[agi] Numenta: article on Jeff Hawkins' AGI approach
Check out this paper... http://www.numenta.com/Numenta_HTM_Concepts.pdf I think it's a good article. It seems to fairly fully reveal the scope and nature of their current scientific activities, though it says nothing about their plans for commercialization or other practical application. What they have is -- a hierarchical representation of patterns-among-patterns-among-patterns, arranged as a tree (or sometimes a directed acyclic graph) -- a Bayes net type belief updating function helping to determine which patterns are present in a given situation -- a scheme for recognizing temporal patterns in data, used at each node in the memory tree, based on a simple greedy learning algorithm that conceptually resembles Hopfield net learning but is implemented differently This is very nice but please note what is not here, for example -- any kind of cognitive architecture for integrating action, perception and cognition toward the achievement of goals in an environment -- any way of pragmatically representing abstract knowledge rather than relatively simple repetitive patterns in data streams -- any way of learning complex coordinated procedures, plans, etc. etc. The theoretical presumption here is that once you've solve the problem of recognizing moderately complex patterns in perceptual data streams, then you're essentially done with the AGI problem and the rest is just some wrappers placed around your perception code. I don't think so I think they are building a nice perceptual pattern recognition module, and waving their hands around arguing that it actually is just an exemplar for an approach that can be more general. I agree that **philosophically** their approach can be extended beyond perception, in the sense that the combination of hierarchy, probabilistic belief propagation, and pattern recognition is critical to all aspects of intelligence. But the particular ways theyve implemented these themes seems to me not to generalize hardly at all beyond the domain of perceptual pattern recognition. (I note that Novamente also embodies all these themes, but in a different way, oriented more toward cognition than perception.) -- Ben --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI bottlenecks
On 01/06/06, Richard Loosemore [EMAIL PROTECTED] wrote: I had similar feelings about William Pearson's recent message about systems that use reinforcement learning: A reinforcement scenario, from wikipedia is defined as Formally, the basic reinforcement learning model consists of: 1. a set of environment states S; 2. a set of actions A; and 3. a set of scalar rewards in the Reals. Here is my standard response to Behaviorism (which is what the above reinforcement learning model actually is): Who decides when the rewards should come, and who chooses what are the relevant states and actions? The rewards I don't deal with, I am interested in external brain add-ons rather than autonomous systems, so the reward system will be closely coupled to a human in some fashion. The rest of post I was trying to outline a system that could alter what it considered actions and states (and bias, learning algorithms etc). The RL definition was just there as an example to work against. If you find out what is doing *that* work, you have found your intelligent system. And it will probably turn out to be so enormously complex, relative to the reinforcement learning part shown above, that the above formalism (assuming it has not been discarded by then) will be almost irrelevant. The internals of the system will be enormously more complex compared to the reinforcement part I described. But it won't make that irrelevent. What goes on inside a PC is vastly more complex than the system that governs the permissions of what each *nix program can do. This doesn't mean the permission governing system is irrelevent. Like the permissions system in *nix the reinforcement system it is only supposed to govern who is allowed to do what, not what actually happens. Unlike the permission system it is supposed to get that from the affect of the programs on the environment. Without it both sorts of systems would be highly unstable. I see it as a necessity for complete modular flexibility. If you get one of the bits that does the work wrong, or wrong for the current environment, how do you allow it to change? Just my deux centimes' worth. Appreciated. On a more positive note, I do think it is possible for AGI researchers to work together within a common formalism. My presentation at the AGIRI workshop was about that, and when I get the paper version of the talk finalized I will post it somewhere. I'll be interested, but sceptical. Will --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
[agi] Universal Test for AI?...... AGI bottlenecks
What is the universal test for the ability of any given AI SYSTEM to Perceive Reason and Act? Is there such a test? What is the closest test known to date? Dan Goe From : William Pearson [EMAIL PROTECTED] To : agi@v2.listbox.com Subject : Re: [agi] AGI bottlenecks Date : Fri, 2 Jun 2006 14:30:20 +0100 On 01/06/06, Richard Loosemore [EMAIL PROTECTED] wrote: I had similar feelings about William Pearson's recent message about systems that use reinforcement learning: A reinforcement scenario, from wikipedia is defined as Formally, the basic reinforcement learning model consists of: 1. a set of environment states S; 2. a set of actions A; and 3. a set of scalar rewards in the Reals. Here is my standard response to Behaviorism (which is what the above reinforcement learning model actually is): Who decides when the rewards should come, and who chooses what are the relevant states and actions? The rewards I don't deal with, I am interested in external brain add-ons rather than autonomous systems, so the reward system will be closely coupled to a human in some fashion. The rest of post I was trying to outline a system that could alter what it considered actions and states (and bias, learning algorithms etc). The RL definition was just there as an example to work against. If you find out what is doing *that* work, you have found your intelligent system. And it will probably turn out to be so enormously complex, relative to the reinforcement learning part shown above, that the above formalism (assuming it has not been discarded by then) will be almost irrelevant. The internals of the system will be enormously more complex compared to the reinforcement part I described. But it won't make that irrelevent. What goes on inside a PC is vastly more complex than the system that governs the permissions of what each *nix program can do. This doesn't mean the permission governing system is irrelevent. Like the permissions system in *nix the reinforcement system it is only supposed to govern who is allowed to do what, not what actually happens. Unlike the permission system it is supposed to get that from the affect of the programs on the environment. Without it both sorts of systems would be highly unstable. I see it as a necessity for complete modular flexibility. If you get one of the bits that does the work wrong, or wrong for the current environment, how do you allow it to change? Just my deux centimes' worth. Appreciated. On a more positive note, I do think it is possible for AGI researchers to work together within a common formalism. My presentation at the AGIRI workshop was about that, and when I get the paper version of the talk finalized I will post it somewhere. I'll be interested, but sceptical. Will --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Numenta: article on Jeff Hawkins' AGI approach
The theoretical presumption here is that once you've solve the problem of recognizing moderately complex patterns in perceptual data streams, then you're essentially done with the AGI problem and the rest is just some wrappers placed around your perception code. I don't think so I think they are building a nice perceptual pattern recognition module, and waving their hands around arguing that it actually is just an exemplar for an approach that can be more general. Some parts of the article definitely overemphasize the potential for perceptual pattern recognition to account for a large number of cognitive processes. But I think that, ultimately, Hawkins et al probably agree with your characterization of perception. For instance, they spend some time discussing the need to hook up an external episodic memory module in order to get more powerful behavior. So surely, from an AGI perspective, they believe that HTM would be just one (albeit important) element in a more complex system. Mike --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
[agi] Other approches to Seed AI? .... Numenta: article on Jeff Hawkins' AGI approach
What are the other methods of approach to Seed AI? Dan Goe From : Mike Ross [EMAIL PROTECTED] To : agi@v2.listbox.com Subject : Re: [agi] Numenta: article on Jeff Hawkins' AGI approach Date : Fri, 2 Jun 2006 10:54:31 -0400 The theoretical presumption here is that once you've solve the problem of recognizing moderately complex patterns in perceptual data streams, then you're essentially done with the AGI problem and the rest is just some wrappers placed around your perception code. I don't think so I think they are building a nice perceptual pattern recognition module, and waving their hands around arguing that it actually is just an exemplar for an approach that can be more general. Some parts of the article definitely overemphasize the potential for perceptual pattern recognition to account for a large number of cognitive processes. But I think that, ultimately, Hawkins et al probably agree with your characterization of perception. For instance, they spend some time discussing the need to hook up an external episodic memory module in order to get more powerful behavior. So surely, from an AGI perspective, they believe that HTM would be just one (albeit important) element in a more complex system. Mike --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Numenta: article on Jeff Hawkins' AGI approach
Hi all,The way I've read Hawkins' and company's work so far is that they view HTM as a cognitive engine that, while perceptually based, would essentially drive other cognitive functions, including behavior. I think you're right that they would agree that these additional cognitive functions would likely need extensions to the architecture. I think that perception has gotten short shrift in AI for a long time, so I'm very hapy to see that they're taking this approach (I am biased, however, being a Master's student under Stan Franklin at the University of Memphis working on -- you guessed it -- the perception module for Stan's LIDA system). -- ScottOn 6/2/06, Mike Ross [EMAIL PROTECTED] wrote: The theoretical presumption here is that once you've solve the problem of recognizing moderately complex patterns in perceptual data streams, then you're essentially done with the AGI problem and the rest is just some wrappers placed around your perception code.I don't think soI think they are building a nice perceptual pattern recognition module, and waving their hands around arguing that it actually is just an exemplar for an approach that can be more general. Some parts of the article definitely overemphasize the potential forperceptual pattern recognition to account for a large number ofcognitive processes.But I think that, ultimately, Hawkins et alprobably agree with your characterization of perception.For instance, they spend some time discussing the need to hook up anexternal episodic memory module in order to get more powerfulbehavior.So surely, from an AGI perspective, they believe that HTMwould be just one (albeit important) element in a more complex system. Mike---To unsubscribe, change your address, or temporarily deactivate your subscription,please go to http://v2.listbox.com/member/[EMAIL PROTECTED] To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Numenta: article on Jeff Hawkins' AGI approach
One of the more interesting ideas the Numenta people have is of how a perceptual system could be used in a motor-control system by hooking up expectations to actual commands. I think its fair to say that Numenta is pushing towards AGI from the animalistic perspective. Once they hook up some memory and tie it in with a control system, it seems they have a good chance of getting something thats about as smart as some dumb animals. To imagine how animals think, I always like to imagine the part of my consciousness that is driving a car while Im driving and having a conversation. The conversation control is the human part of me. The car control is the animal mind. Im guessing that if Numenta makes a lot of progress, they can get that animal mind. But the work described in that paper doesnt seem to have much to do with the human aspect of mind. Mike On 6/2/06, Mike Ross [EMAIL PROTECTED] wrote: The theoretical presumption here is that once you've solve the problem of recognizing moderately complex patterns in perceptual data streams, then you're essentially done with the AGI problem and the rest is just some wrappers placed around your perception code. I don't think so I think they are building a nice perceptual pattern recognition module, and waving their hands around arguing that it actually is just an exemplar for an approach that can be more general. Some parts of the article definitely overemphasize the potential for perceptual pattern recognition to account for a large number of cognitive processes. But I think that, ultimately, Hawkins et al probably agree with your characterization of perception. For instance, they spend some time discussing the need to hook up an external episodic memory module in order to get more powerful behavior. So surely, from an AGI perspective, they believe that HTM would be just one (albeit important) element in a more complex system. Mike --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Numenta: article on Jeff Hawkins' AGI approach
... they have a good chance of getting something thats about as smart as some dumb animalsI agree, Mike, and it seems to me that, from an AGI perspective (as opposed to an AI perspective), this is an excellent goal to have. On 6/2/06, Mike Ross [EMAIL PROTECTED] wrote: One of the more interesting ideas the Numenta people have is of how aperceptual system could be used in a motor-control system by hookingup expectations to actual commands.I think its fair to say that Numenta is pushing towards AGI from the animalistic perspective.Oncethey hook up some memory and tie it in with a control system, it seemsthey have a good chance of getting something thats about as smart as some dumb animals.To imagine how animals think, I always like toimagine the part of my consciousness that is driving a car while Imdriving and having a conversation.The conversation control is thehuman part of me.The car control is the animal mind. Im guessing that if Numenta makes a lot of progress, they can get thatanimal mind.But the work described in that paper doesnt seem to havemuch to do with the human aspect of mind. To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI bottlenecks
Will, Comments taken, but the direction of my critique may have gotten lost in the details: Suppose I proposed a solution to the problem of unifying quantum mechanics and gravity, and suppose I came out with a solution that said that the unified theory involved (a) a specific interface to quantum theory, which I spell out in great detail, and (b) ditto for an interface with geometrodynamics, and (c) a linkage component, to be specified. Physicists would laugh at this. What linkage component?! they would say. And what makes you *believe* that once you sorted out the linkage component, the two interfaces you just specified would play any role whatsoever in that linkage component? They would point out that my linkage component was the meat of the theory, and yet I had referred to in such a way that it seemed as though it was just an extra, to be sorted out later. This is exactly what happened to Behaviorism, and the idea of Reinforcement Learning. The one difference was that they did not explicitly specify an equivalent of my (c) item above: it was for the cognitive psychologists to come along later and point out that Reinforcement Learning implicitly assumed that something in the brain would do the job of deciding when to give rewards, and the job of deciding what the patterns actually were and that that something was the part doing all the real work. In the case of all the experiments in the behaviorist literature, the experimenter substituted for those components, making them less than obvious. Exactly the same critique bears on anyone who suggests that Reinforcement Learning could be the basis for an AGI. I do not believe there is still any reply to that critique. Richard Loosemore William Pearson wrote: On 01/06/06, Richard Loosemore [EMAIL PROTECTED] wrote: I had similar feelings about William Pearson's recent message about systems that use reinforcement learning: A reinforcement scenario, from wikipedia is defined as Formally, the basic reinforcement learning model consists of: 1. a set of environment states S; 2. a set of actions A; and 3. a set of scalar rewards in the Reals. Here is my standard response to Behaviorism (which is what the above reinforcement learning model actually is): Who decides when the rewards should come, and who chooses what are the relevant states and actions? The rewards I don't deal with, I am interested in external brain add-ons rather than autonomous systems, so the reward system will be closely coupled to a human in some fashion. The rest of post I was trying to outline a system that could alter what it considered actions and states (and bias, learning algorithms etc). The RL definition was just there as an example to work against. If you find out what is doing *that* work, you have found your intelligent system. And it will probably turn out to be so enormously complex, relative to the reinforcement learning part shown above, that the above formalism (assuming it has not been discarded by then) will be almost irrelevant. The internals of the system will be enormously more complex compared to the reinforcement part I described. But it won't make that irrelevent. What goes on inside a PC is vastly more complex than the system that governs the permissions of what each *nix program can do. This doesn't mean the permission governing system is irrelevent. Like the permissions system in *nix the reinforcement system it is only supposed to govern who is allowed to do what, not what actually happens. Unlike the permission system it is supposed to get that from the affect of the programs on the environment. Without it both sorts of systems would be highly unstable. I see it as a necessity for complete modular flexibility. If you get one of the bits that does the work wrong, or wrong for the current environment, how do you allow it to change? Just my deux centimes' worth. Appreciated. On a more positive note, I do think it is possible for AGI researchers to work together within a common formalism. My presentation at the AGIRI workshop was about that, and when I get the paper version of the talk finalized I will post it somewhere. I'll be interested, but sceptical. Will --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Numenta: article on Jeff Hawkins' AGI approach
On 6/2/06, Ben Goertzel wrote: Mike You note that Numenta's approach seems oriented toward implementing an animal-level mind... I agree, and I do think this is a fascinating project, and an approach that can ultimately succeed... but I think that for it to succeed Hawkins will have to introduce a LOT of deep concepts that he is currently ignoring in his approach. Most critically he ignores the complex, chaotic dynamics of brain systems... I suppose part of the motivation for starting with animal mind is that the human mind is just a minor adjustment to the animal mind, which is sorta true genetically and evolutionarily But on the other hand, just because animal brains evolved into human brains, doesn't mean that every system with animal-brain functionality has similar evolve-into-human-brain potentiality snip Just from a computer systems design perspective, I think this project is admirable. I think it is safe to claim that all the big computer design disasters occurred because they tried to do too much all at once. ''We want it all, and we want it now!'. Ben may be correct in claiming that major elements are being omitted, but if they even get an animal level intelligence running, this will be a remarkable achievement. They will be world leaders and will learn a lot about designing such systems. Even if it cannot progress to higher levels of intelligence, the experience gained will set their technicians well on the road to the next generation design. BillK --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] procedural vs declarative knowledge
On 6/2/06, Charles D Hixson [EMAIL PROTECTED] wrote: Rule of thumb:First get it working, doing what you want.Thenoptimize.When optimizing, first check your algorithms,then check tosee where time is actually spent.Apply extensive optimization only to the most used 10% (or less) of the code.If you need to optimize morethan that, then you need to either redesign from the base, or get afaster machine.Expect that you will need to redesign pieces so often while in development that it's better to chose the form of code that's easiest tounderstand, redesign, and fix than to optimize it.Only whendevelopment is essentially complete is it time to give optimization forspeed or size serious consideration. That said, do you agree that some applications call for a 'ground up' build mentality? For example, adding security after an application is nearly finished is usually a terrible approach (despite being incredibly common) To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
[agi] Limits to Size and Resources...Procedural vs declarative knowledge
The question remains of the limits of MB/GB size and resource requirements. Execution time for a given process There are limits... Does anyone have any idea of the size of Executable code of fully developed AI System versus a seed AI system? Dan Goe From : Mike Dougherty [EMAIL PROTECTED] To : agi@v2.listbox.com Subject : Re: [agi] procedural vs declarative knowledge Date : Fri, 2 Jun 2006 15:51:34 -0400 On 6/2/06, Charles D Hixson [EMAIL PROTECTED] wrote: Rule of thumb: First get it working, doing what you want. Then optimize. When optimizing, first check your algorithms, then check to see where time is actually spent. Apply extensive optimization only to the most used 10% (or less) of the code. If you need to optimize more than that, then you need to either redesign from the base, or get a faster machine. Expect that you will need to redesign pieces so often while in development that it's better to chose the form of code that's easiest to understand, redesign, and fix than to optimize it. Only when development is essentially complete is it time to give optimization for speed or size serious consideration. That said, do you agree that some applications call for a 'ground up' build mentality? For example, adding security after an application is nearly finished is usually a terrible approach (despite being incredibly common) --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] procedural vs declarative knowledge
Mike Dougherty wrote: On 6/2/06, *Charles D Hixson* [EMAIL PROTECTED] mailto:[EMAIL PROTECTED] wrote: Rule of thumb: First get it working, doing what you want. Then optimize. When optimizing, first check your algorithms, then check to see where time is actually spent. Apply extensive optimization only to the most used 10% (or less) of the code. If you need to optimize more than that, then you need to either redesign from the base, or get a faster machine. Expect that you will need to redesign pieces so often while in development that it's better to chose the form of code that's easiest to understand, redesign, and fix than to optimize it. Only when development is essentially complete is it time to give optimization for speed or size serious consideration. That said, do you agree that some applications call for a 'ground up' build mentality? For example, adding security after an application is nearly finished is usually a terrible approach (despite being incredibly common) That's not a ground up build. That's getting the design right before you commit to anything final. I do agree that getting the security right should be done early. I would assert that it needs to be designed into the system rather than being patched on. The problem is that right is incredibly picky here. You need it not to be so cumbersome that the system is overwhelming either to compute or to use. And for any general purpose system it needs to be adaptable to a wide variety of circumstances, when you won't necessarily be able to predict in advance, e.g., what peripherals are available. (E.g.: you can't depend on visual recognition in low light environments, or if there might not be cameras available.) --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]