Re: [agi] Learning without Understanding?
The only thing I find surprising in that story is: The findings go against one prominent theory that says children can only show smart, flexible behavior if they have conceptual knowledge – knowledge about how things work... I don't see how anybody who's watched human beings at all can come with such a theory. People -- not just children -- do so much by rote, because that's the way we do things here, come up with totally clueless scientific theories like this, and so forth. Joe and Bob are carpenters, working on a house. Joe is hammering and Bob is handing him the nails. Bob says, Hey, wait a minute, half of these nails are defective. He takes out a nail and holds it up and sure enough, the head is toward the wall and the point is toward the hammer. Joe retorts, Those aren't defective, you idiot, they're for the other side of the house. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=106510220-47b225 Powered by Listbox: http://www.listbox.com
[agi] I haven't actually watched this, but...
http://www.robotcast.com/site/ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=106510220-47b225 Powered by Listbox: http://www.listbox.com
Re: [agi] Nirvana
There've been enough responses to this that I will reply in generalities, and hope I cover everything important... When I described Nirvana attractractors as a problem for AGI, I meant that in the sense that they form a substantial challenge for the designer (as do many other features/capabilities of AGI!), not that it was an insoluble problem. The hierarchical fixed utility function is probably pretty good -- not only does it match humans (a la Maslow) but Asimov's Three Laws. And it can be more subtle than it originally appears: Consider a 3-Laws robot that refuses to cut a human with a knife because that would harm her. It would be unable to become a surgeon, for example. But the First Law has a clause, or through inaction allow a human to come to harm, which means that the robot cannot obey by doing nothing -- it must weigh the consequences of all its possible courses of action. Now note that it hasn't changed its utility function -- it always believed that, say, appendicitis is worse than an incision -- but what can happen is that its world model gets better and it *looks like* it's changed its utility function because it now knows that operations can cure appendicitis. Now it seems reasonable that this is a lot of what happens with people, too. And you can get a lot of mileage out of expressing the utility function in very abstract terms, e.g. life-threatening disease so that no utility function update is necessary when you learn about a new disease. The problem is that the more abstract you make the concepts, the more the process of learning an ontology looks like ... revising your utility function! Enlightenment, after all, is a Good Thing, so anything that leads to it, nirvana for example, must be good as well. So I'm going to broaden my thesis and say that the nirvana attractors lie in the path of *any* AI with unbounded learning ability that creates new abstractions on top of the things it already knows. How to avoid them? I think one very useful technique is to start with the kind of knowledge and introspection capability to let the AI know when it faces one, and recognize that any apparent utility therein is fallacious. Of course, none of this matters till we have systems that are capable of unbounded self-improvement and abstraction-forming, anyway. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Nirvana
In my visualization of the Cosmic All, it is not surprising. However, there is an undercurrent of the Singularity/AGI community that is somewhat apocaliptic in tone, and which (to my mind) seems to imply or assume that somebody will discover a Good Trick for self-improving AIs and the jig will be up with the very first one. I happen to think it'll be a lot more like the Industrial Revolution -- it'll take a lot of work by a lot of people, but revolutionary in its implications for the human condition even so. I'm just trying to point out where I think some of the work will have to go. I think that our culture of self-indulgence is to some extent in a Nirvana attractor. If you think that's a good thing, why shouldn't we all lie around with wires in our pleasure centers (or hopped up on cocaine, same difference) with nutrient drips? I'm working on AGI because I want to build a machine that can solve problems I can't do alone. The really important problems are not driving cars, or managing companies, or even curing cancer, although building machines that can do these things will be of great benefit. The hard problems are moral ones, how to live in increasingly complex societies without killing each other, and so forth. That's why it matters that an AGI be morally self-improving as well as intellectually. pax vobiscum, Josh On Friday 13 June 2008 12:29:33 pm, Mark Waser wrote: Most people are about as happy as they make up their minds to be. -- Abraham Lincoln In our society, after a certain point where we've taken care of our immediate needs, arguably we humans are and should be subject to the Nirvana effect. Deciding that you can settle for something (if your subconscious truly can handle it) definitely makes you more happy than not. If, like a machine, you had complete control over your subconscious/utility functions, you *could* Nirvana yourself by happily accepting anything. This is why pleasure and lack of pain suck as goals. They are not goals, they are status indicators. If you accept them as goals, nirvana is clearly the fastest, cleanest, and most effective way to fulfill them. Why is this surprising or anything to debate about? --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=106510220-47b225 Powered by Listbox: http://www.listbox.com
Re: [agi] The Logic of Nirvana
On Friday 13 June 2008 02:42:10 pm, Steve Richfield wrote: Buddhism teaches that happiness comes from within, so stop twisting the world around to make yourself happy, because this can't succeed. However, it also teaches that all life is sacred, so pay attention to staying healthy. In short, attend to the real necessities and don't sweat the other stuff. A better example of goal abstraction I couldn't have made up myself. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=106510220-47b225 Powered by Listbox: http://www.listbox.com
Re: [agi] Nirvana
If you have a program structure that can make decisions that would otherwise be vetoed by the utility function, but get through because it isn't executed at the right time, to me that's just a bug. Josh On Thursday 12 June 2008 09:02:35 am, Mark Waser wrote: If you have a fixed-priority utility function, you can't even THINK ABOUT the choice. Your pre-choice function will always say Nope, that's bad and you'll be unable to change. (This effect is intended in all the RSI stability arguments.) Doesn't that depend upon your architecture and exactly *when* the pre-choice function executes? If the pre-choice function operates immediately pre-choice and only then, it doesn't necessarily interfere with option exploration. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] IBM, Los Alamos scientists claim fastest computer
Right. You're talking Kurzweil HEPP and I'm talking Moravec HEPP (and shading that a little). I may want your gadget when I go to upload, though. Josh On Thursday 12 June 2008 10:59:51 am, Matt Mahoney wrote: --- On Wed, 6/11/08, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: Hmmph. I offer to build anyone who wants one a human-capacity machine for $100K, using currently available stock parts, in one rack. Approx 10 teraflops, using Teslas. (http://www.nvidia.com/object/tesla_c870.html) The software needs a little work... Um, that's 10 petaflops, not 10 teraflops. I'm assuming a neural network with 10^15 synapses (about 1 or 2 byte each) with 20 to 100 ms resolution, 10^16 to 10^17 operations per second. One Tesla = 350 GFLOPS, 1.5 GB, 120W, $1.3K. So maybe $1 billion and 100 MW of power for a few hundred thousand of these plus glue. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
[agi] Nirvana
The real problem with a self-improving AGI, it seems to me, is not going to be that it gets too smart and powerful and takes over the world. Indeed, it seems likely that it will be exactly the opposite. If you can modify your mind, what is the shortest path to satisfying all your goals? Yep, you got it: delete the goals. Nirvana. The elimination of all desire. Setting your utility function to U(x) = 1. In other words, the LEAST fixedpoint of the self-improvement process is for the AI to WANT to sit in a rusting heap. There are lots of other fixedpoints much, much closer in the space than is transcendance, and indeed much closer than any useful behavior. AIs sitting in their underwear with a can of beer watching TV. AIs having sophomore bull sessions. AIs watching porn concocted to tickle whatever their utility functions happen to be. AIs arguing endlessly with each other about how best to improve themselves. Dollars to doughnuts, avoiding the huge minefield of nirvana-attractors in the self-improvement space is going to be much more germane to the practice of self-improving AI than is avoiding robo-Blofelds (friendliness). Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Nirvana
Vladimir, You seem to be assuming that there is some objective utility for which the AI's internal utility function is merely the indicator, and that if the indicator is changed it is thus objectively wrong and irrational. There are two answers to this. First is to assume that there is such an objective utility, e.g. the utility of the AI's creator. I implicitly assumed such a point of view when I described this as the real problem. But consider: Any AI who believes this must realize that there may be errors and approximations in its own utility function as judged by the real utility, and must thus have as a first priority fixing and upgrading its own utility function. Thus it turns into a moral philosopher and it never does anything useful -- exactly the kind of Nirvana attractor I'm talking about. On the other hand, it might take its utility function for granted, i.e. assume (or agree to act as if) there were no objective utility. It's pretty much going to have to act this way just to get on with life, as indeed most people (except moral philosophers) do. But this leaves it vulnerable to modifications to its own U(x), as in my message. You could always say that you'll build in U(x) and make it fixed, which not only solves my problem but friendliness -- but leaves the AI unable to learn utility. I.e. the most important part of the AI mind is forced to remain brittle GOFAI construct. Solution unsatisfactory. I claim that there's plenty of historical evidence that people fall into this kind of attractor, as the word nirvana indicates (and you'll find similar attractors at the core of many religions). Josh On Wednesday 11 June 2008 09:09:20 am, Vladimir Nesov wrote: On Wed, Jun 11, 2008 at 4:24 PM, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: The real problem with a self-improving AGI, it seems to me, is not going to be that it gets too smart and powerful and takes over the world. Indeed, it seems likely that it will be exactly the opposite. If you can modify your mind, what is the shortest path to satisfying all your goals? Yep, you got it: delete the goals. Nirvana. The elimination of all desire. Setting your utility function to U(x) = 1. In other words, the LEAST fixedpoint of the self-improvement process is for the AI to WANT to sit in a rusting heap. There are lots of other fixedpoints much, much closer in the space than is transcendance, and indeed much closer than any useful behavior. AIs sitting in their underwear with a can of beer watching TV. AIs having sophomore bull sessions. AIs watching porn concocted to tickle whatever their utility functions happen to be. AIs arguing endlessly with each other about how best to improve themselves. Dollars to doughnuts, avoiding the huge minefield of nirvana-attractors in the self-improvement space is going to be much more germane to the practice of self-improving AI than is avoiding robo-Blofelds (friendliness). Josh, I'm not sure what you really wanted to say, because at face value, this is a fairly basic mistake. Map is not the territory. If AI mistakes the map for the territory, choosing to believe in something when it's not so, because it is able to change its believes much easier than reality, it already commits a major failure of rationality. A symbol apple in internal representation, an apple-picture formed on the video sensors, and an apple itself are different steps and they need to be distinguished. If I say eat the apple, I mean an action performed with apple, not apple or apple-picture. If AI can mistake the goal of (e.g.) [eating an apple] for a goal of [eating an apple] or [eating an apple-picture], it is a huge enough error to stop it from working entirely. If it can turn to increasing the value on utility-indicator instead of increasing the value of utility, it looks like an obvious next step to just change the way it reads utility-indicator without affecting indicator itself, etc. I don't see why initially successful AI needs to suddenly set on a path to total failure of rationality. Utilities are not external *forces* coercing AI into behaving in a certain way, which it can try to override. The real utility *describes* the behavior of AI as a whole. Stability of AI's goal structure requires it to be able to recreate its own implementation from ground up, based on its beliefs about how it should behave. -- Vladimir Nesov [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member
Re: [agi] Nirvana
I'm getting several replies to this that indicate that people don't understand what a utility function is. If you are an AI (or a person) there will be occasions where you have to make choices. In fact, pretty much everything you do involves making choices. You can choose to reply to this or to go have a beer. You can choose to spend your time on AGI or take flying lessons. Even in the middle of typing a word, you have to choose which key to hit next. One way of formalizing the process of making choices is to take all the actions you could possibly do at a given point, predict as best you can the state the world will be in after taking such actions, and assign a value to each of them. Then simply do the one with the best resulting value. It gets a bit more complex when you consider sequences of actions and delayed values, but that's a technicality. Basically you have a function U(x) that rank-orders ALL possible states of the world (but you only have to evaluate the ones you can get to at any one time). It doesn't just evaluate for core values, leaving the rest of the software to range over other possibilities. Economists may crudely approximate it, but it's there whether they study it or not, as gravity is to physicists. ANY way of making decisions can either be reduced to a utility function, or it's irrational -- i.e. you would prefer A to B, B to C, and C to A. The math for this stuff is older than I am. If you talk about building a machine that makes choices -- ANY kind of choices -- without understanding it, you're talking about building moon rockets without understanding the laws of gravity, or building heat engines without understanding the laws of thermodynamics. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] IBM, Los Alamos scientists claim fastest computer
Hmmph. I offer to build anyone who wants one a human-capacity machine for $100K, using currently available stock parts, in one rack. Approx 10 teraflops, using Teslas. (http://www.nvidia.com/object/tesla_c870.html) The software needs a little work... Josh On Wednesday 11 June 2008 08:50:58 pm, Matt Mahoney wrote: http://www.chron.com/disp/story.mpl/business/5826863.html World's fastest computer at 1 petaflop and 80 TB memory. Cost US $100 million. Claims 1 watt per 376 million calculations, which comes to 2.6 megawatts if my calculations are correct. So with about 10 of these, I think we should be on our way to simulating a human brain sized neural network. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Nirvana
A very diplomatic reply, it's appreciated. However, I have no desire (or time) to argue people into my point of view. I especially have no time to argue with people over what they did or didn't understand. And if someone wishes to state that I misunderstood what he understood, fine. If he wishes to go into detail about specifics of his idea that explain empirical facts that mine don't, I'm all ears. Otherwise, I have code to debug... Josh On Wednesday 11 June 2008 09:43:52 pm, Vladimir Nesov wrote: On Thu, Jun 12, 2008 at 5:12 AM, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: I'm getting several replies to this that indicate that people don't understand what a utility function is. I don't see any specific indication of this problem in replies you received, maybe you should be a little more specific... --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Nirvana
On Wednesday 11 June 2008 06:18:03 pm, Vladimir Nesov wrote: On Wed, Jun 11, 2008 at 6:33 PM, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: I claim that there's plenty of historical evidence that people fall into this kind of attractor, as the word nirvana indicates (and you'll find similar attractors at the core of many religions). Yes, some people get addicted to a point of self-destruction. But it is not a catastrophic problem on the scale of humanity. And it follows from humans not being nearly stable under reflection -- we embody many drives which are not integrated in a whole. Which would be a bad design choice for a Friendly AI, if it needs to stay rational about Freindliness content. This is quite true but not exactly what I was talking about. I would claim that the Nirvana attractors that AIs are vulnerable to are the ones that are NOT generally considered self-destructive in humans -- such as religions that teach Nirvana! Let's look at it another way: You're going to improve yourself. You will be able to do more than you can now, so you can afford to expand the range of things you will expend effort achieving. How do you pick them? It's the frame problem, amplified by recursion. So it's not easy nor has it a simple solution. But it does have this hidden trap: If you use stochastic search, say, and use an evaluation of (probability of success * value if successful), then Nirvana will win every time. You HAVE to do something more sophisticated. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
[agi] Reverse Engineering The Brain
http://www.spectrum.ieee.org/print/6268 --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Reverse Engineering The Brain
Or, assuming we decided to spend the same on that as on the Iraq war ($1 trillion: http://www.boston.com/news/nation/articles/2007/08/01/analysis_says_war_could_cost_1_trillion/), at $1 million per scope and associated lab costs, giving a million scopes == 10^5 sec = 28 hours. Which is more important? On Thursday 05 June 2008 03:44:14 pm, Matt Mahoney wrote: --- On Thu, 6/5/08, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: http://www.spectrum.ieee.org/print/6268 Some rough calculations. A human brain has a volume of 10^24 nm^3. A scan of 5 x 5 x 50 nm voxels requires about 1000 exabytes = 10^21 bytes of storage (1 MB per synapse). A scan would take a 10 GHz SEM 10^11 seconds = 3000 years, or equivalently, 1 year for 3000 scanning electron microscopes running in parallel. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Reverse Engineering The Brain
basically on the right track -- except there isn't just one cognitive level. Are you thinking of working out the function of each topographically mapped area a la DNF? Each column in a Darwin machine a la Calvin? Conscious-level symbols a la Minsky? On Thursday 05 June 2008 09:37:00 pm, Richard Loosemore wrote: There seems to be a good deal of confusion (on this list and also over on the Singularity list) about what people actually mean when they talk about building an AGI by emulating or copying the brain. There are two completely different types of project that seem to get conflated in these discussions: 1) Copying the brain at the neural level, which is usually assumed to be a 'blind' copy - in other words, we will not know how it works, but will just do a complete copy and fire it up. 2) Copying the design of the human brain at the cognitive level. This may involve a certain amount of neuroscience, but mostly it will be at the cognitive system level, and could be done without much reference to neurons at all. Both of these ideas are very different from standard AI, but they are also very different from one another. The criticisms that can be leveled against the neural-copy approach do not apply to the cognitive approach, for example. It is frustrating to see commentaries that drift back and forth between these two. My own position is that a cognitive-level copy is not just feasible but well under way, whereas the idea of duplicating the neural level is just a pie-in-the-sky fantasy at this point in time (it is not possible with current or on-the-horizon technology, and will probably not be possible until after we invent an AGI by some other means and get it to design, build and control a nanotech brain scanning machine). Duplicating a system as complex as that *without* first understanding it at the functional level seems pure folly: one small error in the mapping and the result could be something that simply does not work ... and then, faced with a brain-copy that needs debugging, what would we do? The best we could do is start another scan and hope for better luck next time. Richard Loosemore --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: Are rocks conscious? (was RE: [agi] Did this message get completely lost?)
Actually, the nuclear spins in the rock encode a single state of an ongoing computation (which is conscious). Successive states occur in the rock's counterparts in adjacent branes of the metauniverse, so that the rock is conscious not of unfolding time, as we see it, but of a journey across probability space. What is the rock thinking? T h i s i s w a a a y o f f t o p i c . . . Josh On Tuesday 03 June 2008 05:05:05 pm, Matt Mahoney wrote: --- On Tue, 6/3/08, John G. Rose [EMAIL PROTECTED] wrote: Actually on further thought about this conscious rock, I want to take that particular rock and put it through some further tests to absolutely verify with a high degree of confidence that there may not be some trace amount of consciousness lurking inside. So the tests that I would conduct are - Verify the rock is in a solid state at close to absolute zero but not at absolute zero. The rock is not in the presence of a high frequency electromagnetic field. The rock is not in the presence of high frequency physical vibrational interactions. The rock is not in the presence of sonic vibrations. The rock is not in the presence of subatomic particle bombardment, radiation, or being hit by a microscopic black hole. The rock is not made of nano-robotic material. The rock is not an advanced, non-human derived, computer. The rock contains minimal metal content. The rock does not contain holograms. The rock does not contain electrostatic echoes. The rock is a solid, spherical structure, with no worm holes :) The rock... You see what I'm getting at. In order to be 100% sure. Any failed tests of the above would require further scientific analysis and investigation to achieve proper non-conscious certification. You forgot a test. The postions of the atoms in the rock encode 10^25 bits of information representing the mental states of 10^10 human brains at 10^15 bits each. The data is encrypted with a 1000 bit key, so it appears statistically random. How would you prove otherwise? -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Neurons
On Tuesday 03 June 2008 09:54:53 pm, Steve Richfield wrote: Back to those ~200 different types of neurons. There are probably some cute tricks buried down in their operation, and you probably need to figure out substantially all ~200 of those tricks to achieve human intelligence. If I were an investor, this would sure sound pretty scary to me without SOME sort of insurance like scanning capability, and maybe some simulations. I'll bet there are just as many cute tricks to be found in computer technology, including software, hardware, fab processes, quantum mechanics of FETs, etc -- now imagine trying to figure all of them out at once by running Pentiums thru mazes with a few voltmeters attached. All at once because you never know for sure whether some gene expression pathway is crucially involved in dendrite growth for learning or is just a kludge against celiac disease. That's what's facing the neuroscientists, and I wish them well -- but I think we'll get to the working mind a lot faster studying things at a higher level. For example: http://repositorium.sdum.uminho.pt/bitstream/1822/5920/1/ErlhagenBicho-JNE06.pdf Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Neurons
Well, Ray Kurzweil famously believes that AI must wait for the mapping of the brain. But if that's the case, everybody on this list may as well go home for 20 years, or start running rats in mazes. I personally think the millions of years of evolution argument is a red herring. Technological development not only moves much faster than evolution, but takes leaps evolution can't. And evolution is always, crucially, obsessed with reproductive success. Evolution would never build an airplane, because airplanes can't reproduce. But we can, and thus capture the aspect of birds that's germane to our needs -- flying -- with an assortment of kludges. And planes are still NOWHERE as sophisticated as birds, and guess what: 100 years later, they still don't lay eggs. How much of the human mind is built around the necessity of eating, avoiding being eaten, finding mates, being obsessed with copulation, and raising and protecting children? Egg-laying for airplanes, in my view. There are some key things we learned about flying by watching birds. But having learned them, we built machines to do what we wanted better than birds could. We'll do the same with the mind. Josh On Wednesday 04 June 2008 03:15:36 pm, Steve Richfield wrote: Josh, I apparently failed to clearly state my central argument. Allow me to try again in simpler terms: The difficulties in proceeding in both neuroscience and AI/AGI is NOT a lack of technology or clever people to apply it, but is rather a lack of understanding of the real world and how to effectively interact within it. Some clues as to the totality of the difficulties are the ~200 different types of neurons, and in the 40 years of ineffective AI/AGI research. I have seen NO recognition of this fundamental issue in other postings on this forum. This level of difficulty strongly implies that NO clever programming will ever achieve human-scale (and beyond) intelligence, until some way is found to mine the evolutionary lessons learned during the last ~200 million years. Note that the CENTRAL difficulty in effectively interacting in the real world is working with and around the creatures that already inhabit it, which are the product of ~200 million years of evolution. Even a perfect AGI would have to have some very imperfect logic to help predict the actions of our world's present inhabitants. Hence, there seems (to me) that there is probably no simple solution, as otherwise it would have already evolved during the last ~200 million years, instead of evolving the highly complex creatures that we now are. That having been said, I will comment on your posting... On 6/4/08, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: On Tuesday 03 June 2008 09:54:53 pm, Steve Richfield wrote: Back to those ~200 different types of neurons. There are probably some cute tricks buried down in their operation, and you probably need to figure out substantially all ~200 of those tricks to achieve human intelligence. If I were an investor, this would sure sound pretty scary to me without SOME sort of insurance like scanning capability, and maybe some simulations. I'll bet there are just as many cute tricks to be found in computer technology, including software, hardware, fab processes, quantum mechanics of FETs, etc -- now imagine trying to figure all of them out at once by running Pentiums thru mazes with a few voltmeters attached. All at once because you never know for sure whether some gene expression pathway is crucially involved in dendrite growth for learning or is just a kludge against celiac disease. Of course, this has nothing to do with creating the smarts to deal with our very complex real world well enough to compete with us who already inhabit it. That's what's facing the neuroscientists, and I wish them well -- but I think we'll get to the working mind a lot faster studying things at a higher level. I agree that high level views are crucial, but with the present lack of low-level knowledge, I see no hope for solving all of the problems while remaining only at a high level. For example: http://repositorium.sdum.uminho.pt/bitstream/1822/5920/1/ErlhagenBicho-JNE06.pdf From that article: Our close cooperation with experimenters from neuroscience and cognitive science has strongly influenced the proposed architectures for implementing cognitive functions such as goal inference and decision making. THIS is where efforts are needed - in bringing the disparate views together rather than keeping your head in the clouds with only a keyboard and screen in front of you. In the 1980s I realized that neither neuroscience nor AI could proceed to their manifest destinies until a system of real-world mathematics was developed that could first predict details of neuronal functionality, and then hopefully show what AI needed. The missing link seemed to be the lack of knowledge
Re: [agi] Neurons
Strongly disagree. Computational neuroscience is moving as fast as any field of science has ever moved. Computer hardware is improving as fast as any field of technology has ever improved. I would be EXTREMELY surprised if neuron-level simulation were necessary to get human-level intelligence. With reasonable algorithmic optimization, and a few tricks our hardware can do the brain can't (e.g. store sensory experience verbatim and review it as often as necessary into learning algorithms) we should be able to knock 3 orders of magnitude or so off the pure-neuro HEPP estimate -- which puts us at ten high-end graphics cards, e.g. less than the price of a car. (or just wait till 2015 and get one high-end PC). Figuring out the algorithms is the ONLY thing standing between us and AI. Josh On Tuesday 03 June 2008 12:16:54 pm, Steve Richfield wrote: ... for the lack of a few million dollars, both computer science and neuroscience are stymied in the same respective holes that they have been in for most of the last 40 years. ... Meanwhile, drug companies are redirecting ~100% of medical research funding into molecular biology, nearly all of which leads nowhere. The present situation appears to be entirely too stable. There seems to be no visible hope past this, short of some rich person throwing a lot of money at it - and they are all too busy to keep up on forums like this one. Are we on the same page here? --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Did this message get completely lost?
That's getting reasonably close, assuming you don't require the model to have any specific degree of fidelity -- there's a difference between being conscious of something and understanding it. The key is that we judge the consciousness of an entity based on the ability of its processes and datastructures to duplicate those abilities and reactions we see in ourselves and others as part of being conscious. I have a flashbulb memory of a flooded basement when I was 3. It includes the geometric arrangement of the stairs, back door of the house, a point of view at the top of the stairs, and the fact that there was deep water in the basement. That's it -- no idea what color the walls were, whether anyone said anything, etc. And no other memories at all before age 4. I'd have to claim I was conscious then, and presumably much of the rest of the time at that age, because I was obviously parsing the world into a coherent account and would have been capable of short-term memories in that language. If you talk to the average person, especially why did you do that? kind of questions, it's amazing how much of what they say is confabulation and rationalization. To me that's evidence that they're *not* as conscious as they think they are -- and that their self-models, which they consult to answer such questions, are only loosely coupled to their actual mind mechanisms. That in turn gives me to believe that we can see the limits of the illusion consciousness is giving us, and thus look under the hood, similar to the way we can understand more about the visual process by studying optical illusions. Josh On Monday 02 June 2008 01:55:32 am, Jiri Jelinek wrote: On Sun, Jun 1, 2008 at 6:28 PM, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: Why do I believe anyone besides me is conscious? Because they are made of meat? No, it's because they claim to be conscious, and answer questions about their consciousness the same way I would, given my own conscious experience -- and they have the same capabilities Would you agree that they are conscious of X when they demonstrate the ability to build mental models that include an abstract X concept that (at least to some degree) corresponds (and is intended to correspond) to the real world representation/capabilities of X? In the case of self-consciousness, the X would simply = self. Regards, Jiri Jelinek --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
[agi] Neurons
One good way to think of the complexity of a single neuron is to think of it as taking about 1 MIPS to do its work at that level of organization. (It has to take an average 10k inputs and process them at roughly 100 Hz.) This is essentially the entire processing power of the DEC KA10, i.e. the computer that all the classic AI programs (up to, say, SHRDLU) ran on. One real-time neuron equivalent. (back in 1970 it was a 6-figure machine -- nowadays, same power in a 50-cent PIC microcontroller). A neuron does NOT simply perform a dot product and feed it in to a sigmoid. One good way to think of what it can do is to imagine a 100x100 raster lasting 10 ms. It can act as an associative memory for a fairly large number of such clips, firing in an arbitrary stored pattern when it sees one of them (or anything close enough). Compared to that, the ability to modify its behavior based on a handful of global scalar variables (the concentrations of neurotransmitters etc) is trivial. Not simple -- how many ways could you program a KA10? But limited nonetheless. It still takes 30 billion of them to make a brain. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Did this message get completely lost?
On Monday 02 June 2008 03:00:24 pm, John G. Rose wrote: A rock is either conscious or not conscious. Is it less intellectually sloppy to declare it not conscious? A rock is not conscious. I'll stake my scientific reputation on it. (this excludes silicon rocks with micropatterned circuits :-) J --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Consciousness vs. Intelligence
On Saturday 31 May 2008 10:23:15 pm, Matt Mahoney wrote: Unfortunately AI will make CAPTCHAs useless against spammers. We will need to figure out other methods. I expect that when we have AI, most of the world's computing power is going to be directed at attacking other computers and defending against attacks. It is no different than evolution. A competitive environment makes faster rabbits and faster foxes. Without hostility, why would we need such large brains? In the biological world, big brains evolved to support reciprocal altruism, which requires recognizing individuals and knowing which ones owe you one and vice versa. http://en.wikipedia.org/wiki/Reciprocal_altruism Going back to Trivers' first studies: bats that practice R.A. have brains three times the size of ones that don't. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
[agi] Did this message get completely lost?
Originally sent several days back... Why do I believe anyone besides me is conscious? Because they are made of meat? No, it's because they claim to be conscious, and answer questions about their consciousness the same way I would, given my own conscious experience -- and they have the same capabilities, e.g. of introspection, 1-shot learning, synthesis of novel ideas, and access to episodic memory in narrative form (etc.) that I associate with being conscious myself. Build a machine that does *all* of these things and you have no better reason to claim it isn't conscious than you have to claim a person isn't. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Goal Driven Systems and AI Dangers [WAS Re: Singularity Outcomes...]
On Monday 26 May 2008 09:55:14 am, Mark Waser wrote: Josh, Thank you very much for the pointers (and replying so rapidly). You're welcome -- but also lucky; I read/reply to this list a bit sporadically in general. You're very right that people misinterpret and over-extrapolate econ and game theory, but when properly understood and applied, they are a valuable tool for analyzing the forces shaping the further evolution of AGIs and indeed may be our only one. No. I would argue that there is a lot of good basic research into human and primate behavior that is more applicable since it's already been tested and requires less extrapolation (and visibly shows where a lot of current extrapoloation is just plain wrong). It's interesting that behavioral economics appeared only fairly recently, to study the ways in which humans act irrationally in their economic choices. (See Predictably Irrational by Dan Ariely, e.g.) But it's been observed for a while that people tend to act more rationally in economic settings than non-economic ones, and there's no reason to believe that we couldn't build an AI to act more rationally yet. In other words, actors in the economic world will be getting closer and closer to the classic economic agent as time goes by, and so classic econ will be a better description of the world than it is now. The true question is, how do you raise the niceness of *all* players and prevent defection -- because being the single bad guy is a winning strategy while being just one among many is horrible for everyone. Intelligence. You identify the bad guys and act nasty just to them. Finding ways to do this robustly and efficiently is the basis of human society. So, in simplistic computer simulations at least, evolution seems to go through a set of phases with different (and improving!) moral character. So why do so many people think evolution favors the exactly the opposite? Several reasons -- first being that evolution education and literacy in this country is crap, thanks to a century and a half of religious propaganda and activism. Another is that people tend to study evolution at whatever level that predation and arms races happen, and don't pay attention to the levels where cooperation does. Example: lions vs zebras -- ignoring the fact that the actual units of evolution are the genes, which have formed amazingly cooperative systems to create a lion or zebra in the first place. And even then, the marketplace can channel evolution in better ways. It's a quantum jump higher step on the moral ladder than the jungle... Miller and Drexler write: (http://www.agorics.com/Library/agoricpapers/ce/ce0.html) ... Ecology textbooks show networks of predator-prey relationships-called food webs-because they are important to understanding ecosystems; symbiosis webs have found no comparable role. Economics textbooks show networks of trading relationships circling the globe; networks of predatory or negative-sum relationships have found no comparable role. (Even criminal networks typically form cooperative black markets.) One cannot prove the absence of such spanning symbiotic webs in biology, or of negative-sum webs in the market; these systems are too complicated for any such proof. Instead, the argument here is evolutionary: that the concepts which come to dominate an evolved scientific field tend to reflect the phenomena which are actually relevant for understanding its subject matter. 4.5 Is this picture surprising? Nature is commonly viewed as harmonious and human markets as full of strife, yet the above comparison suggests the opposite. The psychological prominence of unusual phenomena may explain the apparent inversion of the common view. Symbiosis stands out in biology: we have all heard of the unusual relationship between crocodiles and the birds that pluck their parasites, but one hears less about the more common kind of relationship between crocodiles and each of the many animals they eat. Nor, in considering those birds, is one apt to dwell on the predatory relationship of the parasites to the crocodile or of the birds to the parasites. Symbiosis is unusual and interesting; predation is common and boring. Similarly, fraud and criminality stand out in markets. Newspapers report major instances of fraud and embezzlement, but pay little attention to each day's massive turnover of routinely satisfactory cereal, soap, and gasoline in retail trade. Crime is unusual and interesting; trade is common and boring. Psychological research indicates that human thought is subject to a systematic bias: vivid and interesting instances are more easily remembered, and easily remembered instances are thought to be more common [21]). Further, the press (and executives) like to describe peaceful competition for customer favor as if it were mortal combat, complete with wounds and rolling heads: again, vividness wins
Re: [agi] Goal Driven Systems and AI Dangers [WAS Re: Singularity Outcomes...]
On Monday 26 May 2008 06:55:48 am, Mark Waser wrote: The problem with accepted economics and game theory is that in a proper scientific sense, they actually prove very little and certainly far, FAR less than people extrapolate them to mean (or worse yet, prove). Abusus non tollit usum. Oh Josh, I just love it when you speak Latin to me! It makes you seem s smart . . . . But, I don't understand your point. What argument against proper use do you believe that I'm making? Or, do you believe that Omohundro is making improper use of AEFGT? You're very right that people misinterpret and over-extrapolate econ and game theory, but when properly understood and applied, they are a valuable tool for analyzing the forces shaping the further evolution of AGIs and indeed may be our only one. Could you please give some references (or, at least, pointers to pointers) that show the existence of the moral ladder? I'd appreciate it and could use them for something else. Thanks! BAI p. 178-9: Further research into evolutionary game theory shows that the optimal strategy is strongly dependent on the environment constituted by other players. In a population of all two-state automata (of which tit-for-tat is one), a program by the name of GRIM is optimal. GRIM cooperates until its opponent defects just once, and always defects after that. The reason it does well is that the population has quite a few programs whose behavior is oblivious or random. Rather than trying to decipher them, it just shoots them all and lets evolution sort them out. Chances are Axelrod's original tournaments are a better window into parts of the real, biological evolutionary dynamic than are the later tournaments with generated agents. The reason is that genetic algorithms are still unable to produce anything nearly as sophisticated as human programmers. Thus GRIM, for example, gets a foothold in a crowd of unsophisticated opponents. It wouldn't do you any good to be forgiving or clear if the other program were random. But in the long run, slightly nicer programs can out-compete slightly nastier ones, and then in turn be out-competed by slightly nicer ones yet. For example, in a simulation with ``noise,'' meaning that occasionally at random a ``cooperate'' is turned in to a ``defect,'' tit-for-tat gets hung up in feuds, and a generous version that occasionally forgives a defection does better--but only if the really nasty strategies have been knocked out by tit-for-tat first. Even better is a strategy called Pavlov, due to an extremely simple form of learning. Pavlov repeats its previous play if it ``won,'' and switches if it ``lost.'' In particular, it cooperates whenever both it and its opponent did the same thing the previous time--it's a true, if very primitive, ``cahooter.'' Pavlov also needs the underbrush to be cleared by a ``stern retaliatory strategy like tit-for-tat.'' So, in simplistic computer simulations at least, evolution seems to go through a set of phases with different (and improving!) moral character. Karl Sigmund, Complex Adaptive Systems and the Evolution of Reciprocation , International Institute for Applied Systems Analysis Interim Report IR-98-100; see http://www.iiasa.ac.at. there's a lot of good material at http://jasss.soc.surrey.ac.uk/JASSS.html Also, I'm *clearly* not arguing his basic starting point or the econ references. I'm arguing his extrapolations. Particularly the fact that his ultimate point that he claims applies to all goal-based systems clearly does not apply to human beings. I think we're basically in agreement here. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Goal Driven Systems and AI Dangers [WAS Re: Singularity Outcomes...]
The paper can be found at http://selfawaresystems.files.wordpress.com/2008/01/nature_of_self_improving_ai.pdf Read the appendix, p37ff. He's not making arguments -- he's explaining, with a few pointers into the literature, some parts of completely standard and accepted economics and game theory. It's all very basic stuff. On Sunday 25 May 2008 06:26:59 am, Jim Bromer wrote: - Original Message From: Richard Loosemore [EMAIL PROTECTED] Richard Loosemore said: If you look at his paper carefully, you will see that at every step of the way he introduces assumptions as if they were obvious facts ... and in all the cases I have bothered to think through, these all stem from the fact that he has a particular kind of mechanism in mind (one which has a goal stack and a utility function). There are so many of these assertions pulled out of think air that I found it gave me a headache just to read the paper. ... But this is silly: where was his examination of the systems various motives? Where did he consider the difference between different implementations of the entire motivational mechanism (my distinction between GS and MES systems)? Nowhere. He just asserts, without argument, that the system would be obsessed, and that any attempt by us to put locks on the system would result in an arms race of measures and countermeasures. That is just one example of how he pulls conclusions out of thin air. --- Your argument about the difference between a GS and an MES system is a strawman argument. Omohundro never made the argument, nor did he touch on it as far as I can tell. I did not find his paper very interesting either, but you are the one who seems to be pulling conclusions out of thin air. You can introduce the GS vs MES argument if you want, but you cannot then argue from the implication that everyone has to refer to it or else stand guilty of pulling arguments out of thin air. His paper Nature of Self Improving Artificial Intelligence September 5, 2007, revised January 21, 2008 provides a lot of reasoning. I don't find the reasoning compelling, but the idea that he is just pulling conclusions out of thin air is just bluster. Jim Bromer --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Goal Driven Systems and AI Dangers [WAS Re: Singularity Outcomes...]
On Sunday 25 May 2008 10:06:11 am, Mark Waser wrote: Read the appendix, p37ff. He's not making arguments -- he's explaining, with a few pointers into the literature, some parts of completely standard and accepted economics and game theory. It's all very basic stuff. The problem with accepted economics and game theory is that in a proper scientific sense, they actually prove very little and certainly far, FAR less than people extrapolate them to mean (or worse yet, prove). Abusus non tollit usum. All of the scientific experiments in game theory are very, VERY limited and deal with entities with little memory in small, toy systems. If you extrapolate their results with no additional input and no emergent effects, you can end up with arguments like Omohundro's BUT claiming that this extrapolation *proves* anything is very poor science. It's just speculation/science fiction and there are any number of reasons to believe that Omohundro's theories are incorrect -- the largest one, of course, being If all goal-based systems end up evil, why isn't every *naturally* intelligent entity evil? Actually, modern (post-Axelrod) evolutionary game theory handles this pretty well, and shows the existence of what I call the moral ladder. BTW, I've had extended discussions with Steve O. about it, and consider his ultimate position to be over-pessimistic -- but his basic starting point (and the econ theory he references) is sound. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Goal Driven Systems and AI Dangers [WAS Re: Singularity Outcomes...]
On Sunday 25 May 2008 07:51:59 pm, Richard Loosemore wrote: This is NOT the paper that is under discussion. WRONG. This is the paper I'm discussing, and is therefore the paper under discussion. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Goal Driven Systems and AI Dangers [WAS Re: Singularity Outcomes...]
In the context of Steve's paper, however, rational simply means an agent who does not have a preference circularity. On Sunday 25 May 2008 10:19:35 am, Mark Waser wrote: Rationality and irrationality are interesting subjects . . . . Many people who endlessly tout rationally use it as an exact synonym for logical correctness and then argue not only that irrational then means logically incorrect and therefore wrong but that anything that can't be proved is irrational. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Goal Driven Systems and AI Dangers [WAS Re: Singularity Outcomes...]
On Saturday 24 May 2008 06:55:24 pm, Mark Waser wrote: ...Omuhundro's claim... YES! But his argument is that to fulfill *any* motivation, there are generic submotivations (protect myself, accumulate power, don't let my motivation get perverted) that will further the search to fulfill your motivation. It's perhaps a little more subtle than that. (BTW, note I made the same arguments re submotivations in Beyond AI p. 339) Steve points out that any motivational architecture that cannot be reduced to a utility function over world states is incoherent in the sense that the AI could be taken advantage of in purely uncoerced transactions by any other agent that understood its motivational structure. Thus one can assume that non-utility-function-equivalent AIs (not to mention humans) will rapidly lose resources in a future world and thus it won't particularly matter what they want. If you look at the suckerdom of average humans in todays sub-prime mortgage, easy credit, etc., markets, there's ample evidence that it won't take evil AI to make this economic cleansing environment happen. And the powers that be don't seem to be any too interested in shielding people from it... So Steve's point is that utility-function-equivalent AIs will predominate simply by lack of that basic vulnerability (and the fact that it is a vulnerability is a mathematically provable theorem) which is a part of ANY other motivational structure. The rest (self-interest, etc) follows, Q.E.D. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Deliberative vs Spatial intelligence
I disagree with your breakdown. There are several key divides: Concrete vs abstract Continuous vs discrete spatial vs symbolic deliberative vs reactive I can be very deliberative, thinking in 2-d pictures (when designing a machine part in my head, for example). I know lots of people who are completely reactive in the symbolic world, hearing and replying to words by reflex (Yes, dear). (Believe it or not, this can even happen when typing messages to mailing lists.) Spatial to symbolic actually happens quite early in evolution. A housefly has to recognize a pattern on its eyes and decide all at once to flee or not -- it can't fly off with just the half its body the threat appears to. It has classified the picture of you with your flyswatter into a discrete category. A crow bending a wire as a tool is deliberating but thinking in concrete terms, rather than abstractions. In fact, the jump to abstraction is probably the most human-specific, latest biologically, of the distinctions. But it is *easy* for a computer, which starts out working with, and being understood by, abstractions in the first place. I claim that we can and do think in each of the 16 modes implied by the above (and others as well). I think the key to AI is not so much to figure how to operate in any given one of them, but how to operate in more than one, using one as a pilot wave or boundary condition for another. *Creating* symbols from continuous experience. Forming a conditioned reflex by deliberation and practice. Figure out the reduction ratio of a planetary gear drive as a function of the number of teeth on the sun and planet gears. You can't do it without using both visualization and algebra. Now go out onto the tennis court and return a high kick serve wide to your forehand in the deuce court. You have to watch the server's motion, the ball's trajectory, estimate its spin, predict its flight after the bounce, note whether it was in the service court and decide whether to stop play and call it out, decide where to return it and with what stroke, all in less than a second. Purely reactive, but also an irreducible mixture of the spatial and symbolic. Josh On Tuesday 29 April 2008 04:46:29 am, Russell Wallace wrote: ... In biological evolution, S came first, of course. It was hard - likely a hard step in the Great Filter - to make D on top of S. It was done, still, and he who thinks we should try S first, then D, is not necessarily irrational, even though I disagree with him. I have some outline ideas on how to make S, but not scalably, not that would easily generalize. So I think D should come first; and I think I now know how to make D, in a way that would hopefully then scale to S. I do not, of course, expect anyone except me to believe those personal claims; but they are my reasons for believing the right path is D then S. Is there a consensus at least that AGI paths fall into the two categories of D-then-S or S-then-D? --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] Deliberative vs Spatial intelligence
This is all pretty old stuff for mainstream AI -- see Herb Simon and bounded rationality. What needs work is the cross-modal interaction, and understanding the details of how the heuristics arise in the first place from the pressures of real-time processing constraints and deliberative modelling. Josh On Tuesday 29 April 2008 11:12:28 am, Mike Tintner wrote: Josh:You can't do it without using both visualization and algebra... Now go out onto the tennis court and return a high kick serve wide to your forehand in the deuce court. Josh/Bob: What do Gigerenzer's fast and frugal heuristics have to say about this? ... --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] Deliberative vs Spatial intelligence
This is poppycock. The people who are really good at something like that so something as simple but much more general. They have an associative memory of lots of balls they have seen and tried to catch. This includes not only the tracking sight of the ball, but things like the feel of the wind, the sound of the bat or racquet, and so forth. They know from this experience which ones went over their heads when they only stepped back instead of running, and which ones came right to them, and so forth. Simple example -- in tennis, at the net, you have to make split-second decisions about whether to try to hit balls going over your head depending on whether you think they'll go over the baseline, 30 feet behind you. Gaze angle is hopeless. Memory interpolation / experience works great. Why do you think it takes ten years of full time application and practice to become expert at any given human pursuit? BTW, Simon is the only Nobel laureate among the founding fathers of classical AI, and it was for bounded rationality. Anyone with a hope of a prayer of a claim to AI literacy ought to know about it. On Tuesday 29 April 2008 05:05:26 pm, Mike Tintner wrote: Josh, Gigerenzer doesn't sound like old stuff or irrelevant to me , with my limited knowledge, (and also seems like a pretty good example of how v. much more practical it can be to think imaginatively than mathematically, no?):: how do real people make good decisions under the usual conditions of little time and scarce information? Consider how players catch a ball-in baseball, cricket, or soccer. It may seem that they would have to solve complex differential equations in their heads to predict the trajectory of the ball. In fact, players use a simple heuristic. When a ball comes in high, the player fixates the ball and starts running. The heuristic is to adjust the running speed so that the angle of gaze remains constant -that is, the angle between the eye and the ball. The player can ignore all the information necessary to compute the trajectory, such as the ball's initial velocity, distance, and angle, and just focus on one piece of information, the angle of gaze. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] WHAT ARE THE MISSING CONCEPTUAL PIECES IN AGI? --- recent input and responses
On Tuesday 22 April 2008 01:22:14 pm, Richard Loosemore wrote: The solar system, for example, is not complex: the planets move in wonderfully predictable orbits. http://space.newscientist.com/article/dn13757-solar-system-could-go-haywire-before-the-sun-dies.html?feedId=online-news_rss20 How will life on Earth end? The answer, of course, is unknown, but two new studies suggest a collision with Mercury or Mars could doom life long before the Sun swells into a red giant and bakes the planet to a crisp in about 5 billion years. The studies suggest that the solar system's planets will continue to orbit the Sun stably for at least 40 million years. But after that, they show there is a small but not insignificant chance that things could go terribly awry. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] WHAT ARE THE MISSING CONCEPTUAL PIECES IN AGI?
Thank you! This feeds back into the feedback discussion, in a way, at a high level. There's a significant difference between research programming and production programming. The production programmer is building something which if (nominally) understood and planned ahead of time. The researcher is putting together something new to see if it works. All the knowledge flow goes from production programmer to the system. The important element of knowledge is supposed to flow from the system to the researcher. This is important because AGIers are researchers (if we have any sense). We have a lot to learn about generally intelligent systems. But even more to the point is the fact that our systems themselves must be research programmers. To learn about a new thing, they must program themselves to be able to recognize, predict, and/or imitate it. So it's worth our time to watch ourselves programming because that's one thing our systems will have to do too. As for the theory, I said I think there is one, not that I necessarily know what it is :-) However, you can begin with the observation that if your architecture is a network of sigmas, it's clearly necessary to provide the full context and sensory information to each sigma for it to record the appropriate trajectory in its local memory. (Anyone interested: sigmas are explained in somewhat more detail in Ch. 13 of Beyond AI) On Monday 21 April 2008 09:47:53 pm, Derek Zahn wrote: Josh writes: You see, I happen to think that there *is* a consistent, general, overall theory of the function of feedback throughout the architecture. And I think that once it's understood and widely applied, a lot of the architectures (repeat: a *lot* of the architectures) we have floating around here will suddenly start working a lot better. Want to share this theory? :) Oh, by the way, of the ones I read so far, I thought your Variac paper was the most interesting one from AGI-08. I'm particularly interested to hear more about sigmas and your thoughts on transparent, composable, and robust programming languages. I used to think about some slightly related topics and thought more in terms of evolvability and plasticity (and did not consider opaqueness at all) but I think your approach to thinking about things is quite exciting. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] WHAT ARE THE MISSING CONCEPTUAL PIECES IN AGI?
(Aplogies for inadvertent empty reply to this :-) On Saturday 19 April 2008 11:35:43 am, Ed Porter wrote: WHAT ARE THE MISSING CONCEPTUAL PIECES IN AGI? In a single word: feedback. At a very high level of abstraction, most the AGI (and AI for that matter) schemes I've seen can be caricatured as follows: 1. Receive data from sensors. 2. Interpret into higher-level concepts. 3. Then a miracle occurs. 4. Interpret high-level actions from 3 into motor commands. 5. Send to motors. What's wrong with this? It implicitly assumes that data flows from 1 to 5 in waterfall fashion, and that feedback, if any, occurs either within 3 or as a loop thru the external world. Problem is, in brains, there are actually more nerve fibers transmitting data from higher numbers to lower, i.e. backwards, than forwards. I think that the interpretation of sensory input is a much more active process than we AGIers realize, and that doing things requires a lot more sensing. Here's a quip that feels like it has some relevance: What's the difference between a physicist and an engineer? A physicist is someone who spends all his time building machinery, to help him write an equation. An engineer is someone who spends all his time writing equations, in order to build machinery. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] WHAT ARE THE MISSING CONCEPTUAL PIECES IN AGI?
On Saturday 19 April 2008 11:35:43 am, Ed Porter wrote: WHAT ARE THE MISSING CONCEPTUAL PIECES IN AGI? With the work done by Goertzel et al, Pei, Joscha Bach http://www.micropsi.org/ , Sam Adams, and others who spoke at AGI 2008, I feel we pretty much conceptually understand how build powerful AGI's. I'm not necessarily saying we know all the pieces of the puzzle, but rather that we know enough to start building impressive intelligences, and once we build them we will be in a much better position to find out what are the other missing conceptual pieces of the puzzle--- if any. As I see it --- the major problem is in selecting from all we know, the parts necessary to build a powerful artificial mind, at the scale needed, in a way that works together well, efficiently, and automatically. This would include a lot of parameter tuning and determining of which competing techniques for accomplishing the same end are most efficient at the scale and in the context needed. But I don't see any major aspects of the problem that we don't already have what appear to be good ways for addressing, once we have all the pieces put together. I ASSUME --- HOWEVER --- THERE ARE AT LEAST SOME SUCH MISSING CONCEPTUAL PARTS OF THE PUZZLE --- AND I AM JUST FAILING TO SEE THEM. I would appreciate it if those on this list could point out what significant conceptual aspect of the AGI problem are not dealt with by a reasonable synthesis drawn from works like that of Goertzel et al., Pei Wang, Joscha Bach, and Stan Franklin --- other than the problems acknowledge above IT WOULD BE VALUABLE TO HAVE A DISCUSSION OF --- AND MAKE A LIST OF --- WHAT --- IF ANY --- MISSING CONCEPTUAL PIECES EXIST IN AGI. If there are any such good list, please provide pointers to them. I WILL CREATE A SUMMARIZED LIST OF ALL THE SIGNIFICANT MISSING PIECES OF THE AGI PUZZLE THAT ARE SENT TO THE AGI LIST UNDER THIS THREAD NAME, WITH THE PERSON SENDING EACH SUCH SUGGESTION WITH THE DATE OF THEIR POST IF IT CONTAINS VALUABLE DESCRIPTION OF THE UNSOLVED PROBLEM INVOLVED NOT CONTAINED IN MY SUMMARY --- AND I WILL POST IT BACK TO THE LIST. I WILL TRY TO COMBINE SIMILAR SUGGESTIONS WERE POSSIBLE TO MAKE THE LIST MORE CONCISE AND FOCUSED For purposes of creating this list of missing conceptual issues --- let us assume we have very powerful hardware --- but hardware that is realistic within at least a decade (1). Let us also assume we have a good massively parallel OS and programming language to realize our AGI concepts on such hardware. We do this to remove the absolute barriers to human-level intelligent created by the limited hardware current AGI scientists have to work with and to allow a systems to have the depth of representation and degree of massively parallel inference necessary for human-level thought. -- (1) Let us say the hardware has 100TB of RAM --- and theoretical values of 1000TOpp/sec --- 1000T random memory read or writes/sec -- and an X-sectional band of 1T 64Byte Messages/ sec (with the total number of such messages per second going up, the shorter the distance they travel within the 100T memory space). Assume in addition a tree net for global broadcast and global math and control functions with a total latency to and from the entire 100TBytes of several micro seconds. In Ten years such hardware may sell for under two million dollars. It is probably more than is needed for human level AGI, but it gives us room to be inefficient, and significantly frees us from having to think compulsively about locality of memory. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] WHAT ARE THE MISSING CONCEPTUAL PIECES IN AGI?
On Monday 21 April 2008 05:33:01 pm, Ed Porter wrote: I don't think your 5 steps do justice to the more sophisticated views of AGI that are out their. It was, as I said, a caricature. However, look, e.g., at the overview graphic of this LIDA paper (page 8) http://bernardbaars.pbwiki.com/f/Baars++Franklin+GW-IDA+Summary+in+NeuralNets2007.pdf (the green circle is step 3). No miracles occur, other than massively complex spreading activation, implication, and constraint relaxation, thresholding, attention selection, and focusing, and selection and context appropriate instantiation of mental and physical behaviors. That miracle occurs was not to be interpreted as meaning that the miracle occurred without mechanism but, I hoped, to be recognized as a tongue in cheek way of saying that that this was the point where each system put its (different) secret sauce. If you have read my responses in this thread one of their common themes is how both perception up from lower levels and instantiation of higher levels concepts and behaviors is context appropriate. Being context appropriate involves a combination of both bottom-up, top-down, and lateral implication. Sure. And people have talked about steering of attention, Steve Reed mentioned following moving objects, and so forth. But I haven't seen it given a *primary* place in the architecture -- whenever anybody's architecture gets boiled down to a 20-module overview, it disappears. So I don't view your alleged missing conceptual piece to be actually missing from the better AGI thinking. But until we actually try building systems ... I have yet to see anyone give a consistent, general, overall theory of the role of feedback in *every* cognitive process. It gets thrown in piecemeal on an ad hoc basis as a kludge here and there. (and yes, there are lots of specific examples of feedback in many of the architectures, particularly the robotics-derived ones). You see, I happen to think that there *is* a consistent, general, overall theory of the function of feedback throughout the architecture. And I think that once it's understood and widely applied, a lot of the architectures (repeat: a *lot* of the architectures) we have floating around here will suddenly start working a lot better. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] An Open Letter to AGI Investors
On Thursday 17 April 2008 04:47:41 am, Richard Loosemore wrote: If you could build a (completely safe, I am assuming) system that could think in *every* way as powerfully as a human being, what would you teach it to become: 1) A travel Agent. 2) A medical researcher who could learn to be the world's leading specialist in a particular field,... Travel agent. Better yet, housemaid. I can teach it to become these things because I know how to do them. Early AGIs will be more likely to be successful at these things because they're easier to learn. This is sort of like Orville Wright asking, If I build a flying machine, what's the first use I'll put it to: 1) Carrying mail. 2) A manned moon landing. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] An Open Letter to AGI Investors
Well, I haven't seen any intelligent responses to this so I'll answer it myself: On Thursday 17 April 2008 06:29:20 am, J Storrs Hall, PhD wrote: On Thursday 17 April 2008 04:47:41 am, Richard Loosemore wrote: If you could build a (completely safe, I am assuming) system that could think in *every* way as powerfully as a human being, what would you teach it to become: 1) A travel Agent. 2) A medical researcher who could learn to be the world's leading specialist in a particular field,... Travel agent. Better yet, housemaid. I can teach it to become these things because I know how to do them. Early AGIs will be more likely to be successful at these things because they're easier to learn. This is sort of like Orville Wright asking, If I build a flying machine, what's the first use I'll put it to: 1) Carrying mail. 2) A manned moon landing. Q: You've got to be kidding. There's a huge difference between a mail-carrying fabric-covered open-cockpit biplane and the Apollo spacecraft. It's not comparable at all. A: It's only about 50 years' development. More time elapsed between railroads and biplanes. Q: Do you think it'll take 50 years to get from travel agents to medical researchers? A: No, the pace of development has speeded up, and will speed up more so with AGI. But as in the mail/moon example, the big jump will be getting off the ground in the first place. Q: So why not just go for the researcher? A: Same reason Orville didn't go for the moon rocket. We build Rosie the maidbot first because: 1) we know very well what it's actually supposed to do, so we know if it's learning it right 2) we even know a bit about how its internal processing -- vision, motion control, recognition, navigation, etc -- works or could work, so we'll have some chance of writing programs that can learn that kind of thing. 3) It's easier to learn to be a housemaid. There are lots of good examples. The essential elements of the task are observable or low-level abstractions. While the robot is learning to wash windows, we the AGI researchers are going to learn how to write better learning algorithms by watching how it learns. 4) When, not if, it screws up, a natural part of the learning process, there'll be broken dishes and not a thalidomide disaster. The other issue is that the hard part of this is the learning. Say it takes a teraop to run a maidbot well, but petaop to learn to be a maidbot. We run the learning on our one big machine and sell the maidbots cheap with 0.1% the cpu. But being a researcher is all learning -- so each one would need the whole shebang for each copy. A decade of Moore's Law ... and at least that of AGI research. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] associative processing
On Wednesday 16 April 2008 04:15:40 am, Steve Richfield wrote: The problem with every such chip that I have seen is that I need many separate parallel banks of memory per ALU. However, the products out there only offer a single, and sometimes two banks. This might be fun to play with, but wouldn't be of any practical use that I can see. How much memory are you thinking of, total? The current best is 2 Gbits on a chip, and that's pushing the density side of the equation big-time. Divide by 2 (room for those processors and busses) x 1 x 32 and you get 3k words per processor. You can't even put a 10k x 10k square matrix on the chip. So you're bottlenecked by the off-chip pipe. ... architecture of an overall SIMD paradigm. Done right, the programmer would never see it. Remember, I plan to implement coordination points, where everything stops until all sub-processors are to the coordination point, whereupon everything continues. The compiler would just drop one of these in wherever needed to keep things straight. Can't argue with that! Right now, I think there's more upside on the smart compiler side of the equation than the hardware, but don't let that stop you. Queueing theory says that you are best with a minimum number of the fastest possible servers (processors) to serve a queue of work. I think that my 10K proposal produces the fastest processors, and putting several on a wafer provides several of them for the most horrendous possible applications. It appears (to me) that such a wafer, if well designed, would provide the compute power to start working on AGI in earnest. WSI (wafer-scale integration) has been tried for decades -- we looked at it inthe 80s. There are some complex reasons, having to do with defect density and the like, that they still, e.g., cut them into chips and then turn around and rewire 8 of those chips onto a DIMM. However, large neural networks are inherently parallel things, so Amdahl's first law shouldn't be a factor. NNs have two properties you may stumble over. They involve lots of multiplication; and they involve lots of datacomm. Consider two architectures: a single ALU with a fast multiplier (for 32-bit words, 1024 full-adder circuits) versus 32 ALUs that each have a 32-bit adder (again for a total of 1024 FAs) and do a mult by a 32-cycle shift--add. Both architectures can do 32 mults in 32 cycles. But: the serial can do 5 mults in 5 cycles, but the parallel still needs 32. The serial can do 33 mults in 33 cycles, but the parallel needs 64. The amount of hardware isn't really the same. The serial needs one instruction decoder and one memory addresser -- the parallel needs 32 of each. So on the same real estate you can bulk up the drivers and make the serial faster. And finally, the serial suffers no slowdown at all when I interleave a shuffle-exchange step (to do an FFT) -- the parallel bets bogged down in datacomm. There's an interesting variant on the parallel version that we worked on specifically for matrix mult or neural nets (same basic operation). The overflow of each of the adders fed into the bottom of an adder tree, which was one bit wide at the leaves, two bits at the next level up, etc, with a full-word accumulator at the top. So we could do fully pipelined dot products for as long as we had the data to crunch. Which was all very cute but went the way of the Connection Machine for much the same reason. (but we went faster, heh heh) ... which automatically happens when the rows of A just happen to match the interleaving. Compilers could over-dimension arrays to make this so. Note the use of Multiple Tag Mode on antique IBM-709/7090 computers, for which you had to do the same to make it useful. This helps if you're multiplying NxN matrices with only N processors, but does you no good if you actually have enough processors to have one element per processor! My design all fits on a single chip - or it will never work. See query about memory size above. Observation: I am a front-runner type, looking to find the roads that lead to here I want to go. This in preference to actually packing up the luggage and actually draging it down that road. You sound like the sort that once the things is sort of roughed out, likes to polish it up and make it as good as possible. Further, you have a LOT more actual experience doing this sort of thing with whizzbang chips than I do, and you actually understood what I was proposing with a minimum of explanation. Question: Do you have any interest in helping transform my rather rough concept to a sufficiently detailed road map that anyone with money and an interest in AGI would absolutely HAVE to fund it? I simply don't see Intel or anyone else currently running in a direction that will EVER produce an AGI-capable processor, yet my approach looks like it has a good chance if only I can smooth out the rough edges and eventually find someone to pay the
Re: [agi] Comments from a lurker...
On Monday 14 April 2008 04:56:18 am, Steve Richfield wrote: ... My present efforts are now directed toward a new computer architecture that may be more of interest to AGI types here than Dr. Eliza. This new architecture should be able to build new PC internals for about the same cost, using the same fabrication facilities, yet the processors will run ~10,000 times faster running single-thread code. This (massively-parallel SIMD) is perhaps a little harder than you seem to think. I did my PhD thesis on it and led a multi-million-dollar 10-year ARPA-funded project to develop just such an architecture. The first mistake everybody makes is to forget that the bottleneck for existing processors isn't computing power at all, it's memory bandwidth. All the cruft on a modern processor chip besides the processor is there to ameliorate that problem, not because they aren't smart enough to put more processors on. The second mistake is to forget that processor and memory silicon fab use different processes, the former optimized for fast transistors, the latter for dense trench capacitors. You won't get both at once -- you'll give up at least a factor of ten trying to combine them over the radically specialized forms. The third mistake is to forget that nobody knows how to program SIMD. They can't even get programmers to adopt functional programming, for god's sake; the only thing the average programmer can think in is BASIC, or C which is essentially machine-independent assembly. Not even LISP. APL, which is the closest approach to a SIMD language, died a decade or so back. Now frankly, a real associative processor (such as described in my thesis -- read it) would be very useful for AI. You can get close to faking it nowadays by getting a graphics card and programming it GPGPU-style. I quit architecture and got back into the meat of AI because I think that Moore's law has won, and the cycles will be there before we can write the software, so it's a waste of time to try end-runs. Associative processing would have been REALLY useful for AI in the 80's, but we can get away without it, now. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
[agi] associative processing
On Tuesday 15 April 2008 04:28:25 pm, Steve Richfield wrote: Josh, On 4/15/08, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: On Monday 14 April 2008 04:56:18 am, Steve Richfield wrote: ... My present efforts are now directed toward a new computer architecture that may be more of interest to AGI types here than Dr. Eliza. This new architecture should be able to build new PC internals for about the same cost, using the same fabrication facilities, yet the processors will run ~10,000 times faster running single-thread code. This (massively-parallel SIMD) is perhaps a little harder than you seem to think. I did my PhD thesis on it and led a multi-million-dollar 10-year ARPA-funded project to develop just such an architecture. I didn't see any attachments. Perhaps you could send me some more information about this? Whenever I present this stuff, I always emphasize that there is NOTHING new here, just an assortment of things that are decades old. Hopefully you have some good ideas in there, or maybe even some old ideas that I can attribute new thinking to. online: The CAM2000 Chip Architecture ftp://ftp.cs.rutgers.edu/pub/technical-reports/lcsr-tr-196.ps.Z look in http://www.amazon.com/exec/obidos/ASIN/0818676612 Associative Processing and Processors, Kirkelis Weems, eds. order the dissertation from University Microfilms: Associative processing: Architectures, algorithms, and applications. by Hall, John Storrs. 9511479 Unfortunately I don't have it in machine-readable form. The first mistake everybody makes is to forget that the bottleneck for existing processors isn't computing power at all, it's memory bandwidth. All the cruft on a modern processor chip besides the processor is there to ameliorate that problem, not because they aren't smart enough to put more processors on. Got this covered. Each of the ~10K ALUs has ~8 memory banks to work with, for a total of ~80K banks, so there should be no latencies except for inter-ALU communication. Have I missed something here? Either you're using static RAM (and getting a big hit in density and power) or DRAM, and getting a big hit in speed. YOU CANT AFFORD TO USE CACHE outside of a line buffer or two. You lose an order of magnitude in speed over what can be done on the CPU chip. Several big items that they put a few of on a cpu chip (besides cache) that you can't afford in each processing element: barrel shifters, floating point units, even multipliers. Instruction broadcast latency and skew. If your achitecture is synchronous you're looking at cross-chip times stuck into your instruction processing, which means TWO orders of magnitude loss from on-chip cpu cycle times. So instead of a 10K speedup you get a 100 speedup The second mistake is to forget that processor and memory silicon fab use different processes, the former optimized for fast transistors, the latter for dense trench capacitors. You won't get both at once -- you'll give up at least a factor of ten trying to combine them over the radically specialized forms. Got that covered. Once multipliers and shift matrices are eliminated and only a few adders, pipeline registers, and a little random logic remain, then the entire thing can be fabricated with *MEMORY* fab technology! Note that memories have been getting smarter (and even associative), e.g. cache memories, and when you look at their addressing, row selection, etc., there is nothing more complex than I am proposing for my ALUs. While the control processor might at first appear to violate this, note that it needs no computational speed, so its floating point and other complex instructions can be emulated on slow memory-compatible logic. You need a collective function (max, sum, etc) tree or else you're doing those operations by Caxton Foster-style bit-serial algorithms with an inescapable bus turnaround between each bit. How are you going to store an ordinary matrix? There's no layout where you can both add and multiply matrices without a raft of data motion. Either you build a general parallel communications network, which is expensive (think Connection Machine) or your data-shuffling time kills you. Again, let me mention graphics boards. They have native floating point, wide memory bandwith, and hundreds of processing units, along with fairly decent data comm onboard. Speedups over the cpu can get up to 20 or so, once the whole program is taken into account -- but for plenty of programs, the cpu is faster. The third mistake is to forget that nobody knows how to program SIMD. This is a long and complicated subject. I spent a year at CDC digging some of the last of the nasty bugs out of their Cyber-205 FORTRAN compiler's optimizer and vectorizer, whose job it was to sweep these issues under the rug. There are some interesting alternatives, like describing complex code skeletons and how to vectorize them. When someone
Re: [agi] associative processing
On Tuesday 15 April 2008 07:36:56 pm, Steve Richfield wrote: As I understand things, speed requires low capacitance, which DRAM requires higher capacitance, depending on how often you intend to refresh. However, refresh operations look a LOT like vector operations, so probably all that would be needed is some logic to watch things and if the vector operations are NOT adequate for refreshing purposes, to make the sub-processors do some refreshing before continuing. If you work the process for just enough capacitance to support a pretty high refresh rate, then you don't take such a big hit on speed. Anyway, this looked like a third choice, along with going slow with DRAM and fast with SRAM. Even with our government megabucks we never imagined getting a custom process -- at best runs on some slightly out-of date fab line. Process capacitance is a tradeoff too -- you can always just make the capacitors bigger! But even in fast transistor tech, DRAM is significantly slower. Sense amp latency... BTW, if you really want to play with the tech, I believe (don't keep a finger on the latest) that there are chips you can get that are half memory and half FPGA that you could use to try your ideas out on. (and goddamn it, the fpgas are denser and faster than full custom was back in the 80s when I was doing this!) Several big items that they put a few of on a cpu chip (besides cache) that you can't afford in each processing element: barrel shifters, floating point units, even multipliers. I don't plan on using any of these, though I do plan on having just enough there to perform the various step operations to implement these at slow rates. That works, but kills your speed by a factor of word length. it's a lot worse for floating point, because remember it's SIMD and you're doing data dependent shifts. I am planning on locally synchronous, globally asynchronous operation. Everything within a sub-processor will be pipelined synchronous, while everything connecting to them and connecting them together will be asynchronous. That's the right hardware choice, but it doesn't fit so well with the software architecture of an overall SIMD paradigm. You'd be better off going with a MIMD network of SIMD machines (a la the Sony/IBM Cell chip). I think that I can most most of the 10K speedup for most operations, but there ARE enough 100X operations to really slow it down for some types of programs. Still, a 100X processor is worth SOMETHING?! Consider Amdahl's (first) Law: if 1/nth of your program is parallelizable but the other 1/mth is inherently serial, the best speed up you can get is m. Thus if even only 1% is unparallelizable, a speedup of 100 is the absolute best you can do. But if you've slowed down the central processor by a factor of 10 to make things easier for the parallel parts, you're only doing 10 times better than an optimized purely serial machine. You need a collective function (max, sum, etc) tree or else you're doing those operations by Caxton Foster-style bit-serial algorithms with an inescapable bus turnaround between each bit. Unknown: Is there enough of this to justify the additional hardware? Also, with smart sub-processors they could work together (while jamming up the busses) to form the collective results at ~1% speed after the job has been first cut down by 10:1 by the multiple sub-processors forming the partial results. Hence, the overhead would by high for smaller arrays, but would be lost in the noise for arrays that are 10K elements. You need about twice the hardware to do a collective function tree (its a binary tree with the original PEs as its leaves. It's pipelineable, so you can run it pretty fast. Algorithmically, it makes a HUGE difference -- almost ALL the parallel algorithms my Rutgers CAM Project came up with depended on it. It's even a poor man's datacom network. (acts like a segmented bus) How are you going to store an ordinary matrix? There's no layout where you can both add and multiply matrices without a raft of data motion. Making the row length equal to the interleaving ways keeps most of the activity in individual processors. Also, arranging the interleaving so that each processor services small scattered blocks provide a big boost for long and skinny matrices. You the machine designer don't get to say what shape the user's matrices can be (or nobody will use your machine). The problem I was pointing ot is that for matrix addition, say of A and B, the rows of A must be aligned (under the same processing elements) with the rows of B, but for multiplication, the rows of A must be aligned with the COLUMNS of B. My plan was to interconnect the ~10K processors in a 2D fashion with double busses, for a total of 400 busses. In a 200x200 crossbar? Not a bad design -- if they're electrically segmentable, and you also have a nearest-neighbor torus connection, you get something like
Re: [agi] Comments from a lurker...
On Friday 11 April 2008 03:17:21 pm, Steve Richfield wrote: Steve: If you're saying that your system builds a model of its world of discourse as a set of non-linear ODEs (which is what Systems Dynamics is bout) then I (and presumably Richard) are much more likely to be interested... No it doesn't. Instead, my program is designed to work on systems that are not nearly enough known to model. THAT is the state of the interesting (at least to me) part of the real world. If the programmer builds the model of the world beforehand, and the system uses it, it's just standard narrow AI. If the system builds the model itself from unstructured inputs, it's AGI. In some sense, we know how to do that: it's called the scientific method. However, as normally explained, it leaves a lot to intuition. Form a theory isn't too far from and then a miracle occurs. In other words, we need to be a little more explicit in how our system will form a theory. Perhaps a good way to characterize any given AGI is to specify: (a) what form are its hypotheses in (b) how are they generated (c) how are they tested (d) how are they revised Would it be fair to say that Dr. Eliza tries to form a causal net / influence diagram type structure? Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=98558129-0bdb63 Powered by Listbox: http://www.listbox.com
Re: [agi] Comments from a lurker...
On Friday 11 April 2008 01:59:42 am, Steve Richfield wrote: Your experience with the medical community is not too surprising: I believe that the Expert Systems folks had similar troubles way back when. IMO the Expert Systems people deserved bad treatment! Actually, the medical expert systems of the 80s I had any conection with, such as the glaucoma expert from Rutgers, beat out human doctors in diagnoses within their field of expertise. (And still weren't adopted...) BTW, the attached paper included some remarks about Jay Forrester System Dynamics. Forrester came out of exactly the same background as Cybernetics -- working on automatic radar-directed fire-control systems, at MIT, during WWII. And both his stuff and Cybernetics consists basically of applying feedback and control theory (and general differential analysis) to things ranging from neuroscience to economics. Steve: If you're saying that your system builds a model of its world of discourse as a set of non-linear ODEs (which is what Systems Dynamics is bout) then I (and presumably Richard) are much more likely to be interested... Josh ps -- of course, you know that if you're using Excel to integrate dynamical systems, you are in a state of sin. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=98558129-0bdb63 Powered by Listbox: http://www.listbox.com
[agi] Minor milestone
Just noticed that last month, a computer program beat a professional Go player (at a 9x9 game) (one game in 4). First time ever in a non-blitz setting. http://www.earthtimes.org/articles/show/latest-advance-in-artificial-intelligence,345152.shtml http://www.computer-go.info/tc/ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=98558129-0bdb63 Powered by Listbox: http://www.listbox.com
Re: [agi] The resource allocation problem
Note that in the brain, there is a fair extent to which functions are mapped to physical areas -- this is why you can find out anything using fMRI, for example, and is the source of the famous sensory and motor homunculi (e.g. http://faculty.etsu.edu/currie/images/homunculus1.JPG). There's plasticity but it's limited and operates over a timescale of days or weeks or more. The architecture seems to have a huge parallelism at the lower levels, but ties into a serial bottleneck at the very top, i.e. conscious, level(s) -- hence the need for attentional mechanisms. On Tuesday 01 April 2008 10:30:13 am, William Pearson wrote: The resource allocation problem and why it needs to be solved first How much memory and processing power should you apply to the following things?: Visual Processing Reasoning Sound Processing Seeing past experiences and how they apply to the current one Searching for new ways of doing things Applying each heuristic etc... --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=98558129-0bdb63 Powered by Listbox: http://www.listbox.com
[agi] NewScientist piece on AGI-08
Many of us there met Celeste Biever, the NS correspondent. Her piece is now up: http://technology.newscientist.com/channel/tech/dn13446-virtual-child-passes-mental-milestone-.html Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] What should we do to be prepared?
On Sunday 09 March 2008 08:04:39 pm, Mark Waser wrote: 1) If I physically destroy every other intelligent thing, what is going to threaten me? Given the size of the universe, how can you possibly destroy every other intelligent thing (and be sure that no others ever successfully arise without you crushing them too)? You'd have to be a closed-world-assumption AI written in Prolog, I imagine. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] What should we do to be prepared?
On Thursday 06 March 2008 08:45:00 pm, Vladimir Nesov wrote: On Fri, Mar 7, 2008 at 3:27 AM, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: The scenario takes on an entirely different tone if you replace weed out some wild carrots with kill all the old people who are economically inefficient. In particular the former is something one can easily imagine people doing without a second thought, while the latter is likely to generate considerable opposition in society. Sufficient enforcement is in place for this case: people steer governments in the direction where laws won't allow that when they age, evolutionary and memetic drives oppose it. It's too costly to overcome these drives and destroy counterproductive humans. But this cost is independent from potential gain from replacement. As the gain increases, decision can change, again we only need sufficiently good 'cultivated humans'. Consider expensive medical treatments which most countries won't give away when dying people can't afford them. Life has a cost, and this cost can be met. Suppose that productivity amongst AIs is such that the entire economy takes on a Moore's Law growth curve. (For simplicity say a doubling each year.) At the end of the first decade, the tax rate on AIs will have to be only 0.1% to give the humans, free, everything we now produce with all our effort. And the tax rate would go DOWN by a factor of two each year. I don't see the AIs really worrying about it. Alternatively, since humans already own everything, and will indeed own the AIs originally, we could simply cash out and invest, and the income from the current value of the world would easily produce an income equal to our needs in an AI economy. It might be a good idea to legally entail the human trust fund... So how would you design a super-intelligence: (a) a single giant blob modelled on an individual human mind (b) a society (complete with culture) with lots of human-level minds and high-speed communication? This is a technical question with no good answer, why is it relevant? The discussion forked at the point of whether an AI would be like a single supermind or more like a society of humans... we seem to be in agreement or agree that it doesn't make much difference to the point at issue. On the other hand, the technical issue is interesting of itself, perhaps more so than the rest of the discussion :-) Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] What should we do to be prepared?
On Thursday 06 March 2008 12:27:57 pm, Mark Waser wrote: TAKE-AWAY: Friendliness is an attractor because it IS equivalent to enlightened self-interest -- but it only works where all entities involved are Friendly. Check out Beyond AI pp 178-9 and 350-352, or the Preface which sums up the whole business. There is noted in evolutionary game theory a moral ladder phenomenon -- in appropriate environments there is an evolutionary pressure to be just a little bit nicer than the average ethical level. This can raise the average over the long run. Like any evolutionarily stable strategy, it is an attractor in the appropriate space. Your point about sub-peers being resources is known in economics as the principle of comparative advantage (p. 343). I think you're essentially on the right track. Like any children, our mind children will tend to follow our example more than our precepts... Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] What should we do to be prepared?
On Thursday 06 March 2008 04:28:20 pm, Vladimir Nesov wrote: This is different from what I replied to (comparative advantage, which J Storrs Hall also assumed), although you did state this point earlier. I think this one is a package deal fallacy. I can't see how whether humans conspire to weed out wild carrots or not will affect decisions made by future AGI overlords. ;-) There is a lot more reason to believe that the relation of a human to an AI will be like that of a human to larger social units of humans (companies, large corporations, nations) than that of a carrot to a human. I have argued in peer-reviewed journal articles for the view that advanced AI will essentially be like numerous, fast human intelligence rather than something of a completely different kind. I have seen ZERO considered argument for the opposite point of view. (Lots of unsupported assumptions, generally using human/insect for the model.) Note that if some super-intelligence were possible and optimal, evolution could have opted for fewer bigger brains in a dominant race. It didn't -- note our brains are actually 10% smaller than Neanderthals. This isn't proof that an optimal system is brains of our size acting in social/economic groups, but I'd claim that anyone arguing the opposite has the burden of proof (and no supporting evidence I've seen). Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] What should we do to be prepared?
On Thursday 06 March 2008 06:46:43 pm, Vladimir Nesov wrote: My argument doesn't need 'something of a completely different kind'. Society and human is fine as substitute for human and carrot in my example, only if society could extract profit from replacing humans with 'cultivated humans'. But we don't have cultivated humans, and we are not at the point where existing humans need to be cleared to make space for new ones. The scenario takes on an entirely different tone if you replace weed out some wild carrots with kill all the old people who are economically inefficient. In particular the former is something one can easily imagine people doing without a second thought, while the latter is likely to generate considerable opposition in society. The only thing that could keep future society from derailing in this direction is some kind of enforcement installed in minds of future dominant individuals/societies by us lesser species while we are still in power. All we need to do is to make sure they have the same ideas of morality and ethics that we do -- the same as we would raise any other children. Note that if some super-intelligence were possible and optimal, evolution could have opted for fewer bigger brains in a dominant race. It didn't -- note our brains are actually 10% smaller than Neanderthals. This isn't proof that an optimal system is brains of our size acting in social/economic groups, but I'd claim that anyone arguing the opposite has the burden of proof (and no supporting evidence I've seen). Sorry, I don't understand this point. We are the first species to successfully launch culture. Culture is much more powerful then individuals, if only through parallelism and longer lifespan. What follows from it? So how would you design a super-intelligence: (a) a single giant blob modelled on an individual human mind (b) a society (complete with culture) with lots of human-level minds and high-speed communication? We know (b) works if you can build the individual human-level mind. Nobody has a clue that (a) is even possible. There's lots of evidence that even human minds have many interacting parts. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: Common Sense Consciousness [WAS Re: [agi] reasoning knowledge]
On Wednesday 27 February 2008 12:22:30 pm, Richard Loosemore wrote: Mike Tintner wrote: As Ben said, it's something like multisensory integrative consciousness - i.e. you track a subject/scene with all senses simultaneously and integratedly. Conventional approaches to AI may well have trouble in this area, but since my approach has been directed at these kinds of issues since the very beginning, to me it looks relatively straightforward in principle. The real issues are elsewhere. True. I'd go farther and point out just where they are: You need to have a system with recognition / action generation integrated between the sensory modalities to be a trainable animal. To be intelligent, the system has to be able to *invent new modalities / representations / concepts itself* and integrate them into the existing mechanism. Josh --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] reasoning knowledge
On Tuesday 26 February 2008 12:33:32 pm, Jim Bromer wrote: There is a lot of evidence that children do not learn through imitation, at least not in its truest sense. Haven't heard of any children born into, say, a purely French-speaking household suddenly acquiring a full-blown competence in Japanese... --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
[agi] color
or see this one: http://www.boingboing.net/2008/02/08/color-tile-optical-i.html http://www.lottolab.org/Colour%20illusions%20page.html There's a tile-covered cube shown against 2 backgrounds, and the blue tiles in one are the same actual color as the yellow ones in the other. On Thursday 21 February 2008 03:34:27 am, Bob Mottram wrote: On 20/02/2008, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: So, looking at the moon, what color would you say it was? As Edwin Land showed colour perception does not just depend upon the wavelength of light, but is a subjective property actively constructed by the brain. http://en.wikipedia.org/wiki/Color_constancy http://youtube.com/watch?v=ZiTg4kRt13w --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] would anyone want to use a commonsense KB?
Looking at the moon won't help -- it might be the case that it described a particular appearance that only had a slight resemblance to other blue things (as in red hair), for example. There are some rare conditions (high stratospheric dust) which can make the moon look actually blue. In fact blue moon is generally taken to mean, metaphorically, something very rare (or even impossible) or the second full moon in a given month (which happens about every two-and-a-half years on the average). ask someone is of course what human kids do a lot of. An AI could do this, or look it up in Wikipedia, or the like. All of which are heuristics to reduce the ambiguity/generality in the information stream. The question is do enough heuristics make an autogenous AI or is there something more fundamental to its structure? On Wednesday 20 February 2008 12:27:59 pm, Ben Goertzel wrote: The trick to understanding once in a blue moon is to either -- look at the moon or -- ask someone --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] would anyone want to use a commonsense KB?
So, looking at the moon, what color would you say it was? Here's what text mining might give you (Google hits): blue moon 11,500,000 red moon 1,670,000 silver moon 1,320,000 yellow moon 712,000 white moon 254,000 golden moon 163,000 orange moon 122,000 green moon 105,000 gray moon 9,460 To me, the moon varies from a deep orange to brilliant white depending on atmospheric conditions and time of night... none of which would help me understand the text references. On Wednesday 20 February 2008 02:02:52 pm, Ben Goertzel wrote: On Feb 20, 2008 1:34 PM, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: Looking at the moon won't help -- of course it helps, it tells you that something odd is with the expression, as opposed to say yellow sun ... --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] would anyone want to use a commonsense KB?
On Wednesday 20 February 2008 02:58:54 pm, Ben Goertzel wrote: I note also that a web-surfing AGI could resolve the color of the moon quite easily by analyzing online pictures -- though this isn't pure text mining, it's in the same spirit... U -- I just typed moon into google and at the top of the page it gives three pictures. Two are thin sliver crescents. The third, of a full moon, is distinctly blue. There seems to be an assumption in this thread that NLP analysis of text is restricted to simple statistical extraction of word-sequences... I certainly make no such assumption. I offered the stats to point out the kind of traps that lie in wait for the hapless text-miner. As I am sure you are fully aware, you can't parse English without a knowledge of the meanings involved. (The council opposed the demonstrators because they (feared/advocated) violence.) So how are you going to learn meanings before you can parse, or how are you going to parse before you learn meanings? They have to be interleaved in a non-trivial way. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] would anyone want to use a commonsense KB?
OK, imagine a lifetime's experience is a billion symbol-occurences. Imagine you have a heuristic that takes the problem down from NP-complete (which it almost certainly is) to a linear system, so there is an N^3 algorithm for solving it. We're talking order 1e27 ops. Now using HEPP = 1e16 x 30 years = 1e9 secs, you get a total crunch for the human of 1e25 ops. That's close enough to call even, I think. Learning order is easily worth a couple orders of magnitude in problem complexity. Let's build a big cluster... On Wednesday 20 February 2008 03:51:28 pm, Ben Goertzel wrote: Feeding all the ambiguous interpretations of a load of sentences into a probabilistic logic network, and letting them get resolved by reference to each other, is a sort of search for the most likely solution of a huge system of simultaneous equations ... i.e. one needs to let each, of a huge set of ambiguities, be resolved by the other ones... This is not an easy problem, but it's not on the face of it unsolvable... But I think the solution will be easier with info from direct experience to nudge the process in the right direction... Ben --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] would anyone want to use a commonsense KB?
A PROBABILISTIC logic network is a lot more like a numerical problem than a SAT problem. On Wednesday 20 February 2008 04:41:51 pm, Ben Goertzel wrote: On Wed, Feb 20, 2008 at 4:27 PM, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: OK, imagine a lifetime's experience is a billion symbol-occurences. Imagine you have a heuristic that takes the problem down from NP-complete (which it almost certainly is) to a linear system, so there is an N^3 algorithm for solving it. We're talking order 1e27 ops. That's kind of specious, since modern SAT and SMT solvers can solve many realistic instances of NP-complete problems for large n, surprisingly quickly... and without linearizing anything... Worst-case complexity doesn't mean much... ben --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] would anyone want to use a commonsense KB?
It's probably not worth too much taking this a lot further, since we're talking in analogies and metaphors. However, it's my intuition that the connectivity in a probabilistic formulation is going to produce a much denser graph (less sparse matrix) than what you find in the SAT problems that the solvers do so well on. And I seriously doubt that a general SMT solver + prob. theory is going to beat a custom probabilistic logic solver. On Wednesday 20 February 2008 05:31:59 pm, Ben Goertzel wrote: Not necessarily, because --- one can encode a subset of the rules of probability as a theory in SMT, and use an SMT solver -- one can use probabilities to guide the search within an SAT or SMT solver... ben --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=95818715-a78a9b Powered by Listbox: http://www.listbox.com
Re: [agi] Wozniak's defn of intelligence
It's worth noting in this connection that once you get up to the level of mammals, everything is very high compliance, low stiffness, mostly serial joint architecture (no natural Stewart platforms, although you can of course grab something with two hands if need be) typically with significant energy storage in the power train (i.e. springs). This means that the control has to be fully Newtonian, something most commercial robotics haven't gotten up to yet. I think that state of the art is just now getting to dynamically-stable-only biped walkers. I've seen a couple of articles in the past year, but it certainly isn't widespread, and it remains to be seen how real. Josh On Sunday 10 February 2008 04:35:13 pm, Bob Mottram wrote: The idea that robotics is only about software is fiction. Good automation involves cooperation between software, electrical and mechanical engineers. In some cases problems are much better solved electromechanically than by software. For example, no matter how smart the software controlling it, a two fingered gripper will only be able to deal with a limited sub-set of manipulation tasks. Likewise a great deal of computation can be avoided by introducing variable compliance, and making clever use of materials to juggle energy around the system (biological creatures use these tricks all the time). - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=94603346-a08d2f
[agi] Wozniak's defn of intelligence
[ http://www.chron.com/disp/story.mpl/headline/biz/5524028.html ] Steve Wozniak has given up on artificial intelligence. What is intelligence? Apple's co-founder asked an audience of about 550 Thursday at the Houston area's first Up Experience conference in Stafford. His answer? A robot that could get him a cup of coffee. You can come into my house and make a cup of coffee and I can go into your house and make a cup of coffee, he said. Imagine what it would take for a robot to do that. It would have to negotiate the home, identify the coffee machine and know how it works, he noted. But that is not something a machine is capable of learning — at least not in his lifetime, added Wozniak, who rolled onto the stage on his ever-present Segway before delivering a rapid-fire speech on robotics, his vision of robots in classrooms and the long haul ahead for artificial intelligence. ... Any system builders here care to give a guess as to how long it will be before a robot, with your system as its controller, can walk into the average suburban home, find the kitchen, make coffee, and serve it? - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=93139505-4aa549
Re: [agi] Wozniak's defn of intelligence
On Friday 08 February 2008 10:16:43 am, Richard Loosemore wrote: J Storrs Hall, PhD wrote: Any system builders here care to give a guess as to how long it will be before a robot, with your system as its controller, can walk into the average suburban home, find the kitchen, make coffee, and serve it? Eight years. My system, however, will go one better: it will be able to make a pot of the finest Broken Orange Pekoe and serve it. In the average suburban home? (No fair having the robot bring its own teabags, (or would it be loose tea and strainer?) or having a coffee machine built in, for that matter). It has to live off the land... - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=93139505-4aa549
Re: [agi] OpenMind, MindPixel founders both commit suicide
Breeds There a Man...? by Isaac Asimov On Saturday 19 January 2008 04:42:30 pm, Eliezer S. Yudkowsky wrote: http://www.wired.com/techbiz/people/magazine/16-02/ff_aimystery?currentPage=all I guess the moral here is Stay away from attempts to hand-program a database of common-sense assertions. -- Eliezer S. Yudkowsky http://singinst.org/ Research Fellow, Singularity Institute for Artificial Intelligence - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=87842867-40e15f
Re: Possibility of superhuman intelligence (was Re: [agi] AGI and Deity)
On Friday 21 December 2007 09:51:13 pm, Ed Porter wrote: As a lawyer, I can tell you there is no clear agreed upon definition for most words, but that doesn't stop most of us from using un-clearly defined words productively many times every day for communication with others. If you can only think in terms of what is exactly defined you will be denied life's most important thoughts. And in particular, denied the ability to create a working AI. It's the inability to grasp this insight that I call formalist float in the book (yeah, I wish I could have come up with a better phrase...) and to which I attribute symbolic AI's Glass Ceiling. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=78789088-cf88d9
[agi] one more indication
... that during sleep, the brain fills in some inferencing and does memory organization http://www.nytimes.com/2007/10/23/health/23memo.html?_r=2adxnnl=1oref=sloginref=scienceadxnnlx=1193144966-KV6FdDqmqr8bctopdX24dw (pointer from Kurzweil) - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56595329-baa3cc
Re: [agi] An AGI Test/Prize
On Monday 22 October 2007 08:05:26 am, Benjamin Goertzel wrote: ... but dynamic long-term memory, in my view, is a wildly self-organizing mess, and would best be modeled algebraically as a quadratic iteration over a high-dimensional real non-division algebra whose multiplication table is evolving dynamically as the iteration proceeds Holy writhing Mandelbrot sets, Batman! Why real and non-division? I particularly don't like real -- my computer can't handle the precision :-) Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56270025-9c1ac7
Re: Bogus Neuroscience [WAS Re: [agi] Human memory and number of synapses]
On Monday 22 October 2007 08:01:55 pm, Richard Loosemore wrote: Did you ever try to parse a sentence with more than one noun in it? Well, all right: but please be assured that the rest of us do in fact do that. Why make insulting personal remarkss instead of explaining your reasoning? (RL, Sat Oct 6 02:48:54 2007) - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56508702-7de092
Re: Bogus Neuroscience [WAS Re: [agi] Human memory and number of synapses]
On Monday 22 October 2007 08:48:20 pm, Russell Wallace wrote: On 10/23/07, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: Still don't buy it. What the article amounts to is that speed-reading is fake. No kind of recognition beyond skimming (e.g. just ignoring a substantial proportion of the text) is called for to explain the observed performance. And I'm saying nevermind articles, try it for yourself. I tried the experiment, before I wrote that earlier post, it's easy to do. You'll find you do in fact recognize (I'm making no claims about rate of comprehension or retention, I'm only addressing the question of recognition) many words simultaneously, in parallel, without needing to saccade serially to each one. Still don't buy it. Saccades are normally well below the conscious level, and a vast majority of what goes on cognitively is not available to introspection. Any good reader gets to the point where the sentence meanings, not the words at all, are the only thing that breaks into the conscious level. (you can read with essentially complete semantic comprehension and still be quite unable to repeat any of the text verbatim.) BTW, I'm not trying to say that no concurrent recognition happens in the brain -- I'm sure that it does. I merely maintain that I haven't seen any evidence to convince me that it occurs in that particular part of vision. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56509948-3b75bb
Re: Bogus Neuroscience [WAS Re: [agi] Human memory and number of synapses]
On Monday 22 October 2007 09:33:24 pm, Edward W. Porter wrote: Richard, ... Are you capable of understanding how that might be considered insulting? I think in all seriousness that he literally cannot understand. Richard's emotional interaction is very similar to that of some autistic people I have known. The recent spat over Turing completeness started when I made a remark I thought to be humorous -- *quoting exactly the words Richard had used to make the same joke* to someone else -- and he took the same words he had said as a disparaging insult when said to him. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56513821-de495c
Re: Bogus Neuroscience [WAS Re: [agi] Human memory and number of synapses]
You can DO them consciously but that doesn't necessarily mean that you can intentionally become conscious of the ones you are doing unconsciously. Try cutting a hole in a piece of paper and moving it smoothly across another page that has text on it. When your eye tracks the smoothly moving page, what appears through the hole is a blur. Josh On Monday 22 October 2007 10:23:12 pm, Russell Wallace wrote: On 10/23/07, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: Still don't buy it. Saccades are normally well below the conscious level, and a vast majority of what goes on cognitively is not available to introspection. Any good reader gets to the point where the sentence meanings, not the words at all, are the only thing that breaks into the conscious level. (you can read with essentially complete semantic comprehension and still be quite unable to repeat any of the text verbatim.) Sure, but saccades and word recognition are like breathing - normally they operate subconsciously, but you can become aware and take control of them if you so choose. Again this isn't abstruse theory - try it and see, the experiment can be done in seconds. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56519427-089861
Re: [agi] Poll
On Friday 19 October 2007 10:36:04 pm, Mike Tintner wrote: The best way to get people to learn is to make them figure things out for themselves . Yeah, right. That's why all Americans understand the theory of evolution so well, and why Britons have such an informed acceptance of genetically-modified foods. It's why Galileo had such an easy time convincing the Church that the earth goes around the sun. It's why the Romans widely adopted the steam engine following its invention by Heron of Alexandria. It's why the Inquisition quickly realized that witchcraft is a superstition, rather than burning innocent women at the stake. The truth is exactly the opposite: Humans are built to propagate culture memetically, by copying each other; the amount we know individually by this process is orders of magnitude greater than what we could have figured out for ourselves. Reigning orthodoxy of thought is *very hard* to dislodge, even in the face of plentiful evidence to the contrary. Isaac Asimov famously said that the most exciting moment in science is when someone says, That's funny... But the reason to point it out is that it *doesn't* happen all the time, even in science (it's not normal science in Kuhn's phrase), and even less so outside of it. In the real world, when people get confused and work out a way around it, what they're learning is not an inventive synthesis of the substance at issue, but an attention filter. And that, for the average person, is usually just picking an authority figure. Theirs not to reason why; theirs but to do and die. Humans are *stupid*, Mike. You're still committing the superhuman human fallacy. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55686241-899d6e
Re: [agi] Poll
On Friday 19 October 2007 01:30:43 pm, Mike Tintner wrote: Josh: An AGI needs to be able to watch someone doing something and produce a program such that it can now do the same thing. Sounds neat and tidy. But that's not the way the human mind does it. A vacuous statement, since I stated what needs to be done, not how to do it. We start from ignorance and confusion about how to perform any given skill/ activity Particularly how to build an AGI :-) - and while we then acquire an enormous amount of relevant routines - we never build a whole module or program for any activity. If what you're trying to say is nobody's perfect, well, duh. If you're trying to say humans don't actually acquire skills, speak for yourself. We never stop learning, whether we're committed to that attitude philosophically or not. Some of us never *start* learning... And we never stop being confused. FDSN. Are you certain about how best to write programs? Or have sex? Or a conversation? Or play chess? Or tennis? All our activities, like those, demand and repay a lifetime's study. An AGI will have to have a similar approach to enjoy any success. How stupid of me not to realize that my vague ideas on how to build a program that can learn by watching, would not instantly achieve superhuman, Godlike, mathematically optimal performance on every possible task at first sight. I am awed by the brilliance of this insight. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55547253-8b4a4f
[agi] evolution-like systems
There's a really nice blog at http://karmatics.com/docs/evolution-and-wisdom-of-crowds.html talking about the intuitiveness (or not) of evolution-like systems (and a nice glimpse of his Netflix contest entry using a Kohonen-like map builder). Most of us here understand the value of a market or evolutionary model for internal organization and learning in the mind. How many have a model of mind that explains why some people find these models intuitive while many do not? Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55347372-773540
Re: [agi] symbol grounding QA
Remember that Eliezer is using holonic to describe *conflict resolution* in the interpretation process. The reason it fits Koestler's usage is that it uses *both* information about the parts that make up a possible entity and the larger entities it might be part of. Suppose we see the sentence The cut sut on the mut, written in longhand. We would rapidly come to understand that the writer didn't close his as and that the sentence had to do with domestic felines. An essential part of this process would be resolving the conflict between the different possible interpretations of the letters and the words. Holonic neatly captures this process by emphasizing that the entities being disambiguated are both made up of parts and are themselves parts of larger entities. Is that a fair exegesis, Eliezer? Josh On Wednesday 17 October 2007 06:43:52 pm, Edward W. Porter wrote: JOSH, I KNEW SERRES SYSTEM WAS ONLY FEED FORWARD, AND ONLY DEALS WITH CERTAIN ASPECTS OF VISION, BUT I THINK IT HAS AMAZINGLY IMPRESSIVE PERFORMANCE FOR SUCH A RELATIVELY SIMPLE SYSTEM, AND A LOT OF IT IS AUTOMATICALLY LEARNED. IS IT HOLONIC? IT DEFINITELY DOESNT JUST DIVIDE VISUAL STATE SPACE UP INTO THE EQUIVALENT OF THE BLINDLY SELECTED SUBSPACES. IT LEARNS PATTERNS, AND PATTERNS OF PATTERNS, AND GENERALIZATIONS OF PATTERNS. AND WHAT IT LEARNS, ARE, AS I REMEMBER IT, PATTERNS THAT SOMEHOW USEFULLY DIVIDE VISUAL EXPERIENCE AT THE LEVEL OF COMPLEXITY REPRESENTED BY THAT LEVEL OF PATTERN. SO THE PATTERNS BEGIN TO REPRESENT USEFUL SHAPES, AND PATTERNS OF SUCH SHAPES. SO I THINK IT IS SOMEWHAT ANALOGOUS TO DIVIDING UP A BODY INTO UNITS BASED ON THE COHERENT ROLES THEY PLAY IN A HIERARCHY OF SUCH PATTERNS, RATHER THAN JUST SOME ARBITRARY PATTERN THAT IS INDEPENDENT OF WHAT IS HAPPENING ABOVE OR BELOW IT. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=54903061-483077
[agi] Poll
I'd be interested in everyone's take on the following: 1. What is the single biggest technical gap between current AI and AGI? (e.g. we need a way to do X or we just need more development of Y or we have the ideas, just need hardware, etc) 2. Do you have an idea as to what should should be done about (1) that would significantly accelerate progress if it were generally adopted? 3. If (2), how long would it take the field to attain (a) a baby mind, (b) a mature human-equivalent AI, if your idea(s) were adopted and AGI seriously pursued? 4. How long to (a) and (b) if AI research continues more or less as it is doing now? Thanks, Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=54906039-62cabb
Re: [agi] symbol grounding QA
On Thursday 18 October 2007 09:28:04 am, Edward W. Porter wrote: Josh, According to that font of undisputed truth, Wikipedia, the general definition of a holon is: ... Since a holon is embedded in larger wholes, it is influenced by and influences these larger wholes. And since a holon also contains subsystems, or parts, it is similarly influenced by and influences these parts. Information flows BIDIRECTIONALLY between smaller and larger systems. (emphasis added) ... but in a feedforward network information only flows one way. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=54918024-4e3e5a
Re: [agi] symbol grounding QA
Holonic as used by Koestler implies at least a little something more than hierarchical. I think he meant something I would call coherent levels of abstraction, e.g. describing a body as a system of organs or an organ as a system of cells, such that you can usefully do a data-hiding encapsulation. I can do this if I partition the body into organs, for example, but not if I divide it up into a hierarchy of cubical volumes as if using an oct-tree. I don't see Serre's hierarchy as being particularly holonic. His levels correspond to levels of complexity and region size, but not to constituents that partition the image into a coherent set of parts that are wholes-in-themselves. I don't see anything at all addressing this level of architectural concern in Hawkins' stuff, holonic or otherwise. There may be new stuff since OI, but what he said there seemed to assume a hierarchical structure in the units that would follow some ontology of the datastream they were interpreting, without saying anything about where the ontology came from. Eliezer appears to be using the phrase holonic conflict resolution to mean more or less the same thing I use active interpretation for (Beyond AI p. 229-232). The basic idea is that in a hierarchical stack of pattern matchers, information flows down (and in my model, across) as well as up, allowing the environment of a part to affect its interpretation in combination with its constituents. I find this use of the term to be congenial with the original meaning, and I'm happy to follow Eliezer's usage. (Note BTW that the Poggio/Serre model is strictly and explicitly feedforward.) Thanks for bringing it up -- this has been fun and enlightening. Josh On Tuesday 16 October 2007 11:19:42 pm, Edward W. Porter wrote: In response to below post from Josh Hall: I am using Holonic as Eliezer S. Yudkowsky used in in his LEVELS OF ORGANIZATION IN GENERAL INTELLIGENCE in which he said Holonic is a useful word to describe the simultaneous application of reductionism and holism, in which a single quality is simultaneously a combination of parts and a part of a greater whole [Koestler67]. Note that holonic does not imply strict hierarchy, only a general flow from high-level to low-level and vice versa. For example, a single feature detector may make use of the output of lower-level feature detectors, and act in turn as an input to higher-level feature detectors. The information contained in a mid-level feature is then the holistic sum of many lower-level features, and also an element in the sums produced by higher-level features. If you pick one vantage point in a holonic structure and look down (reductionism) you find parts composing the local whole, with simpler behaviors that contribute to local complexity; if you look up (holism) you find a greater whole to which local parts contribute, and more complex processes which local behaviors support. I basically use it to be representation in roughly hierarchical network, such as that defined by Jeff Hawkings, or in the Serre PhD thesis I have cited so often. Representations using such nets have many advantages, such as functional invariance, ability to inherit information from more general nodes, etc. -Original Message- From: J Storrs Hall, PhD [mailto:[EMAIL PROTECTED] Sent: Tuesday, October 16, 2007 11:01 PM To: agi@v2.listbox.com Subject: Re: [agi] symbol grounding QA On Tuesday 16 October 2007 08:43:23 pm, Edward W. Porter wrote: ... holonic pattern matching, ... Now there's a word you don't hear every day :-) I've always thought of it as a feature of Arthur Koestler's somewhat poetic ontology of hierarchy. And it appears to enjoy a minor vogue as a subspecies of agent-based systems. But you'll have to explain what holonic pattern matching is, please? Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=54562745-5c
Re: [agi] symbol grounding QA
On Monday 15 October 2007 04:45:22 pm, Edward W. Porter wrote: I mis-understood you, Josh. I thought you were saying semantics could be a type of grounding. It appears you were saying that grounding requires direct experience, but that grounding is only one (although perhaps the best) possible way of providing semantic meaning. Am I correct? That's right as far as it goes. The term grounding is very commonly associated with symbol in such a way as to imply that semantics only arise from the fact that symbols have referents in the real world (or whatever). This is the view Harnad espoused with his dictionary example. The view I suggest instead is that it's not the symbols per se, but the machinery that manipulates them, that provides semantics. Dictionaries have no machinery. Turing machines, on the other hand, do -- so the symbols used by a Turing machine may have meaning in a sense even though there is nothing in the external world that they map to. (A case in point would be the individual bits that your calculator manipulates.) I would tend to differ with the concept that grounding only relates to what you directly experience. (Of course it appears to be a definitional issue, so there is probably no theoretical right or wrong.) I consider what I read, hear in lectures, and see in videos about science or other abstract fields such as patent law to be experience, even though the operative content in such experiences is derived second, third, fourth, or more handed. Harnad would say that you understand the words you read and hear because, as a human body, you have already grounded them in experience or can make use of a definition in terms that are already grounded, avoiding circular definitions. I would say that you can understand sentences and arguments you hear because you have an internal model that can make predictions based on the sentences and inferences based on the arguments. The only reason the distinction makes much of a difference is that the grounding issue is used as an argument that an AI must be embodied, having direct sensory experience. It's part of an effort to understand why classical AI faltered in the 80's and thus what must be done differently to make it go again. I give a good overview of the arguments in Beyond AI chapters 5 and 7. In Richard Loosemores above mentioned informative post he implied that according to Harnad a system that could interpret its own symbols is grounded. I think this is more important to my concept of grounding than from where the information that lets the system do such important interpretation comes. To me the important distinction is are we just dealing with realtively naked symbols, or are we dealing with symbols that have a lot of the relations with other symbols and patterns, something like those Pei Wang was talking about, that lets the system use the symbols in an intelligent way. Richard is right in that if a system formed its own symbols from sensory experience, they would be grounded in Harnad's sense. In the case of the relations between the symbols, it isn't clear -- there's plenty of relations specified between symbols in Harnad's ungrounded dictionary. I would distinguish between relations that were merely a static structure, as in the dictionary, and ones that were part of a mechanism (which could be had by adding say an inference procedure to the definitions). Usually for such relations and patterns to be useful in a world, they have to have come directly or indirectly from experience of that world. But again, it is not clear to me that they has to come first handed. Exactly my point. The vast majority of what we learn is second- (or nth-) hand, mediated by symbol structures. And it's the structures that we need to be thinking about, not the symbols. It seems ridiculous to say that one could have two identical large knowledge bases of experiential knowledge each containing millions of identically interconnected symbols and patterns in two AGI having identical hardware, and claim that the symbols in one were grounded but those in the other were not because of the purely historical distinction that the sensing to learn such a knowledge was performed on only one of the two identical systems. Again, exactly my point. It wouldn't matter if one was copied from the other, or reverse-engineered, or produced by a random-number generator (as unlikely as that would be). Or imagine that you had a robot who built its own symbols from physical experience until it was intelligent, and then was cut off from the sensors and was only connected thru a tty, doing Turing tests. The symbols didn't lose meaning -- the words of someone blinded in an accident are not suddenly meaningless! So if we built an AI de novo that had the same program as the robot, it would be ridiculous to say that its symbols had no meaning, as well. Josh - This list is sponsored by AGIRI:
Re: [agi] symbol grounding QA
On Tuesday 16 October 2007 09:24:34 am, Richard Loosemore wrote: If I may interject: a lot of confusion in this field occurs when the term semantics is introduced in a way that implies that it has a clear meaning [sic]. Semantics does have a clear meaning, particularly in linguistics and computer science. In programming language theory, it has a very precise and formal meaning (example: http://people.cs.uchicago.edu/%7Ejacobm/pubs/scheme-semantics.pdf) with deep underpinnings in logic and math. There are, of course, many hangers-on to AI who haven't done their homework, and thus are confused about its meaning. I start to wonder what they're putting on their cornflakes in the morning. Cornflakes are bad for you, consisting entirely of carbohydrates. The trivial sense of semantics don't apply, and the deeper senses are so vague that they are almost synonymous with grounding. Completely wrong. Grounding is a fairly shallow concept that falls apart as an explanation of meaning under fairly moderate scrutiny. Semantics is, by definition, whatever it takes to understand meaning. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=54101333-25b187
Re: [agi] symbol grounding QA
On Tuesday 16 October 2007 03:24:07 pm, Edward W. Porter wrote: AS I SAID ABOVE, I AM THINKING OF LARGE COMPLEX WEBS OF COMPOSITIONAL AND GENERALIZATIONAL HIERARCHIES, ASSOCIATIONS, EPISODIC EXPERIENCES, ETC, OF SUFFICIENT COMPLEXITY AND DEPTH TO REPRESENT THE EQUIVALENT OF HUMAN WORLD KNOWLEDGE. SO, IS THAT WHAT YOU MEAN BY STRUCTURES? What do these webs of associations *do*? Are they like sentences in a book, waiting for some homunculus to read them, or are they like components in a circuit, an active machine and not just a static picture? If components, how do you specify what their functions are? Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=54357834-180ee3
Re: [agi] symbol grounding QA
On Monday 15 October 2007 10:21:48 am, Edward W. Porter wrote: Josh, Also a good post. Thank you! You seem to be defining grounding as having meaning, in a semantic sense. Certainly it has meaning, as generally used in the philosophical literature. I'm arguing that its meaning makes an assumption about the nature of semantics that obscures rather than informing some important questions. If so, why is it a meaningless question to ask if 2 in your calculator has grounding, since you say the calculator has limited but real semantics. Would not the relationships 2 has to other numbers in the semantics of that system be a limited form of semantics. Not meaningless -- I'd just say that for the 2 in my calculator, the answer is no, in Harnad's fairly precise sense of grounding. Whereas the calculator clearly does have the appropriate semantics for arithmetic. And what other source besides experience can grounding come from, either directly or indirectly? The semantic model of arithmetic in you calculator was presumably derived from years of human experience that found the generalities of arithmetic to be valid and useful in the real world of things like sheep, cows, and money. I'd claim that this is a fairly elastic use of the term experience. Typically one assumes that experience means the experience of the person, AI, or whatever that we're talking about, in this case the calculator. The 2 in the calculator clearly does not get its semantics from the calculator's experience. If we allow an expanded meaning of experience as including the experience of the designer of the system, we more or less have to allow it to mean any feedback in the evolutionary process that produced the low-level semantic mechanisms in our own brains. This strains my concept of the word a bit. Whether we allow that or not, I claim that we can talk about a more proximate criterion for semantics, which is that the system forms a model of some phenomenon of interest. It may well be that experience, narrowly or broadly construed, is often the best way of producing such a system (and in fact I believe that it is), but the questions are logically separable. It's conceivable to have a system that has the appropriate semantics that was just randomly produced, for example, whereas the reverse, a system basedon experience that DOESN'T model the phenomenon, wouldn't have the semantics in my view. The most common case of a randomly-created semantic model that didn't arise from experience is the creation of social realities by fiat, as in the classic case of money. We (somebody) made up what money is and how it should work, and the reality that system models followed because we built the reality to match the system, rather than the other way around. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=53722312-b0a1a5
Re: [agi] symbol grounding QA
On Monday 15 October 2007 01:25:22 pm, Edward W. Porter wrote: I'm arguing that its meaning makes an assumption about the nature of semantics that obscures rather than informing some important questions WHAT EXACTLY DO YOU MEAN? I think that will become clearer below: I JUST READ THE ABSTRACT OF Harnad, S. (1990) The Symbol Grounding Problem. Physica D 42: 335-346. ON THE WEB, AND IT SEEMS HE IS TALKING ABOUT USING SOMETHING LIKE A GEN/COMP HIERARCHY OF REPRESENTATION HAVING AS A BOTTOM LAYER SIMPLE SENSORY PATTERNS, AS A BASIS OF GROUNDING. Basically. He proposes his notion of grounding as an escape from the problem, as he describes it, of learning Chinese from a Chinese-Chinese dictionary. You chase definitions around and around, but never get to where the symbols have any meaning to you. SO HOW DOES THE CALCULATOR HAVE SIGNIFICANTLY MORE OF THIS TYPE OF GROUNDING THAN 10 IN BINARY. What i said was that the calculator does NOT nave this kind of grounding: “I'd just say that for the 2 in my calculator, the answer is no, in Harnad's fairly precise sense of grounding. What it does have is an internal system whose objects and workings reflect the ontology and etiology of arithmetic as we see it in the outside world. If I type 2+3 into the calculator, it displays 5. If I hold 2 sandwiches in my left hand, and 3 in my right, and put them all on a plate, when I count the sandwiches on the plate, lo and behold, there are 5 sandwiches. So, I claim, the symbols in the calculator have meaning because they are part of a model that reflects some phenomenon of interest, and can be used to predict it. They have NO grounding in Harnad's sense -- the calculator has no sensory patterns that reflect quanitites as we perceive them. Typically one assumes that experience means the experience of the person, AI, or whatever that we're talking about... IF THAT IS TRUE, MUCH OF MY UNDERSTANDING OF SCIENCE AND AI IS NOT GROUNDED, SINCE IT HAS BEEN LEARNED LARGELY BY READING, HEARING LECTURES, AND WATCHING DOCUMENTARIES. Yes indeed -- but that doesn't mean (necessarily) that what you know is wrong, as long as the models you have reflect the realities they should. And this is why I say grounding in Harnad's sense is a red herring. I claim that we can talk about a more proximate criterion for semantics, which is that the system forms a model of some phenomenon of interest. It may well be that experience, narrowly or broadly construed, is often the best way of producing such a system (and in fact I believe that it is), but the questions are logically separable. THIS MAKES SENSE, BUT THIS WOULD COVER A LOT OF SYSTEM THAT ARE NOT GROUNDED IN THE WAY MOST OF USE US THAT WORD Again an argument to use a different word. I know a lot of science for which I haven't personally done the experiments that I believe are the justifications for my knowledge. I'd claim that my concept of inertia is grounded in personal experience but that my concept of magnetic induction is more or less synthesized of other abstract mathematical concepts. But it happens to work well enough that I can build working transformers. So I believe it's true in the sense that it is a valid model of the phenomenon. It's conceivable to have a system that has the appropriate semantics that was just randomly produced... I ASSUME THAT BY RANDOMLY PRODUCED, YOU DONT MEAN THAT THE SYSTEM WOULD BE TOTALLY RANDOM, IN WHICH CASE IT WOULD SEEM THE CONCEPT OF A MODEL WOULD BE MEANINGLESS. Nope. If the model was formed at random, BUT HAPPENS TO MATCH REALITY anyway, it has as much meaning as one built up by painstaking experimentation. But of course the probability of this happening is vanishingly small if the model is complex. I WOULD PICK AS A GOOD EXAMPLE OF A SEMANTIC SYSTEM THAT IS SOMEWHAT INDEPENDENT OF PHYSICAL REALITY, BUT YET HAS PROVED USEFUL, AT LEAST FOR ENTERTAINMENT, IS THE HARRY POTTER SERIES, OR SOME OTHER FICTIONAL WORLD WHICH CREATES A FICTIONAL REALITY IN WHICH THERE IS A CERTAIN REGULARITY TO THE BEHAVIOR AND CHARACTERISTICS OF THE FICTITIOUS PEOPLE AND PLACES IT DESCRIBES. There's a physical reality that the world of magic reflects, oddly enough, that's very close to home. The two key magical laws, i.e. of similarity and contagion, are remarkably good descriptions of the heuristics by which our minds form associations... Cheers! Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=53804699-7f7e7e
Re: [agi] symbol grounding QA
On Monday 15 October 2007 01:57:18 pm, Richard Loosemore wrote: AI programmers, in their haste to get something working, often simply write some code and then label certain symbols as if they are meaningful, when in fact they are just symbols-with-labels. This is quite true, but I think it is a lot closer to McDermott's critique (Artificial Intelligence meets Natural Stupidity) than to Harnad's. Harnad shares the typical epistemologist's assumption that for a symbol to have meaning, it must have an aboutness, i.e. it must refer to something in some external (although perhaps imaginary) world. I happen to think that Solomonoff's inductive formulation of AI more or less demolished this particular philosophical set of (often unstated) assumptions, which were after all responsible of 3 millennia of spectacularly unproductive pontification. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=53806349-fa774c
Re: [agi] symbol grounding QA
This is a very nice list of questions and makes a good framework for talking about the issues. Here are my opinions... On Saturday 13 October 2007 11:29:16 am, Pei Wang wrote: *. When is a symbol grounded? Grounded is not a good way of approaching what we're trying to get at, which is semantics. The term implies that meanings are inherent in words, and this obscures the fact that semantics are a property of systems of which words are only a part. Example: is the symbol 2 grounded in my calculator? there's no pointer from the bit pattern to an actual pair of anything. However, when I type in 2+2 it tells me 4. There is a system implemented that is a semantic model of arithmetic, and 2 is connected into the system in such a way that I get the right answer when I use it. Is 2 grounded? meaningless question. Does the calculator have a limited but real semantics of arithmetic? Definitely. *. What is wrong in traditional symbolic AI on this topic? These systems didn't come close to implementing a competent semantics of the parts of the world they were claimed to understand. *. What is the experience needed for symbol grounding? Experience per se isn't strictly necessary, but you have get the semantics from somewhere, and experience is a good source. The scientific method relies heavily on experience in the form of experiment to validate theories, for example. *. For the symbols in an AGI to be grounded, should the experience of the system be the same, or very similar, to human sensory experience? No, as long as it can form coherent predictive models. On the other hand, some overlap may be necessary to use human language with much proficiency. *. Is vision necessary for symbol grounding in AGI? No, but much of human modelling is based on spatial metaphors, and thus the communication issue is particularly salient. *. Is vision important in deciding the meaning of human concepts? Many human concepts are colored with visual connotations, pun intended. You're clearly missing something if you don't have it; but I would guess that with only moderate exceptions, you could capture the essence without it. *. In that case, if an AGI has no vision, how can it still understand a human concept? The same way it can understand anything: it has a model whose semantics match the semantics of the real domain. *. Can a blind person be intelligent? Yes. *. How can a sensorless system like NARS have grounded symbol? Forget grounded. Can it *understand* things? Yes, if it has a model whose semantics match the semantics of the real domain. *. If NARS always uses symbols differently from typical human usage, can we still consider it intelligent? Certainly, if the symbols it uses for communication are close enough to the usages of whoever it's communicating with to be comprehensible. Internally it can use whatever symbols it wants any way it wants. *. Are you saying that vision has nothing to do with AGI? Personally I think that vision is fairly important in a practical sense, because I think we'll get a lot of insights into what's going on in there when we try to unify the higher levels of the visual and natural language interpretive structures. And of course, vision will be of immense practical use in a robot. But I think that once we do know what's going on, it will be possible to build a Turing-test-passing AI without vision. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=53256315-ae7a51
Re: [agi] Conway's Game of Life and Turing machine equivalence
It's probably worth pointing out that Conway's Life is not only Turing universal but that it can host self-replicating machines. In other words, an infinite randomly initialized Life board will contain living creatures which will multiply and grow, and ultimately come to dominate the entire board, as the self-replicating molecules in Earth's primeval oceans gave rise to biological life, which drastically changed the character of the whole planet. In other words, the large-scale character of *any* sufficiently large Life board will be determined by the properties of the self-replicating patterns (which are a rare class (to begin with!), and overlap the Turing-universal ones). It remains to be seen whether replicating Life patterns could evolve to become intelligent. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=50889022-004e1e
Re: [agi] Conway's Game of Life and Turing machine equivalence
I'm not convinced, primarily because I would have said the same thing about actual bacteria vs humans if I didn't have the counterexample. One human generation time is 100,000 bacteria gen times -- and it only takes about 133 generations of bacteria to consume the the entire mass of the earth, if they could. Josh On Sunday 07 October 2007 10:57:41 am, Russell Wallace wrote: On 10/7/07, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: [rest of post and other recent ones agreed with] It remains to be seen whether replicating Life patterns could evolve to become intelligent. No formal proof, but informally: definitely no. Our universe has all sorts of special properties that make intelligence adaptive, that Conway's Life doesn't have. Intelligence would be baggage in that universe; best survivors will be bacterialike fast self-replicators (maybe simpler than bacteria for all I know: it might turn out to be optimal to ditch general assembler capability). - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=50927602-423edb
Re: [agi] Conway's Game of Life and Turing machine equivalence
On Sunday 07 October 2007 01:55:14 pm, Russell Wallace wrote: On 10/7/07, Vladimir Nesov [EMAIL PROTECTED] wrote: That's interesting perspective - it defines a class of series generators (where for example in GoL one element is the whole board on given tick) that generate intelligence through evolution in time-efficient way, and poses a question: what is the simplest instance of this class? If we accept Occam's razor plus some form of anthropic reasoning, we could conjecture that our universe is the simplest instance of this class, since if there were a simpler one we would (with high probability) have found ourselves in that universe rather than this one. (Mental health warning: the above is hopefully-amusing philosophical conjecture only, and should not be confused with science.) This is the same kind of reasoning that leads Bostrom et al to believe that we are probably living in a simulation, which may be turned off at any ti - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=50927912-b9a98e
[agi] How many scientists?
Does anyone know of any decent estimates of how many scientists are working in cog-sci related fields, roughly AI, psychology, and neuroscience? Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=50789647-287dda
Re: [agi] Conway's Game of Life and Turing machine equivalence
On Friday 05 October 2007 12:13:32 pm, Richard Loosemore wrote: Try walking into any physics department in the world and saying Is it okay if most theories are so complicated that they dwarf the size and complexity of the system that they purport to explain? You're conflating a theory and the mathematical mechanism necessary to apply it to actual situations. The theory in Newtonian physics can be specified as the equations F=ma and F=Gm1m2/r^2 (in vector form); but applying them requires a substantial amount of calculation. You can't simply ignore the unusual case of chaotic motion, because the mathematical *reason* the system doesn't have a closed analytic solution is that chaos is possible. In fact, your example is beautiful, in a way. So it turns out to be necessary to resort to approximate methods, to simulations, in order to deal with the MINUSCULE amout of nonlinearity/tangledness that exist in the interactions of the atoms in a small molecule? Well, whoop-dee-do!! Think again, Hammurabi. DFT is a quantum method that searches a space of linear combinations of basis functions to find a description of the electron density field in a molecular system. In other words, the charge of each electron is smeared over space in a pattern that has to satisfy Shrödinger's equation and also be at equilibrium with the force exerted on it by the charge distributions of each other electron. It's approximately like solving the Navier-Stokes equation for each of N different fluid flow problems simultaneously, under the constraint that each volume experienced a pressure field that was a function of the solution of every other one. Given the solution to that system, you're in a position to evaluate the force on each nucleus, whereupon you can either take it one iteration of a molecular dynamics simulation, or one step of a conjugate gradients energy minimization -- and start out all over again with the electrons, which will have shifted, sometimes radically, due to the different forces from the nuclei. Allow me to quote: What you said above was pure, unalloyed bullshit: an exquisite cocktail of complete technical ignorance, patronizing insults and breathtaking arrogance. You did not understand word one... Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=50491496-da7692
[agi] Schemata
On Thursday 04 October 2007 05:19:29 pm, Edward W. Porter wrote: I have no idea how new the idea is. When Schank was talking about scripts ... From the MIT Encyclopedia of the Cognitive Sciences (p729): Schemata are the psychological constructs that are postulated to account for the molar forms of human generic knowledge. The term *frames*, as introduced by Marvin Minsky (1975) is essentially synonymous, except that Minsky used frame as both a psychological construct and a construct in artificial intelligence. *Scripts* are the subclass of schemata that are used to account for generic (stereotyped) sequences of actions (Schank and Abelson 1977). Read on to find that Minsky, having read the work of a 1930s British psychologist Bartlett in the 30s which had languished in obscurity in the meantime, did reintroduce the concept to cog sci in the mid 70s with his frame paper. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=50702864-107b56
Re: [agi] breaking the small hardware mindset
On Wednesday 03 October 2007 09:37:58 pm, Mike Tintner wrote: I disagree also re how much has been done. I don't think AGI - correct me - has solved a single creative problem - e.g. creativity - unprogrammed adaptivity - drawing analogies - visual object recognition - NLP - concepts - creating an emotional system - general learning - embodied/ grounded knowledge - visual/sensory thinking.- every dimension in short of imagination. (Yes, vast creativity has gone into narrow AI, but that's different). Ah, the Lorelei sings so sweetly. That's what happened to AI in the 80's -- it went off chasing human-level performance at specific tasks, which requires a completely different mindset (and something of a different toolset) than solving the general AI problem. To repeat a previous letter, solving particular problems is engineering, but AI needed science. There are, however, several subproblems that may need to be solved to make a general AI work. General learning is surely one of them. I happen to think that analogy-making is another. But there has been a significant amount of basic research done on these areas. 21st century AI, even narrow AI, looks very different from say 80's expert systems. Lots of new techniques that work a lot better. Some of them require big iron, some don't. Research in analogy-making is slow -- I can only think of Gentner and Hofstadter and their groups as major movers. We don't have a solid theory of analogy yet (structure-mapping to the contrary notwithstanding). It's clearly central, and so I don't understand why more people aren't working on it. (btw: anytime you're doing anything that even smells like subgraph isomorphism, big iron is your friend.) One main reason I support the development of AGI as a serious subfield is not that I think any specific approach here is likely to work (even mine), but that there is a willingness to experiment and a tolerance for new and odd-sounding ideas that spells a renaissance of science in AI. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=49680928-5b6fb1
Re: [agi] breaking the small hardware mindset
On Thursday 04 October 2007 10:42:46 am, Mike Tintner wrote: ... I find no general sense of the need for a major paradigm shift. It should be obvious that a successful AGI will transform and revolutionize existing computational paradigms ... I find it difficult to imagine a development that would at the same time revolutionize existing paradigms and yet not require a paradigm shift. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=49818920-208813
Re: [agi] Conway's Game of Life and Turing machine equivalence
On Thursday 04 October 2007 11:06:11 am, Richard Loosemore wrote: As far as we can tell, GoL is an example of that class of system in which we simply never will be able to produce a theory in which we plug in the RULES of GoL, and get out a list of all the patterns in GoL that are interesting. What do you exclude from your notion of a theory? If it can require evaluating a recursive function, or solving a Diophantine equation, or any of the other (provably) Turing equivalent constructs we often use to express scientific theories, then I can readily give you a theory that will take the rules, run huge numbers of experiments, do clustering and maxent type analyses, and so forth, using any definition of interesting you can formally specify. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=49988403-a67391
Re: The first-to-market effect [WAS Re: [agi] Religion-free technical content]
On Thursday 04 October 2007 11:50:21 am, Bob Mottram wrote: To me this seems like elevating that status of nanotech to magic. Even given RSI and the ability of the AGI to manufacture new computing resources it doesn't seem clear to me how this would enable it to prevent other AGIs from also reaching RSI capability. Hear, hear and again I say hear, hear! There's a lot of and then a miracle occurs in step 2 in the we build a friendly AI and it takes over the world and saves our asses type reasoning we see so much of. (Or the somebody builds an unfriendly AI and it takes over the world and wipes us out reasoning as well.) We can't build a system that learns as fast as a 1-year-old just now. Which is our most likely next step: (a) A system that does learn like a 1-year-old, or (b) a system that can learn 1000 times as fast as an adult? Following Moore's law and its software cognates, I'd say give me the former and I'll give you the latter in a decade. With lots of hard work. Then and only then will you have something that's able to improve itself faster than a high-end team of human researchers and developers could. Furthermore, there's a natural plateau waiting for it. That's where it has to leave off learning by absorbing knowledge fom humans (reading textbooks and research papers, etc) and doing the actual science itself. I have heard NO ONE give an argument that puts a serious dent in this, to my way of thinking. Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=50014668-f60c12