weekend's lengthy exchange, can you explain briefly the one-shot complex PD? I'm unsure how a program could evaluate and learn to predict the behavior of its opponent if it only gets 1-shot. Obviously I'm missing something. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] Reply to Bill Hubbard's post: Mon, 10 Feb 2003
There are simple external conditions that provoke protective tendencies in humans following chains of logic that seem entirely natural to us. Our intuition that reproducing these simple external conditions serve to provoke protective tendencies in AIs is knowably wrong, failing an unsupported specific complex miracle. Well said. Or to put it another way, you see Friendliness in AIs as pretty likely regardless, and you think I'm going to all these lengths to provide a guarantee. I'm not. I'm going to all these lengths to create a *significant probability* of Friendliness. You're mischaracterizing my position. I'm certainly not saying we'll get friendliness for free, but was trying to reason by analogy (perhaps in a flawed way), that our best chance of success may be to model AGI's based on our innate tendencies wherever possible. Human behavior is a knowable quality. I perceived, based on the character of your discussion, that you would be unsatisfied with anything short of a formal, mathetmatical proof that any given AGI would not destroy us before giving the assent to turning it on. If that characterization was incorrect, the fault is mine. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] AI Morality -- a hopeless quest
I don't think any human alive has the moral and ethical underpinnings to allow them to resist the corruption of absolute power in the long run. We are all kept in check by our lack of power, the competition of our fellow humans, the laws of society, and the instructions of our peers. Remove a human from that support framework and you will have a human that will warp and shift over time. We are designed to exist in a social framework, and our fragile ethical code cannot function properly in a vacuum. This says two things to me. First, we should try to create friendly AI's. Second, we have no hope of doing it. We will forge ahead anyway because progress is always inevitable. We'll do as good a job as we can. At some point humans will be obsolete, but that's no reason to turn back. I'm also a strong proponent of the idea that humans can be made much better with the addition of enhancements, first through external add-ons (gargoyle type apparati which enhance our minds through UI's that are as intuitively useful as a hammer), and later through direct enhancement of our brains. In summary, I think we are getting ahead of ourselves in thinking we even have the capacity to predict what a friendly AI will be, especially if said AI is hyperintelligent and self-modifying. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] AI Morality -- a hopeless quest
I am exceedingly glad that I do not share your opinion on this. Human altruism *is* possible, and indeed I observe myself possessing a significant measure of it. Anyone doubting thier ability to 'resist corruption' should not IMO be working in AGI, but should be doing some serious introspection/study of thier goals and motivations. (No offence intended, Brad) Michael Roy Ames None taken. I'm altruistic myself, to a fault oftentimes. I have no doubt of my ability to help my fellow man. I bend over backwards to help complete strangers without a care because it makes me feel good. I am a friendly person. But that word fellow is the key. It implies peers, relative equals. I don't think I, or you, or anyone, can expect our personal ethical frameworks to function properly in a situation like that a hyperintelligent AI will face. Tell me this, have you ever killed an insect because it bothered you? -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] AI Morality -- a hopeless quest
I can't imagine the military would be interested in AGI, by its very definition. The military would want specialized AI's, constructed around a specific purpose and under their strict control. An AGI goes against everything the military wants from its weapons and agents. They train soldiers for a long time specifically to beat the GI out of them (har har, no pun intended) so that they behave in a predictable manner in a dangerous situation. And while I'm not entirely optimistic about the practicality of building ethics into AI's, I think we should certainly try, and that rules military funding right out. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] Consciousness
A good, if somewhat lightweight, article on the nature of mind and whether = silicon can eventually manifest conscioussness.. http://www.theage.com.au/articles/2003/02/09/1044725672185.html Kevin I don't know if consciousness debates are verbotten here or not, but I will say that I grow weary of Penrose worming his way into every debate/article with his hand-waving about quantum phenomenae. Their only application to the debate is that they are unknown and therefore a subject of mystery, like consciousness. The implied inference used by many, including this author, is that they are therefore related. He makes a good point about the failure of the neuron replacement thought experiment, but slipping and there is much in quantum physics to suggest it might be into the last paragraph left a bad taste in my mouth. Ascribing the unknown to quantum physics, merely because it is mysterious, is no different than ascribing it to the Almighty. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] AGI morality
There might even be a benefit to trying to develop an ethical system for the earliest possible AGIs - and that is that it forces everyone to strip the concept of an ethical system down to its absolute basics so that it can be made part of a not very intelligent system. That will probably be helpful in getting the clarity we need for any robust ethical system (provided we also think about the upgrade path issues and any evolutionary deadends we might need to avoid). Cheers, Philip I'm sure this idea is nothing new to this group, but I'll mention it anyway out of curiosity. A simple and implementable means of evaluating and training the ethics of an early AGI (one existing in a limited FileWorld type environment), would engage the AGI in variants of prisoner's dilemna with either humans or a copy of itself. The payoff matrix(CC, CD, DD) could be varied to provide a number of different ethical situtations. Another idea is that the prisoner's dilemna could then be internalized, and the AGI could play the game between internal actors, with the Self evaluating their actions and outcomes. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] A thought.
Philip, I can understand the brain structure we see in intelligent animals would emerge from a process of biological evolution where no conscious design is involved (ie. specialised non conscious functions emerge first, generalised processes emerge later), but why should AGI design emulate this given that we can now apply conscious design processes, in addition to the traditional evolutionary incremental trial and error methods? Cheers, Philip An excellent question. I don't think there's any long term need for AGI to follow evolution's path, and there are certainly some benefits to eschewing that approach. However, I don't think we're yet at a point in which we can afford to ignore the structure of the brain as a rubric. It seems to make the most sense that if we are going to develop an AGI that we can communicate with and understand, there's no reason to start from scratch. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] A thought.
3) Any successful AGI system is also going to have components in two other categories: a) specialized-intelligence components that solve particular problems in ways having little or nothing to do with truly general intelligence capability b) specialized-intelligence components that are explicitly built on top of components having truly general intelligence capability Are you willing to explain why you put them in this order, or has this available elsewhere, perhaps on agiri.org? I ask because it's my perspective that the brain is built the other way around, with specialized intelligence modules on the bottom and AGI built on top of them. I know you're not trying to build a brain per se, but I'm curious why you choose this manner to stack ASI and AGI. It's my belief that in the case of our brains, what we call AGI is the seamless combination of many ASI's. Our problem solving looks general, but it really isn't. There's AGI wiring on top to glue it all together, but most of the work is being done subconsciously in specialized regions. -Brad --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
Re: [agi] A thought.
I've just joined this list, it's my first post. Greetings all. 1.5 line summary of me: AI enthusiast since 10 yrs old, CS undergrad degree, 3 mo.'s from finishing psych/neuroscience PHD. Mike, you are correct that an AI must be matched to its enviornment. It's likely that a sentience optimized to function in an alien environment would behave in a way that appeared initially as random noise to a human observer. However when you say this: nd understandable. There is ONE general organizational structure that opti= mizes this AGI for our environment. All deviations from the one design onl= y serve to make the AGI function less effectively. Any significant departu= I could not disagree more. There are an infinite number of ways an AI could be designed within a given social/cultural context. The nature of the designs would allow them to provide different solutions, some of which are more or less effective in any particular situation. Evolution figured this out(forgive my anthropomorphizing), and this is why our minds contain many different forms of intelligence. They all attack any problem in a parallel fashion, share their results, and come to a sort of consensus. res cease to function in any way we would consider intelligent. The SAI of= the future will be vastly more intelligent, powerful, and amazing. It wil= l not be incomprehensible. It will be a lot like us. It might be incomprehensible if it's too much like us. One of the dangers of creating an AI to study brain function is that the result might be even more inscrutable than our brain. -Brad Wyble --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]