RE: [agi] SyNAPSE might not be a joke ---- was ---- Building a machine that can learn from experience
===Colin said== The tacit assumption is that the model's thus implemented on a computer will/can 'behave' indistinguishably from the real thing, when what you are observing is a model of the real thing, not the real thing. ===ED's reply=== I was making no assumption that the model would be behave indistinquisably from the real thing, but instead only that there were meaningful --- and, from a cross-fertilization standpoint, informative --- levels of description at which the computer model and the corresponding brain behavior were similar. ===Colin said== There's a boundary to cross - when you claim to have access to human level intellect - then you are demanding a equivalence with a real human, not a model of a human. ===ED's reply=== When I, and presumably many other AGIers, say human-level AGI, we do not mean an exact functional replica of the human brain or mind. Rather we mean an AGI that can do things like speak and understand natural language, see and understand the meaning of its visual surroundings, reason from the rough equivalent of human-level world knowledge, have common sense, do creative problems solving, and other mental tasks --- substantially as well as most people. Its methods of computation do not have to be exactly like those used in the mind, the major issue is that its competencies be at least roughly as good over a range of talents. ===Colin said== I don't think there's any real issue here. Mostly semantics being mixed a bit. Gotta get back to xmas! Yuletide stuff to you. ===ED's reply=== Agreed. Ed Porter -Original Message- From: Colin Hales [mailto:c.ha...@pgrad.unimelb.edu.au] Sent: Tuesday, December 23, 2008 7:55 PM To: agi@v2.listbox.com Subject: Re: [agi] SyNAPSE might not be a joke was Building a machine that can learn from experience Ed, Comments interspersed below: Ed Porter wrote: Colin, Here are my comments re the following parts of your below post: ===Colin said== I merely point out that there are fundamental limits as to how computer science (CS) can inform/validate basic/physical science - (in an AGI context, brain science). Take the Baars/Franklin IDA project.. It predicts nothing neuroscience can poke a stick at. ===ED's reply=== Different AGI models can have different degrees of correspondence to, and different explanatory relevance to, what is believed to take place in the brain. For example the Thomas Serre's PhD thesis Learning a Dictionary of Shape-Components in Visual Cortex: Comparison with Neurons, Humans and Machines, at from http://cbcl.mit.edu/projects/cbcl/publications/ps/MIT-CSAIL-TR-2006-028.pdf http://cbcl.mit.edu/projects/cbcl/publications/ps/MIT-CSAIL-TR-2006-028.pdf , is a computer simulation which is rather similar to my concept of how a Novamente-like AGI could perform certain tasks in visual perception, and yet it is designed to model the human visual system to a considerable degree. It shows that a certain model of how Serre and Poggio think a certain aspect of the human brain works, does in fact work surprisingly well when simulated in a computer. A surprisingly large number of brain science papers are based on computer simulations, many of which are substantially simplified models, but they do given neuroscientists a way to poke a stick at various theories they might have for how the brain operates at various levels of organization. Some of these papers are directly relevant to AGI. And some AGI papers are directly relevant to providing answers to certain brain science questions. You are quite right! Realistic models can be quite informative and feed back - suggesting new empirical approaches. There can be great cross-fertilisation. However the point is irrelevant to the discussion at hand. The phrase does in fact work surprisingly well when simulated in a computer illustrates the confusion. 'work'? according to whom? surprisingly well? by what criteria? The tacit assumption is that the model's thus implemented on a computer will/can 'behave' indistinguishably from the real thing, when what you are observing is a model of the real thing, not the real thing. HERE If you targeting AGI with a benchmark/target of human intellect or problem solving skills, then the claim made on any/all models is that models can attain that goal. A computer implements a model. To make a claim that a model completely captures the reality upon which it was based, you need to have a solid theory of the relationships between models and reality that is not wishful thinking or assumption, but solid science. Here's where you run into the problematic issue that basic physical sciences have with models. There's a boundary to cross - when you claim to have access to human level intellect - then you are demanding a equivalence with a real human, not a model of a human. ===Colin said== I agree with your : At the other end of things,
Re: [agi] SyNAPSE might not be a joke ---- was ---- Building a machine that can learn from experience
Why is it that people who repeatedly resort to personal abuse like this are still allowed to participate in the discussion on the AGI list? Richard Loosemore Ed Porter wrote: Richard, You originally totally trashed Tononi's paper, including its central core, by saying: It is, for want of a better word, nonsense. And since people take me to task for being so dismissive, let me add that it is the central thesis of the paper that is nonsense: if you ask yourself very carefully what it is he is claiming, you can easily come up with counterexammples that make a mockery of his conclusion.\ When asked to support your statement that you can easily come up with counterexammples that make a mockery of his conclusion you refused. You did so by grossly mis-describing Tononi’s paper (for example it does not include “pages of …math”, of any sort, and particularly not “pages of irrelevant math”) and implying its mis-described faults so offended your delicate sense of AGI propriety that re-reading it enough to find support for your extremely critical (and perhaps totally unfair) condemnation would be either too much work or too emotionally painful. You said the counterexamples to the core of this paper were easy to come up with, but you can’t seem to come up with any. Such stunts have the appearance of being those of a pompous windbag. Ed Porter P.S. Your postscript is not sufficiently clear to provide much support for your position. P.P.S. You below -Original Message- From: Richard Loosemore [mailto:r...@lightlink.com] Sent: Tuesday, December 23, 2008 9:53 AM To: agi@v2.listbox.com Subject: Re: [agi] SyNAPSE might not be a joke was Building a machine that can learn from experience Ed Porter wrote: Richard, Please describe some of the counterexamples, that you can easily come up with, that make a mockery of Tononi's conclusion. Ed Porter Alas, I will have to disappoint. I put a lot of effort into understanding his paper first time around, but the sheer agony of reading (/listening to) his confused, shambling train of thought, the non-sequiteurs, and the pages of irrelevant math that I do not need to experience a second time. All of my original effort only resulted in the discovery that I had wasted my time, so I have no interest in wasting more of my time. With other papers that contain more coherent substance, but perhaps what looks like an error, I would make the effort. But not this one. It will have to be left as an exercise for the reader, I'm afraid. Richard Loosemore P.S. A hint. All I remember was that he started talking about multiple regions (columns?) of the brain exchanging information with one another in a particular way, and then he asserted a conclusion which, on quick reflection, I knew would not be true of a system resembling the distributed one that I described in my consciousness paper (the molecular model). Knowing that his conclusion was flat-out untrue for that one case, and for a whole class of similar systems, his argument was toast. -Original Message- From: Richard Loosemore [mailto:r...@lightlink.com] Sent: Monday, December 22, 2008 8:54 AM To: agi@v2.listbox.com Subject: Re: [agi] SyNAPSE might not be a joke was Building a machine that can learn from experience Ed Porter wrote: I don't think this AGI list should be so quick to dismiss a $4.9 million dollar grant to create an AGI. It will not necessarily be vaporware. I think we should view it as a good sign. Even if it is for a project that runs the risk, like many DARPA projects (like most scientific funding in general) of not necessarily placing its money where it might do the most good --- it is likely to at least produce some interesting results --- and it just might make some very important advances in our field. The article from http://www.physorg.com/news148754667.html said: .a $4.9 million grant.for the first phase of DARPA's Systems of Neuromorphic Adaptive Plastic Scalable Electronics (SyNAPSE) project. Tononi and scientists from Columbia University and IBM will work on the software for the thinking computer, while nanotechnology and supercomputing experts from Cornell, Stanford and the University of California-Merced will create the hardware. Dharmendra Modha of IBM is the principal investigator. The idea is to create a computer capable of sorting through multiple streams of changing data, to look for patterns and make logical decisions. There's another requirement: The finished cognitive computer should be as small as a the brain of a small mammal and use as little power as a 100-watt light bulb. It's a major
[agi] Levels of Self-Awareness?
This is more of a question than a statement. There appears to be several levels of self-awareness, e.g. 1. Knowing that you are an individual in a group, have a name, etc. Even kittens and puppies quickly learn their names, know to watch others when their names are called, etc. 2. Understanding that they have some (limited) ability to modify their own behavior, reactions, etc., so that you can explain to them how something they did was inappropriate, and they can then modify their behavior. Can filter what they say, etc. I know one lady with two Masters degree who apparently has NOT reached this level. 3. Understanding that the process of thinking itself is a skill that no one has completely mastered, that there are advanced techniques to be learned, that there are probably as-yet undiscovered techniques for really advanced capabilities, etc. Further, are capable of internalizing new thinking techniques. There appears to be several people on this list who have apparently NOT reached this level. 4. Any theories as to what the next level might be? Note that the above relates to soul, especially in that an individual at a higher level might look upon individuals at a lower level as soulless creatures. Given that various people span several levels, wouldn't this consign much of the human race as being soulless creatures? Clearly, it would seem that no AGI researcher can program a level of self-awareness that they themselves have not reached, tried and failed to reach, etc. Hence, this may impose a cap on a future AGI's potential abilities, especially if thegold is in #4, #5, etc. Has someone already looked into this? Steve Richfield --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Levels of Self-Awareness?
2008/12/24 Steve Richfield steve.richfi...@gmail.com: Clearly, it would seem that no AGI researcher can program a level of self-awareness that they themselves have not reached, tried and failed to reach, etc. This is not at all clear to me. It is certainly prossible for programmers to program computer to do tasks better than they can (e.g. play chess) and I see no reason why it shouldn't be possible for self awareness. Indeed it would be rather trivial to give an AGI access to its source code. -- Philip Hunt, cabala...@googlemail.com Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Levels of Self-Awareness?
Philip, On 12/24/08, Philip Hunt cabala...@googlemail.com wrote: 2008/12/24 Steve Richfield steve.richfi...@gmail.com: Clearly, it would seem that no AGI researcher can program a level of self-awareness that they themselves have not reached, tried and failed to reach, etc. This is not at all clear to me. It is certainly prossible for programmers to program computer to do tasks better than they can (e.g. play chess) Yes, but these programmers already know how to play chess. They (probably) can't program a game in which they themselves don't have any skill at all. In the case of higher forms of self-awareness, programmers in effect don't even know the rules of the game to be programmed, yet the game will have a vast overall effect on everything the AGI thinks. To illustrate, much human thought goes into dispute resolution - a field rich with advanced concepts that are generally unknown to the general population and AGI programmers. Since this has to much to do with the subtleties of common errors in human thinking, there is no practical way for an AGI to figure this out for itself short of participating in thousands of disputes - that humans would simply not tolerate. Once these concepts are understood, the very act of thinking is changed forever. Someone who is highly trained and experienced in dispute resolution thinks quite differently than you probably do, and probably regards your thinking as immature and generally low-level. In short, their idea of self-awareness is quite different than yours. Regardless of tools, I don't see how such a thing could be programmed except by someone who is already able to think at that level. Then, how about the NEXT level, whatever that might be? and I see no reason why it shouldn't be possible for self awareness. My point is that lower-level self-awareness is MUCH simpler to contemplate than is higher-level, and further, that different people (and AGI researchers) function at various levels. Indeed it would be rather trivial to give an AGI access to its source code. Why should it be any better at modifying its source code than we would be at writing it? The problem of levels still remains. Steve Richfield --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
[agi] Introducing Steve's Theory of Everything in cognition.
Ben, et al, After ~5 months of delay for theoretical work, here are the basic ideas as to how really fast and efficient automatic learning could be made almost trivial. I decided NOT to post the paper (yet), but rather, to just discuss the some of the underlying ideas in AGI-friendly terms. Suppose for a moment that a NN or AGI program (they can be easily mapped from one form to the other), instead of operating on objects (in an object-oriented sense), instead, operates on the rate-of-changes in the probabilities of objects, or dp/dt. Presuming sufficient bandwidth to generally avoid superstitious coincidences, fast unsupervised learning then becomes completely trivial, as like objects cause simultaneous like-patterned changes in the inputs WITHOUT the overlapping effects of the many other objects typically present in the input (with numerous minor exceptions). But, what would Bayesian equations or NN neuron functionality look like in dp/dt space? NO DIFFERENCE (math upon request). You could trivially differentiate the inputs to a vast and complex existing AGI or NN, integrate the outputs, and it would perform *identically* (except for some little details discussed below). Of course, while the transforms would be identical, unsupervised learning would be quite a different matter, as now the nearly-impossible becomes trivially simple. For some things (like short-term memory) you NEED an integrated object-oriented result. Very simple - just integrate the signal. How about muscle movements? Note that muscle actuation typically causes acceleration, which doubly integrates the driving signal, which would require yet another differentiation of a differentiated signal to, when doubly integrated by the mechanical system, produce movement to the desired location. Note that once input values are stored in a matrix for processing, the baby has already been thrown out with the bathwater. You must START with differentiated input values and NOT static measured values. THIS is what the PCA folks have been missing in their century-long quest for an efficient algorithm to identify principal components, as their arrays had already discarded exactly what they needed. Of course you could simply subtract successive samples from one another - at some considerable risk, since you are now sampling at only half the Nyquist-required speed to make your AGI/NN run at its intended speed. In short, if inputs are not being electronically differentiated, then sampling must proceed at least twice as fast as the NN/AGI cycles. But - how about the countless lost constants of integration? They all come out in the wash - except for where actual integration at the outputs is needed. Then, clippers and leaky integrators, techniques common to electrical engineering, will work fine and produce many of the same artifacts (like visual extinction) seen in natural systems. It all sounds SO simple, but I couldn't find any prior work in this direction using Google. However, the collective memory of this group is pretty good, so perhaps someone here knows of some prior effort that did something like this. I would sure like to put SOMETHING in the References section of my paper. Loosemore: THIS is what I was talking about when I explained that there is absolutely NO WAY to understand a complex system through direct observation, except by its useless anomalies. By shifting an entire AGI or NN to operate on derivatives instead of object values, it works *almost* (the operative word in this statement) exactly the same as one working in object-oriented space, only learning is transformed from the nearly-impossible to the trivially simple. Do YOU see any observation-based way to tell how we are operating behind our eyeballs, object-oriented or dp/dt? While there are certainly other explanations for visual extinction, this is the only one that I know of that is absolutely impossible to engineer around. No one has (yet) proposed any value to visual extinction, and it is a real problem for hunters, so if it were avoidable, then I suspect that ~200 million years of evolution would have eliminated it long ago. From this comes numerous interesting corollaries. Once the dp/dt signals are in array form, it would become simple to automatically recognize patterns representing complex phenomena at the level of the neurons/equations in question. Of course, putting it in this array form is effectively a transformation from AGI equations to NN construction, a transformation that has been discussed in prior postings. In short, if you want your AGI to learn at anything approaching biological speeds, it appears that you absolutely MUST transform your AGI structure to a NN-like representation, regardless of the structure of the processor on which it runs. Unless I am missing something really important here, this should COMPLETELY transform the AGI field, regardless of the particular approach taken. Any thoughts? Steve Richfield
Re: [agi] Introducing Steve's Theory of Everything in cognition.
On Thu, Dec 25, 2008 at 9:33 AM, Steve Richfield steve.richfi...@gmail.com wrote: Any thoughts? I can't tell this note from nonsense. You need to work on presentation, if your idea can actually hold some water. If you think you understand the idea enough to express it as math, by all means do so, it'll make your own thinking clearer if nothing else. -- Vladimir Nesov robot...@gmail.com http://causalityrelay.wordpress.com/ --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com
Re: [agi] Introducing Steve's Theory of Everything in cognition.
On Dec 24, 2008, at 10:33 PM, Steve Richfield wrote: Of course you could simply subtract successive samples from one another - at some considerable risk, since you are now sampling at only half the Nyquist-required speed to make your AGI/NN run at its intended speed. In short, if inputs are not being electronically differentiated, then sampling must proceed at least twice as fast as the NN/AGI cycles. Or... you could be using something like compressive sampling, which safely ignores silly things like the Nyquist limit. Cheers, J. Andrew Rogers --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=123753653-47f84b Powered by Listbox: http://www.listbox.com