Re: [agi] An alternative plan to discover self-organization theory
Matt:It is like the way evolution works, except that there is a human in the loop to make the process a little more intelligent. IOW this is like AGI, except that it's narrow AI. That's the whole point - you have to remove the human from the loop. In fact, it also sounds like a misconceived and rather literal idea of evolution as opposed to the reality. From: Matt Mahoney Sent: Monday, June 21, 2010 3:01 AM To: agi Subject: Re: [agi] An alternative plan to discover self-organization theory Steve Richfield wrote: He suggested that I construct a simple NN that couldn't work without self organizing, and make dozens/hundreds of different neuron and synapse operational characteristics selectable ala genetic programming, put it on the fastest computer I could get my hands on, turn it loose trying arbitrary combinations of characteristics, and see what the winning combination turns out to be. Then, armed with that knowledge, refine the genetic characteristics and do it again, and iterate until it efficiently self organizes. This might go on for months, but self-organization theory might just emerge from such an effort. Well, that is the process that created human intelligence, no? But months? It actually took 3 billion years on a planet sized molecular computer. That doesn't mean it won't work. It just means you have to narrow your search space and lower your goals. I can give you an example of a similar process. Look at the code for PAQ8HP12ANY and LPAQ9M data compressors by Alexander Ratushnyak, which are the basis of winning Hutter prize submissions. The basic principle is that you have a model that receives a stream of bits from an unknown source and it uses a complex hierarchy of models to predict the next bit. It is sort of like a neural network because it averages together the results of lots of adaptive pattern recognizers by processes that are themselves adaptive. But I would describe the code as inscrutable, kind of like your DNA. There are lots of parameters to tweak, such as how to preprocess the data, arrange the dictionary, compute various contexts, arrange the order of prediction flows, adjust various learning rates and storage capacities, and make various tradeoffs sacrificing compression to meet memory and speed requirements. It is simple to describe the process of writing the code. You make random changes and keep the ones that work. It is like the way evolution works, except that there is a human in the loop to make the process a little more intelligent. There are also fully automated optimizers for compression algorithms, but they are more limited in their search space. For example, the experimental PPM based EPM by Serge Osnach includes a program EPMOPT that adjusts 20 numeric parameters up or down using a hill climbing search to find the best compression. It can be very slow. Another program, M1X2 by Christopher Mattern, uses a context mixing (PAQ like) algorithm in which the contexts are selected by using a hill climbing genetic algorithm to select a set of 64-bit masks. One version was run for 3 days to find the best options to compress a file that normally takes 45 seconds. -- Matt Mahoney, matmaho...@yahoo.com From: Steve Richfield steve.richfi...@gmail.com To: agi agi@v2.listbox.com Sent: Sun, June 20, 2010 2:06:55 AM Subject: [agi] An alternative plan to discover self-organization theory No, I haven't been smokin' any wacky tobacy. Instead, I was having a long talk with my son Eddie, about self-organization theory. This is his proposal: He suggested that I construct a simple NN that couldn't work without self organizing, and make dozens/hundreds of different neuron and synapse operational characteristics selectable ala genetic programming, put it on the fastest computer I could get my hands on, turn it loose trying arbitrary combinations of characteristics, and see what the winning combination turns out to be. Then, armed with that knowledge, refine the genetic characteristics and do it again, and iterate until it efficiently self organizes. This might go on for months, but self-organization theory might just emerge from such an effort. I had a bunch of objections to his approach, e.g. Q. What if it needs something REALLY strange to work? A. Who better than you to come up with a long list of really strange functionality? Q. There are at least hundreds of bits in the genome. A. Try combinations in pseudo-random order, with each bit getting asserted in ~half of the tests. If/when you stumble onto a combination that sort of works, switch to varying the bits one-at-a-time, and iterate in this way until the best combination is found. Q. Where are we if this just burns electricity for a few months and finds nothing? A. Print out the best combination, break out the wacky tobacy, and come up with even better/crazier
Re: [agi] An alternative plan to discover self-organization theory
Mike Tintner wrote: Matt:It is like the way evolution works, except that there is a human in the loop to make the process a little more intelligent. IOW this is like AGI, except that it's narrow AI. That's the whole point - you have to remove the human from the loop. In fact, it also sounds like a misconceived and rather literal idea of evolution as opposed to the reality. You're right. It is narrow AI. You keep pointing out that we haven't solved the general problem. You are absolutely correct. So, do you have any constructive ideas on how to solve it? Preferably something that takes less than 3 billion years on a planet sized molecular computer. -- Matt Mahoney, matmaho...@yahoo.com From: Mike Tintner tint...@blueyonder.co.uk To: agi agi@v2.listbox.com Sent: Mon, June 21, 2010 7:59:29 AM Subject: Re: [agi] An alternative plan to discover self-organization theory Matt:It is like the way evolution works, except that there is a human in the loop to make the process a little more intelligent. IOW this is like AGI, except that it's narrow AI. That's the whole point - you have to remove the human from the loop. In fact, it also sounds like a misconceived and rather literal idea of evolution as opposed to the reality. From: Matt Mahoney Sent: Monday, June 21, 2010 3:01 AM To: agi Subject: Re: [agi] An alternative plan to discover self-organization theory Steve Richfield wrote: He suggested that I construct a simple NN that couldn't work without self organizing, and make dozens/hundreds of different neuron and synapse operational characteristics selectable ala genetic programming, put it on the fastest computer I could get my hands on, turn it loose trying arbitrary combinations of characteristics, and see what the winning combination turns out to be. Then, armed with that knowledge, refine the genetic characteristics and do it again, and iterate until it efficiently self organizes. This might go on for months, but self-organization theory might just emerge from such an effort. Well, that is the process that created human intelligence, no? But months? It actually took 3 billion years on a planet sized molecular computer. That doesn't mean it won't work. It just means you have to narrow your search space and lower your goals. I can give you an example of a similar process. Look at the code for PAQ8HP12ANY and LPAQ9M data compressors by Alexander Ratushnyak, which are the basis of winning Hutter prize submissions. The basic principle is that you have a model that receives a stream of bits from an unknown source and it uses a complex hierarchy of models to predict the next bit. It is sort of like a neural network because it averages together the results of lots of adaptive pattern recognizers by processes that are themselves adaptive. But I would describe the code as inscrutable, kind of like your DNA. There are lots of parameters to tweak, such as how to preprocess the data, arrange the dictionary, compute various contexts, arrange the order of prediction flows, adjust various learning rates and storage capacities, and make various tradeoffs sacrificing compression to meet memory and speed requirements. It is simple to describe the process of writing the code. You make random changes and keep the ones that work. It is like the way evolution works, except that there is a human in the loop to make the process a little more intelligent. There are also fully automated optimizers for compression algorithms, but they are more limited in their search space. For example, the experimental PPM based EPM by Serge Osnach includes a program EPMOPT that adjusts 20 numeric parameters up or down using a hill climbing search to find the best compression. It can be very slow. Another program, M1X2 by Christopher Mattern, uses a context mixing (PAQ like) algorithm in which the contexts are selected by using a hill climbing genetic algorithm to select a set of 64-bit masks. One version was run for 3 days to find the best options to compress a file that normally takes 45 seconds. -- Matt Mahoney, matmaho...@yahoo.com From: Steve Richfield steve.richfi...@gmail.com To: agi agi@v2.listbox.com Sent: Sun, June 20, 2010 2:06:55 AM Subject: [agi] An alternative plan to discover self-organization theory No, I haven't been smokin' any wacky tobacy. Instead, I was having a long talk with my son Eddie, about self-organization theory. This is his proposal: He suggested that I construct a simple NN that couldn't work without self organizing, and make dozens/hundreds of different neuron and synapse operational characteristics selectable ala genetic programming, put it on the fastest computer I could get my hands on, turn it loose trying arbitrary combinations of characteristics, and see what the winning combination turns out to be. Then, armed with that knowledge, refine the
[agi] Re: High Frame Rates Reduce Uncertainty
Ignoring Steve because we are simply going to have to agree to disagree... And I don't see enough value in trying to understand his paper. I said the math was overly complex, but what I really meant is that the approach is overly complex and so filled with research specific jargon, I don't care to try understand it. It is overly converned with copying the way that the brain does things. I don't care how the brain does it. I care about why the brain does it. Its the same as the analogy of giving a man a fish or teaching him to fish. You may figure out how the brain works, but it does you little good if you don't understand why it works that way. You would have to create a synthetic brain to take advantage of the knowledge, which is not a approach to AGI for many reasons. There are a million other ways, even better ways, to do it than the way the brain does it. Just because the brain accidentally found 1 way out of a million to do it doesn't make it the right way for us to develop AGI. So, moving on I can't find references online, but I've read that the Air Force studied the ability of the human eye to identify aircraft in images that were flashed on a screen at 1/220th of a second. So, clearly, the human eye can at least distinguish 220 fps if it operated that way. Of course, it may not operate on fps second, but that is besides the point. I've also heard other people say that a study has shown that the human eye takes 1000 exposures per second. They had no references though, so it is hearsay. The point was that the brain takes advantage of the fact that with such a high exposure rate, the changes between each image are very small if the objects are moving. This allows it to distinguish movement and visual changes with extremely low uncertainty. If it detects that the changes required to match two parts of an image are too high or the distance between matches is too far, it can reject a match. This allows it to distinguish only very low uncertainty changes and reject changes that have high uncertainty. I think this is a very significant discovery regarding how the brain is able to learn in such an ambiguous world with so many variables that are difficult to disambiguate, interpret and understand. Dave On Fri, Jun 18, 2010 at 2:19 PM, David Jones davidher...@gmail.com wrote: I just came up with an awesome idea. I just realized that the brain takes advantage of high frame rates to reduce uncertainty when it is estimating motion. The slower the frame rate, the more uncertainty there is because objects may have traveled too far between images to match with high certainty using simple techniques. So, this made me think, what if the secret to the brain's ability to learn generally stems from this high frame rate trick. What if we made a system that could process even high frame rates than the brain can. By doing this you can reduce the uncertainty of matches very very low (well in my theory so far). If you can do that, then you can learn about the objects in a video, how they move together or separately with very high certainty. You see, matching is the main barrier when learning about objects. But with a very high frame rate, we can use a fast algorithm and could potentially reduce the uncertainty to almost nothing. Once we learn about objects, matching gets easier because now we have training data and experience to take advantage of. In addition, you can also gain knowledge about lighting, color variation, noise, etc. With that knowledge, you can then automatically create a model of the object with extremely high confidence. You will also be able to determine the effects of light and noise on the object's appearance, which will help match the object invariantly in the future. It allows you to determine what is expected and unexpected for the object's appearance with much higher confidence. Pretty cool idea huh? Dave --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] Re: High Frame Rates Reduce Uncertainty
Your computer monitor flashes 75 frames per second, but you don't notice any flicker because light sensing neurons have a response delay of about 100 ms. Motion detection begins in the retina by cells that respond to contrast between light and dark moving in specific directions computed by simple, fixed weight circuits. Higher up in the processing chain, you detect motion when your eyes and head smoothly track moving objects using kinesthetic feedback from your eye and neck muscles and input from your built in accelerometer in the semicircular canals in your ears. This is all very complicated of course. You are more likely to detect motion in objects that you recognize and expect to move, like people, animals, cars, etc. -- Matt Mahoney, matmaho...@yahoo.com From: David Jones davidher...@gmail.com To: agi agi@v2.listbox.com Sent: Mon, June 21, 2010 9:39:30 AM Subject: [agi] Re: High Frame Rates Reduce Uncertainty Ignoring Steve because we are simply going to have to agree to disagree... And I don't see enough value in trying to understand his paper. I said the math was overly complex, but what I really meant is that the approach is overly complex and so filled with research specific jargon, I don't care to try understand it. It is overly converned with copying the way that the brain does things. I don't care how the brain does it. I care about why the brain does it. Its the same as the analogy of giving a man a fish or teaching him to fish. You may figure out how the brain works, but it does you little good if you don't understand why it works that way. You would have to create a synthetic brain to take advantage of the knowledge, which is not a approach to AGI for many reasons. There are a million other ways, even better ways, to do it than the way the brain does it. Just because the brain accidentally found 1 way out of a million to do it doesn't make it the right way for us to develop AGI. So, moving on I can't find references online, but I've read that the Air Force studied the ability of the human eye to identify aircraft in images that were flashed on a screen at 1/220th of a second. So, clearly, the human eye can at least distinguish 220 fps if it operated that way. Of course, it may not operate on fps second, but that is besides the point. I've also heard other people say that a study has shown that the human eye takes 1000 exposures per second. They had no references though, so it is hearsay. The point was that the brain takes advantage of the fact that with such a high exposure rate, the changes between each image are very small if the objects are moving. This allows it to distinguish movement and visual changes with extremely low uncertainty. If it detects that the changes required to match two parts of an image are too high or the distance between matches is too far, it can reject a match. This allows it to distinguish only very low uncertainty changes and reject changes that have high uncertainty. I think this is a very significant discovery regarding how the brain is able to learn in such an ambiguous world with so many variables that are difficult to disambiguate, interpret and understand. Dave On Fri, Jun 18, 2010 at 2:19 PM, David Jones davidher...@gmail.com wrote: I just came up with an awesome idea. I just realized that the brain takes advantage of high frame rates to reduce uncertainty when it is estimating motion. The slower the frame rate, the more uncertainty there is because objects may have traveled too far between images to match with high certainty using simple techniques. So, this made me think, what if the secret to the brain's ability to learn generally stems from this high frame rate trick. What if we made a system that could process even high frame rates than the brain can. By doing this you can reduce the uncertainty of matches very very low (well in my theory so far). If you can do that, then you can learn about the objects in a video, how they move together or separately with very high certainty. You see, matching is the main barrier when learning about objects. But with a very high frame rate, we can use a fast algorithm and could potentially reduce the uncertainty to almost nothing. Once we learn about objects, matching gets easier because now we have training data and experience to take advantage of. In addition, you can also gain knowledge about lighting, color variation, noise, etc. With that knowledge, you can then automatically create a model of the object with extremely high confidence. You will also be able to determine the effects of light and noise on the object's appearance, which will help match the object invariantly in the future. It allows you to determine what is expected and unexpected for the object's appearance with much higher confidence. Pretty cool idea huh? Dave agi | Archives | Modify Your Subscription
Re: [agi] Re: High Frame Rates Reduce Uncertainty
Thank you Matt. That's very useful input. On Mon, Jun 21, 2010 at 9:57 AM, Matt Mahoney matmaho...@yahoo.com wrote: Your computer monitor flashes 75 frames per second, but you don't notice any flicker because light sensing neurons have a response delay of about 100 ms. Motion detection begins in the retina by cells that respond to contrast between light and dark moving in specific directions computed by simple, fixed weight circuits. Higher up in the processing chain, you detect motion when your eyes and head smoothly track moving objects using kinesthetic feedback from your eye and neck muscles and input from your built in accelerometer in the semicircular canals in your ears. This is all very complicated of course. You are more likely to detect motion in objects that you recognize and expect to move, like people, animals, cars, etc. -- Matt Mahoney, matmaho...@yahoo.com -- *From:* David Jones davidher...@gmail.com *To:* agi agi@v2.listbox.com *Sent:* Mon, June 21, 2010 9:39:30 AM *Subject:* [agi] Re: High Frame Rates Reduce Uncertainty Ignoring Steve because we are simply going to have to agree to disagree... And I don't see enough value in trying to understand his paper. I said the math was overly complex, but what I really meant is that the approach is overly complex and so filled with research specific jargon, I don't care to try understand it. It is overly converned with copying the way that the brain does things. I don't care how the brain does it. I care about why the brain does it. Its the same as the analogy of giving a man a fish or teaching him to fish. You may figure out how the brain works, but it does you little good if you don't understand why it works that way. You would have to create a synthetic brain to take advantage of the knowledge, which is not a approach to AGI for many reasons. There are a million other ways, even better ways, to do it than the way the brain does it. Just because the brain accidentally found 1 way out of a million to do it doesn't make it the right way for us to develop AGI. So, moving on I can't find references online, but I've read that the Air Force studied the ability of the human eye to identify aircraft in images that were flashed on a screen at 1/220th of a second. So, clearly, the human eye can at least distinguish 220 fps if it operated that way. Of course, it may not operate on fps second, but that is besides the point. I've also heard other people say that a study has shown that the human eye takes 1000 exposures per second. They had no references though, so it is hearsay. The point was that the brain takes advantage of the fact that with such a high exposure rate, the changes between each image are very small if the objects are moving. This allows it to distinguish movement and visual changes with extremely low uncertainty. If it detects that the changes required to match two parts of an image are too high or the distance between matches is too far, it can reject a match. This allows it to distinguish only very low uncertainty changes and reject changes that have high uncertainty. I think this is a very significant discovery regarding how the brain is able to learn in such an ambiguous world with so many variables that are difficult to disambiguate, interpret and understand. Dave On Fri, Jun 18, 2010 at 2:19 PM, David Jones davidher...@gmail.comwrote: I just came up with an awesome idea. I just realized that the brain takes advantage of high frame rates to reduce uncertainty when it is estimating motion. The slower the frame rate, the more uncertainty there is because objects may have traveled too far between images to match with high certainty using simple techniques. So, this made me think, what if the secret to the brain's ability to learn generally stems from this high frame rate trick. What if we made a system that could process even high frame rates than the brain can. By doing this you can reduce the uncertainty of matches very very low (well in my theory so far). If you can do that, then you can learn about the objects in a video, how they move together or separately with very high certainty. You see, matching is the main barrier when learning about objects. But with a very high frame rate, we can use a fast algorithm and could potentially reduce the uncertainty to almost nothing. Once we learn about objects, matching gets easier because now we have training data and experience to take advantage of. In addition, you can also gain knowledge about lighting, color variation, noise, etc. With that knowledge, you can then automatically create a model of the object with extremely high confidence. You will also be able to determine the effects of light and noise on the object's appearance, which will help match the object invariantly in the future. It allows you to determine what is expected and unexpected for the object's
Re: [agi] High Frame Rates Reduce Uncertainty
The brain does not get the high frame rate signals as the eye itself only gives brain images at 24 frames per second. Else u wouldn't be able to watch a movie. Any comments? On 6/21/10, Matt Mahoney matmaho...@yahoo.com wrote: Your computer monitor flashes 75 frames per second, but you don't notice any flicker because light sensing neurons have a response delay of about 100 ms. Motion detection begins in the retina by cells that respond to contrast between light and dark moving in specific directions computed by simple, fixed weight circuits. Higher up in the processing chain, you detect motion when your eyes and head smoothly track moving objects using kinesthetic feedback from your eye and neck muscles and input from your built in accelerometer in the semicircular canals in your ears. This is all very complicated of course. You are more likely to detect motion in objects that you recognize and expect to move, like people, animals, cars, etc. -- Matt Mahoney, matmaho...@yahoo.com From: David Jones davidher...@gmail.com To: agi agi@v2.listbox.com Sent: Mon, June 21, 2010 9:39:30 AM Subject: [agi] Re: High Frame Rates Reduce Uncertainty Ignoring Steve because we are simply going to have to agree to disagree... And I don't see enough value in trying to understand his paper. I said the math was overly complex, but what I really meant is that the approach is overly complex and so filled with research specific jargon, I don't care to try understand it. It is overly converned with copying the way that the brain does things. I don't care how the brain does it. I care about why the brain does it. Its the same as the analogy of giving a man a fish or teaching him to fish. You may figure out how the brain works, but it does you little good if you don't understand why it works that way. You would have to create a synthetic brain to take advantage of the knowledge, which is not a approach to AGI for many reasons. There are a million other ways, even better ways, to do it than the way the brain does it. Just because the brain accidentally found 1 way out of a million to do it doesn't make it the right way for us to develop AGI. So, moving on I can't find references online, but I've read that the Air Force studied the ability of the human eye to identify aircraft in images that were flashed on a screen at 1/220th of a second. So, clearly, the human eye can at least distinguish 220 fps if it operated that way. Of course, it may not operate on fps second, but that is besides the point. I've also heard other people say that a study has shown that the human eye takes 1000 exposures per second. They had no references though, so it is hearsay. The point was that the brain takes advantage of the fact that with such a high exposure rate, the changes between each image are very small if the objects are moving. This allows it to distinguish movement and visual changes with extremely low uncertainty. If it detects that the changes required to match two parts of an image are too high or the distance between matches is too far, it can reject a match. This allows it to distinguish only very low uncertainty changes and reject changes that have high uncertainty. I think this is a very significant discovery regarding how the brain is able to learn in such an ambiguous world with so many variables that are difficult to disambiguate, interpret and understand. Dave On Fri, Jun 18, 2010 at 2:19 PM, David Jones davidher...@gmail.com wrote: I just came up with an awesome idea. I just realized that the brain takes advantage of high frame rates to reduce uncertainty when it is estimating motion. The slower the frame rate, the more uncertainty there is because objects may have traveled too far between images to match with high certainty using simple techniques. So, this made me think, what if the secret to the brain's ability to learn generally stems from this high frame rate trick. What if we made a system that could process even high frame rates than the brain can. By doing this you can reduce the uncertainty of matches very very low (well in my theory so far). If you can do that, then you can learn about the objects in a video, how they move together or separately with very high certainty. You see, matching is the main barrier when learning about objects. But with a very high frame rate, we can use a fast algorithm and could potentially reduce the uncertainty to almost nothing. Once we learn about objects, matching gets easier because now we have training data and experience to take advantage of. In addition, you can also gain knowledge about lighting, color variation, noise, etc. With that knowledge, you can then automatically create a model of the object with extremely high confidence. You will also be able to determine the effects of light and noise on the object's appearance, which will help match the object
Re: [agi] High Frame Rates Reduce Uncertainty
I'm reading about the retina motion processing. Maybe the brain does only get 24 frames per second, but the retina may send it information about hypothesized or likely movements. The brain can then do some further processing, such as using including kinesthetic feedback that tells the brain about how the body moved during the time that the images were captured, which Matt mentioned (thanks again). Maybe the important question here is could we potentially create a real camera system to capture extremely high frame rates for the purposes of generating vast amounts of very low uncertainty training data that could be used to create algorithms that do not require such high frame rates. With such training data, you could automatically test algorithms much more quickly. You could potentially even use genetic algorithms to find the right solution, although I'm not a big fan of such an approach. To solve screenshot computer vision, I could potentially generate screenshots with very high frame rates for training data also. Even if the process is supervised, it could probably generate vastly more training data than we've ever had before. Dave On Mon, Jun 21, 2010 at 10:30 AM, deepakjnath deepakjn...@gmail.com wrote: The brain does not get the high frame rate signals as the eye itself only gives brain images at 24 frames per second. Else u wouldn't be able to watch a movie. Any comments? On 6/21/10, Matt Mahoney matmaho...@yahoo.com wrote: Your computer monitor flashes 75 frames per second, but you don't notice any flicker because light sensing neurons have a response delay of about 100 ms. Motion detection begins in the retina by cells that respond to contrast between light and dark moving in specific directions computed by simple, fixed weight circuits. Higher up in the processing chain, you detect motion when your eyes and head smoothly track moving objects using kinesthetic feedback from your eye and neck muscles and input from your built in accelerometer in the semicircular canals in your ears. This is all very complicated of course. You are more likely to detect motion in objects that you recognize and expect to move, like people, animals, cars, etc. -- Matt Mahoney, matmaho...@yahoo.com From: David Jones davidher...@gmail.com To: agi agi@v2.listbox.com Sent: Mon, June 21, 2010 9:39:30 AM Subject: [agi] Re: High Frame Rates Reduce Uncertainty Ignoring Steve because we are simply going to have to agree to disagree... And I don't see enough value in trying to understand his paper. I said the math was overly complex, but what I really meant is that the approach is overly complex and so filled with research specific jargon, I don't care to try understand it. It is overly converned with copying the way that the brain does things. I don't care how the brain does it. I care about why the brain does it. Its the same as the analogy of giving a man a fish or teaching him to fish. You may figure out how the brain works, but it does you little good if you don't understand why it works that way. You would have to create a synthetic brain to take advantage of the knowledge, which is not a approach to AGI for many reasons. There are a million other ways, even better ways, to do it than the way the brain does it. Just because the brain accidentally found 1 way out of a million to do it doesn't make it the right way for us to develop AGI. So, moving on I can't find references online, but I've read that the Air Force studied the ability of the human eye to identify aircraft in images that were flashed on a screen at 1/220th of a second. So, clearly, the human eye can at least distinguish 220 fps if it operated that way. Of course, it may not operate on fps second, but that is besides the point. I've also heard other people say that a study has shown that the human eye takes 1000 exposures per second. They had no references though, so it is hearsay. The point was that the brain takes advantage of the fact that with such a high exposure rate, the changes between each image are very small if the objects are moving. This allows it to distinguish movement and visual changes with extremely low uncertainty. If it detects that the changes required to match two parts of an image are too high or the distance between matches is too far, it can reject a match. This allows it to distinguish only very low uncertainty changes and reject changes that have high uncertainty. I think this is a very significant discovery regarding how the brain is able to learn in such an ambiguous world with so many variables that are difficult to disambiguate, interpret and understand. Dave On Fri, Jun 18, 2010 at 2:19 PM, David Jones davidher...@gmail.com wrote: I just came up with an awesome idea. I just realized that the brain takes advantage of high frame rates
Re: [agi] A fundamental limit on intelligence?!
Steve, You didn't mention this, so I guess I will: larger animals do generally have larger brains, coming close to a fixed brain/body ratio. Smarter animals appear to be the ones with a higher brain/body ratio rather than simply a larger brain. This to me suggests that the amount of sensory information and muscle coordination necessary is the most important determiner of the amount of processing power needed. There could be other interpretations, however. It's also pretty important to say that brains are expensive to fuel. It's probably the case that other animals didn't get as smart as us because the additional food they could get per ounce brain was less than the additional food needed to support an ounce of brain. Humans were in a situation in which it was more. So, I don't think your argument from other animals supports your hypothesis terribly well. One way around your instability if it exists would be (similar to your hemisphere suggestion) split the network into a number of individuals which cooperate through very low-bandwidth connections. This would be like an organization of humans working together. Hence, multiagent systems would have a higher stability limit. However, it is still the case that we hit a serious diminishing-returns scenario once we needed to start doing this (since the low-bandwidth connections convey so much less info, we need waaay more processing power for every IQ point or whatever). And, once these organizations got really big, it's quite plausible that they'd have their own stability issues. --Abram On Mon, Jun 21, 2010 at 11:19 AM, Steve Richfield steve.richfi...@gmail.com wrote: There has been an ongoing presumption that more brain (or computer) means more intelligence. I would like to question that underlying presumption. That being the case, why don't elephants and other large creatures have really gigantic brains? This seems to be SUCH an obvious evolutionary step. There are all sorts of network-destroying phenomena that rise from complex networks, e.g. phase shift oscillators there circular analysis paths enforce themselves, computational noise is endlessly analyzed, etc. We know that our own brains are just barely stable, as flashing lights throw some people into epileptic attacks, etc. Perhaps network stability is the intelligence limiter? If so, then we aren't going to get anywhere without first fully understanding it. Suppose for a moment that theoretically perfect neurons could work in a brain of limitless size, but their imperfections accumulate (or multiply) to destroy network operation when you get enough of them together. Brains have grown larger because neurons have evolved to become more nearly perfect, without having yet (or ever) reaching perfection. Hence, evolution may have struck a balance, where less intelligence directly impairs survivability, and greater intelligence impairs network stability, and hence indirectly impairs survivability. If the above is indeed the case, then AGI and related efforts don't stand a snowball's chance in hell of ever outperforming humans, UNTIL the underlying network stability theory is well enough understood to perform perfectly to digital precision. This wouldn't necessarily have to address all aspects of intelligence, but would at minimum have to address large-scale network stability. One possibility is chopping large networks into pieces, e.g. the hemispheres of our own brains. However, like multi-core CPUs, there is work for only so many CPUs/hemispheres. There are some medium-scale network similes in the world, e.g. the power grid. However, there they have high-level central control and lots of crashes, so there may not be much to learn from them. Note in passing that I am working with some non-AGIers on power grid stability issues. While not fully understood, the primary challenge appears (to me) to be that the various control mechanisms (that includes humans in the loop) violate a basic requirement for feedback stability, namely, that the frequency response not drop off faster then 12db/octave at any frequency. Present control systems make binary all-or-nothing decisions that produce astronomical high-frequency components (edges and glitches) related to much lower-frequency phenomena (like overall demand). Other systems then attempt to deal with these edges and glitches, with predictable poor results. Like the stock market crash of May 6, there is a list of dates of major outages and near-outages, where the failures are poorly understood. In some cases, the lights stayed on, but for a few seconds came ever SO close to a widespread outage that dozens of articles were written about them, with apparently no one understanding things even to the basic level that I am explaining here. Hence, a single theoretical insight might guide both power grid development and AGI development. For example, perhaps there is a necessary capability of components in large
[agi] Fwd: AGI question
Hi I'm new to this list, but I've been thinking about consciousness, cognition and AI for about half of my life (I'm 32 years old). As is probably the case for many of us here, my interests began with direct recognition of the depth and wonder of varieties of phenomenological experiences-- and attempting to comprehend how these constellations of significance fit in with a larger picture of what we can reliably know about the natural world. I am secondarily motivated by the fact that (considerations of morality or amorality aside) AGI is inevitable, though it is far from being a forgone conclusion that powerful general thinking machines will have a first-hand subjective relationship to a world, as living creatures do-- and therefore it is vital that we do as well as possible in understanding what makes systems conscious. A zombie machine intelligence singularity is something I would refer to rather as a holocaust, even if no one were directly killed, assuming these entities could ultimately prevail over the previous forms of life on our planet. I'm sure I'm not the only one on this list who sees a behavioral/ecological level of analysis as the most likely correct level at which to study perception and cognition, and perception as being a kind of active relationship between an organism and an environment. Having thoroughly convinced my self of a non-dualist, embodied, externalist perspective on cognition, I turn to the nature of life itself (and possibly even physics but maybe that level will not be necessary) to make sense of the nature of subjectivity. I like Bohm's or Bateson's panpsychism about systems as wholes, and significance as informational distinctions (which it would be natural to understand as being the basis of subjective experience), but this is descriptive rather than explanatory. I am not a biologist, but I am increasingly interested in finding answers to what it is about living organisms that gives them a unity such that something is something to the system as a whole. The line of investigation that theoretical biologists like Robert Rosen and other NLDS/chaos people have pursued is interesting, but I am unfamiliar with related work that might have made more progress on the system-level properties that give life its characteristic unity and system-level responsiveness. To me, this seems the most likely candidate for a paradigm shift that would produce AGI. In contrast I'm not particularly convinced that modeling a brain is a good way to get AGI, although I'd guess we could learn a few more things about the coordination of complex behavior if we could really understand them. Another way to put this is that obviously evolutionary computation would be more than just boring hill-climbing if we knew what an organism even IS (perhaps in a more precise computational sense). If we can know what an organism is then it should be (maybe) trivial to model concepts, consciousness, and high level semantics to the umpteenth degree, or at least this would be a major hurtle I think. Even assuming a solution to the problem posed above, there is still plenty of room for other minds skepticism in non-living entities implemented on questionably foreign mediums but there would be a lot more reason to sleep well that the science/technology is leading in a direction in which questions about subjectivity could be meaningfully investigated. Rob --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] An alternative plan to discover self-organization theory
(I'm a little late in this conversation. I tried to send this message the other day but I had my list membership configured wrong. -Rob) -- Forwarded message -- From: rob levy r.p.l...@gmail.com Date: Sun, Jun 20, 2010 at 5:48 PM Subject: Re: [agi] An alternative plan to discover self-organization theory To: agi@v2.listbox.com On a related note, what is everyone's opinion on why evolutionary algorithms are such a miserable failure as creative machines, despite their successes in narrow optimization problems? I don't want to conflate the possibly separable problems of biological development and evolution, though they are interrelated. There are various approaches to evolutionary theory such as Lima de Faria's evolution without selection ideas and Reid's evolution by natural experiment that suggest natural selection is not all it's cracked up to be, and that the step of generating, (mutating, combining, ) is where the more interesting stuff happens. Most of the alternatives to Neodarwinian Synthesis I have seen are based in dynamic models of emergence in complex systems. The upshot is, you don't get creativity for free, you actually still need to solve a problem that is as hard as AGI in order to get creativity for free. So, you would need to solve the AGI-hard problem of evolution and development of life, in order to then solve AGI itself (reminds me of the old SNL sketch: first, get a million dollars...). Also, my hunch is that there is quite a bit of overlap between the solutions to the two problems. Rob Disclaimer: I'm discussing things above that I'm not and don't claim to be an expert in, but from what I have seen so far on this list, that should be alright. AGI is by its nature very multidisciplinary which necessitates often being breadth-first, and therefore shallow in some areas. On Sun, Jun 20, 2010 at 2:06 AM, Steve Richfield steve.richfi...@gmail.comwrote: No, I haven't been smokin' any wacky tobacy. Instead, I was having a long talk with my son Eddie, about self-organization theory. This is *his*proposal: He suggested that I construct a simple NN that couldn't work without self organizing, and make dozens/hundreds of different neuron and synapse operational characteristics selectable ala genetic programming, put it on the fastest computer I could get my hands on, turn it loose trying arbitrary combinations of characteristics, and see what the winning combination turns out to be. Then, armed with that knowledge, refine the genetic characteristics and do it again, and iterate until it *efficiently* self organizes. This might go on for months, but self-organization theory might just emerge from such an effort. I had a bunch of objections to his approach, e.g. Q. What if it needs something REALLY strange to work? A. Who better than you to come up with a long list of really strange functionality? Q. There are at least hundreds of bits in the genome. A. Try combinations in pseudo-random order, with each bit getting asserted in ~half of the tests. If/when you stumble onto a combination that sort of works, switch to varying the bits one-at-a-time, and iterate in this way until the best combination is found. Q. Where are we if this just burns electricity for a few months and finds nothing? A. Print out the best combination, break out the wacky tobacy, and come up with even better/crazier parameters to test. I have never written a line of genetic programming, but I know that others here have. Perhaps you could bring some rationality to this discussion? What would be a simple NN that needs self-organization? Maybe a small pot of neurons that could only work if they were organized into layers, e.g. a simple 64-neuron system that would work as a 4x4x4-layer visual recognition system, given the input that I fed it? Any thoughts on how to score partial successes? Has anyone tried anything like this in the past? Is anyone here crazy enough to want to help with such an effort? This Monte Carlo approach might just be simple enough to work, and simple enough that it just HAS to be tried. All thoughts, stones, and rotten fruit will be gratefully appreciated. Thanks in advance. Steve *agi* | Archives https://www.listbox.com/member/archive/303/=now https://www.listbox.com/member/archive/rss/303/ | Modifyhttps://www.listbox.com/member/?;Your Subscription http://www.listbox.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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
RE: [agi] A fundamental limit on intelligence?!
-Original Message- From: Steve Richfield [mailto:steve.richfi...@gmail.com] My underlying thought here is that we may all be working on the wrong problems. Instead of working on the particular analysis methods (AGI) or self-organization theory (NN), perhaps if someone found a solution to large- network stability, then THAT would show everyone the ways to their respective goals. For a distributed AGI this is a fundamental problem. Difference is that a power grid is such a fixed network. A distributed AGI need not be that fixed, it could lose chunks of itself but grow them out somewhere else. Though a distributed AGI could be required to run as a fixed network. Some traditional telecommunications networks are power grid like. They have a drastic amount of stability and healing functions built-in as have been added over time. Solutions for large-scale network stabilities would vary per network topology, function, etc.. Virtual networks play a large part, this would be related to the network's ability to reconstruct itself meaning knowing how to heal, reroute, optimize and grow.. John --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] A fundamental limit on intelligence?!
Abram, On Mon, Jun 21, 2010 at 8:38 AM, Abram Demski abramdem...@gmail.com wrote: Steve, You didn't mention this, so I guess I will: larger animals do generally have larger brains, coming close to a fixed brain/body ratio. Smarter animals appear to be the ones with a higher brain/body ratio rather than simply a larger brain. This to me suggests that the amount of sensory information and muscle coordination necessary is the most important determiner of the amount of processing power needed. There could be other interpretations, however. It is REALLY hard to compare the intelligence of various animals, because of their innate behavior being overlaid. For example, based on ability to follow instruction, cats must be REALLY stupid. It's also pretty important to say that brains are expensive to fuel. It's probably the case that other animals didn't get as smart as us because the additional food they could get per ounce brain was less than the additional food needed to support an ounce of brain. Humans were in a situation in which it was more. So, I don't think your argument from other animals supports your hypothesis terribly well. Presuming for a moment that you are right, then there will be no singularity! No, this is NOT a reductio ad absurdum proof either way. Why no singularity? If there really is a limit to the value of intelligence, then why should we think that there will be anything special about super-intelligence? Perhaps we have been deluding ourselves because we want to think that the reason we aren't all rich is because we just aren't smart enough, when in reality some entirely different phenomenon may be key? Have YOU observed that success in life is highly correlated to intelligence? One way around your instability if it exists would be (similar to your hemisphere suggestion) split the network into a number of individuals which cooperate through very low-bandwidth connections. While helping breadth of analysis, this would seem to absolutely limit analysis depth to that of one individual. This would be like an organization of humans working together. Hence, multiagent systems would have a higher stability limit. Providing they don't get into a war of some sort. However, it is still the case that we hit a serious diminishing-returns scenario once we needed to start doing this (since the low-bandwidth connections convey so much less info, we need waaay more processing power for every IQ point or whatever). I see more problems with analysis depth than with bandwidth limitations. And, once these organizations got really big, it's quite plausible that they'd have their own stability issues. Yes. Steve On Mon, Jun 21, 2010 at 11:19 AM, Steve Richfield steve.richfi...@gmail.com wrote: There has been an ongoing presumption that more brain (or computer) means more intelligence. I would like to question that underlying presumption. That being the case, why don't elephants and other large creatures have really gigantic brains? This seems to be SUCH an obvious evolutionary step. There are all sorts of network-destroying phenomena that rise from complex networks, e.g. phase shift oscillators there circular analysis paths enforce themselves, computational noise is endlessly analyzed, etc. We know that our own brains are just barely stable, as flashing lights throw some people into epileptic attacks, etc. Perhaps network stability is the intelligence limiter? If so, then we aren't going to get anywhere without first fully understanding it. Suppose for a moment that theoretically perfect neurons could work in a brain of limitless size, but their imperfections accumulate (or multiply) to destroy network operation when you get enough of them together. Brains have grown larger because neurons have evolved to become more nearly perfect, without having yet (or ever) reaching perfection. Hence, evolution may have struck a balance, where less intelligence directly impairs survivability, and greater intelligence impairs network stability, and hence indirectly impairs survivability. If the above is indeed the case, then AGI and related efforts don't stand a snowball's chance in hell of ever outperforming humans, UNTIL the underlying network stability theory is well enough understood to perform perfectly to digital precision. This wouldn't necessarily have to address all aspects of intelligence, but would at minimum have to address large-scale network stability. One possibility is chopping large networks into pieces, e.g. the hemispheres of our own brains. However, like multi-core CPUs, there is work for only so many CPUs/hemispheres. There are some medium-scale network similes in the world, e.g. the power grid. However, there they have high-level central control and lots of crashes, so there may not be much to learn from them. Note in passing that I am working with some non-AGIers on power grid stability
Re: [agi] A fundamental limit on intelligence?!
I think a real world solution to grid stability would require greater use of sensory devices (and a some sensory-feedback devices). I really don't know for sure, but my assumption is that electrical grid management has relied mostly on the electrical reactions of the grid itself, and here you are saying that is just not good enough for critical fluctuations in 2010. So while software is also necessary of course, the first change in how grid management should be done is through greater reliance on off-the-grid (or at minimal backup on-grid) sensory devices. I am quite confident, without knowing anything about the subject, that that is what needs to be done because I understand a little about how different groups of people work and I have seen how sensory devices like gps and lidar have fundamentally changed AI projects because they allowed time sensitive critical analysis that was too slow and for contemporary AI to solve. 100 years from now, electrical grid management won't require another layer of sensors because the software analysis of grid fluctuations will be sufficient. On the other hand, grid managers will not remove these additional layers of sensors from the grid a hundred years from now anymore than we telephone engineers would suggest that maybe they should stop using fiber optics because they could get back to 1990 fiber optic capacity and reliability using copper wire with today's switching and software devices. Jim Bromer On Mon, Jun 21, 2010 at 11:19 AM, Steve Richfield steve.richfi...@gmail.com wrote: There has been an ongoing presumption that more brain (or computer) means more intelligence. I would like to question that underlying presumption. That being the case, why don't elephants and other large creatures have really gigantic brains? This seems to be SUCH an obvious evolutionary step. There are all sorts of network-destroying phenomena that rise from complex networks, e.g. phase shift oscillators there circular analysis paths enforce themselves, computational noise is endlessly analyzed, etc. We know that our own brains are just barely stable, as flashing lights throw some people into epileptic attacks, etc. Perhaps network stability is the intelligence limiter? If so, then we aren't going to get anywhere without first fully understanding it. Suppose for a moment that theoretically perfect neurons could work in a brain of limitless size, but their imperfections accumulate (or multiply) to destroy network operation when you get enough of them together. Brains have grown larger because neurons have evolved to become more nearly perfect, without having yet (or ever) reaching perfection. Hence, evolution may have struck a balance, where less intelligence directly impairs survivability, and greater intelligence impairs network stability, and hence indirectly impairs survivability. If the above is indeed the case, then AGI and related efforts don't stand a snowball's chance in hell of ever outperforming humans, UNTIL the underlying network stability theory is well enough understood to perform perfectly to digital precision. This wouldn't necessarily have to address all aspects of intelligence, but would at minimum have to address large-scale network stability. One possibility is chopping large networks into pieces, e.g. the hemispheres of our own brains. However, like multi-core CPUs, there is work for only so many CPUs/hemispheres. There are some medium-scale network similes in the world, e.g. the power grid. However, there they have high-level central control and lots of crashes, so there may not be much to learn from them. Note in passing that I am working with some non-AGIers on power grid stability issues. While not fully understood, the primary challenge appears (to me) to be that the various control mechanisms (that includes humans in the loop) violate a basic requirement for feedback stability, namely, that the frequency response not drop off faster then 12db/octave at any frequency. Present control systems make binary all-or-nothing decisions that produce astronomical high-frequency components (edges and glitches) related to much lower-frequency phenomena (like overall demand). Other systems then attempt to deal with these edges and glitches, with predictable poor results. Like the stock market crash of May 6, there is a list of dates of major outages and near-outages, where the failures are poorly understood. In some cases, the lights stayed on, but for a few seconds came ever SO close to a widespread outage that dozens of articles were written about them, with apparently no one understanding things even to the basic level that I am explaining here. Hence, a single theoretical insight might guide both power grid development and AGI development. For example, perhaps there is a necessary capability of components in large networks, to be able to custom tailor their frequency response curves to not participate on unstable
Re: [agi] A fundamental limit on intelligence?!
John, Your comments appear to be addressing reliability, rather than stability... On Mon, Jun 21, 2010 at 9:12 AM, John G. Rose johnr...@polyplexic.comwrote: -Original Message- From: Steve Richfield [mailto:steve.richfi...@gmail.com] My underlying thought here is that we may all be working on the wrong problems. Instead of working on the particular analysis methods (AGI) or self-organization theory (NN), perhaps if someone found a solution to large- network stability, then THAT would show everyone the ways to their respective goals. For a distributed AGI this is a fundamental problem. Difference is that a power grid is such a fixed network. Not really. Switches may connect or disconnect Canada, equipment is constantly failing and being repaired, etc. In any case, this doesn't seem to be related to stability, other than it being a lot easier to analyze a fixed network rather than a variable network. A distributed AGI need not be that fixed, it could lose chunks of itself but grow them out somewhere else. Though a distributed AGI could be required to run as a fixed network. Some traditional telecommunications networks are power grid like. They have a drastic amount of stability and healing functions built-in as have been added over time. However, there is no feedback, so stability isn't even a potential issue. Solutions for large-scale network stabilities would vary per network topology, function, etc.. However, there ARE some universal rules, like the 12db/octave requirement. Virtual networks play a large part, this would be related to the network's ability to reconstruct itself meaning knowing how to heal, reroute, optimize and grow.. Again, this doesn't seem to relate to millisecond-by-millisecond stability. Steve --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] Fwd: AGI question
rob levy wrote: I am secondarily motivated by the fact that (considerations of morality or amorality aside) AGI is inevitable, though it is far from being a forgone conclusion that powerful general thinking machines will have a first-hand subjective relationship to a world, as living creatures do-- and therefore it is vital that we do as well as possible in understanding what makes systems conscious. A zombie machine intelligence singularity is something I would refer to rather as a holocaust, even if no one were directly killed, assuming these entities could ultimately prevail over the previous forms of life on our planet. What do you mean by conscious? If your brain were removed and replaced by a functionally equivalent computer that simulated your behavior (presumably a zombie), how would you be any different? Why would it matter? -- Matt Mahoney, matmaho...@yahoo.com From: rob levy r.p.l...@gmail.com To: agi agi@v2.listbox.com Sent: Mon, June 21, 2010 11:53:29 AM Subject: [agi] Fwd: AGI question Hi I'm new to this list, but I've been thinking about consciousness, cognition and AI for about half of my life (I'm 32 years old). As is probably the case for many of us here, my interests began with direct recognition of the depth and wonder of varieties of phenomenological experiences-- and attempting to comprehend how these constellations of significance fit in with a larger picture of what we can reliably know about the natural world. I am secondarily motivated by the fact that (considerations of morality or amorality aside) AGI is inevitable, though it is far from being a forgone conclusion that powerful general thinking machines will have a first-hand subjective relationship to a world, as living creatures do-- and therefore it is vital that we do as well as possible in understanding what makes systems conscious. A zombie machine intelligence singularity is something I would refer to rather as a holocaust, even if no one were directly killed, assuming these entities could ultimately prevail over the previous forms of life on our planet. I'm sure I'm not the only one on this list who sees a behavioral/ecological level of analysis as the most likely correct level at which to study perception and cognition, and perception as being a kind of active relationship between an organism and an environment. Having thoroughly convinced my self of a non-dualist, embodied, externalist perspective on cognition, I turn to the nature of life itself (and possibly even physics but maybe that level will not be necessary) to make sense of the nature of subjectivity. I like Bohm's or Bateson's panpsychism about systems as wholes, and significance as informational distinctions (which it would be natural to understand as being the basis of subjective experience), but this is descriptive rather than explanatory. I am not a biologist, but I am increasingly interested in finding answers to what it is about living organisms that gives them a unity such that something is something to the system as a whole. The line of investigation that theoretical biologists like Robert Rosen and other NLDS/chaos people have pursued is interesting, but I am unfamiliar with related work that might have made more progress on the system-level properties that give life its characteristic unity and system-level responsiveness. To me, this seems the most likely candidate for a paradigm shift that would produce AGI. In contrast I'm not particularly convinced that modeling a brain is a good way to get AGI, although I'd guess we could learn a few more things about the coordination of complex behavior if we could really understand them. Another way to put this is that obviously evolutionary computation would be more than just boring hill-climbing if we knew what an organism even IS (perhaps in a more precise computational sense). If we can know what an organism is then it should be (maybe) trivial to model concepts, consciousness, and high level semantics to the umpteenth degree, or at least this would be a major hurtle I think. Even assuming a solution to the problem posed above, there is still plenty of room for other minds skepticism in non-living entities implemented on questionably foreign mediums but there would be a lot more reason to sleep well that the science/technology is leading in a direction in which questions about subjectivity could be meaningfully investigated. Rob agi | Archives | Modify Your Subscription --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] A fundamental limit on intelligence?!
Jim, Yours is the prevailing view in the industry. However, it doesn't seem to work. Even given months of time to analyze past failures, they are often unable to divine rules that would have reliably avoided the problems. In short, until you adequately understand the system that your sensors are sensing, all the readings in the world won't help. Further, when a system is fundamentally unstable, you must have a control system that completely deals with the instability, or it absolutely will fail. The present system meets neither of these criteria. There is another MAJOR issue. Presuming a power control center in the middle of the U.S., the round-trip time at the speed of light to each coast is ~16ms, or two half-cycles at 60Hz. In control terms, that is an eternity. Distributed control requires fundamental stability to function reliably. Times can be improved by having separate control systems for each coast, but the interface would still have to meet fundamental stability criteria (like limiting the rates of change), and our long coasts would still require a full half-cycle of time to respond. Note that faults must be responded to QUICKLY to save the equipment, and so cannot be left to central control systems to operate. So, we end up with the system we now have, that does NOT meet reasonable stability criteria. Hence, we may forever have occasional outages until the system is radically re-conceived. Steve == On Mon, Jun 21, 2010 at 9:17 AM, Jim Bromer jimbro...@gmail.com wrote: I think a real world solution to grid stability would require greater use of sensory devices (and a some sensory-feedback devices). I really don't know for sure, but my assumption is that electrical grid management has relied mostly on the electrical reactions of the grid itself, and here you are saying that is just not good enough for critical fluctuations in 2010. So while software is also necessary of course, the first change in how grid management should be done is through greater reliance on off-the-grid (or at minimal backup on-grid) sensory devices. I am quite confident, without knowing anything about the subject, that that is what needs to be done because I understand a little about how different groups of people work and I have seen how sensory devices like gps and lidar have fundamentally changed AI projects because they allowed time sensitive critical analysis that was too slow and for contemporary AI to solve. 100 years from now, electrical grid management won't require another layer of sensors because the software analysis of grid fluctuations will be sufficient. On the other hand, grid managers will not remove these additional layers of sensors from the grid a hundred years from now anymore than we telephone engineers would suggest that maybe they should stop using fiber optics because they could get back to 1990 fiber optic capacity and reliability using copper wire with today's switching and software devices. Jim Bromer On Mon, Jun 21, 2010 at 11:19 AM, Steve Richfield steve.richfi...@gmail.com wrote: There has been an ongoing presumption that more brain (or computer) means more intelligence. I would like to question that underlying presumption. That being the case, why don't elephants and other large creatures have really gigantic brains? This seems to be SUCH an obvious evolutionary step. There are all sorts of network-destroying phenomena that rise from complex networks, e.g. phase shift oscillators there circular analysis paths enforce themselves, computational noise is endlessly analyzed, etc. We know that our own brains are just barely stable, as flashing lights throw some people into epileptic attacks, etc. Perhaps network stability is the intelligence limiter? If so, then we aren't going to get anywhere without first fully understanding it. Suppose for a moment that theoretically perfect neurons could work in a brain of limitless size, but their imperfections accumulate (or multiply) to destroy network operation when you get enough of them together. Brains have grown larger because neurons have evolved to become more nearly perfect, without having yet (or ever) reaching perfection. Hence, evolution may have struck a balance, where less intelligence directly impairs survivability, and greater intelligence impairs network stability, and hence indirectly impairs survivability. If the above is indeed the case, then AGI and related efforts don't stand a snowball's chance in hell of ever outperforming humans, UNTIL the underlying network stability theory is well enough understood to perform perfectly to digital precision. This wouldn't necessarily have to address all aspects of intelligence, but would at minimum have to address large-scale network stability. One possibility is chopping large networks into pieces, e.g. the hemispheres of our own brains. However, like multi-core CPUs, there is work for only so many CPUs/hemispheres. There
Re: [agi] An alternative plan to discover self-organization theory
rob levy wrote: On a related note, what is everyone's opinion on why evolutionary algorithms are such a miserable failure as creative machines, despite their successes in narrow optimization problems? Lack of computing power. How much computation would you need to simulate the 3 billion years of evolution that created human intelligence? -- Matt Mahoney, matmaho...@yahoo.com From: rob levy r.p.l...@gmail.com To: agi agi@v2.listbox.com Sent: Mon, June 21, 2010 11:56:53 AM Subject: Re: [agi] An alternative plan to discover self-organization theory (I'm a little late in this conversation. I tried to send this message the other day but I had my list membership configured wrong. -Rob) -- Forwarded message -- From: rob levy r.p.l...@gmail.com Date: Sun, Jun 20, 2010 at 5:48 PM Subject: Re: [agi] An alternative plan to discover self-organization theory To: agi@v2.listbox.com On a related note, what is everyone's opinion on why evolutionary algorithms are such a miserable failure as creative machines, despite their successes in narrow optimization problems? I don't want to conflate the possibly separable problems of biological development and evolution, though they are interrelated. There are various approaches to evolutionary theory such as Lima de Faria's evolution without selection ideas and Reid's evolution by natural experiment that suggest natural selection is not all it's cracked up to be, and that the step of generating, (mutating, combining, ) is where the more interesting stuff happens. Most of the alternatives to Neodarwinian Synthesis I have seen are based in dynamic models of emergence in complex systems. The upshot is, you don't get creativity for free, you actually still need to solve a problem that is as hard as AGI in order to get creativity for free. So, you would need to solve the AGI-hard problem of evolution and development of life, in order to then solve AGI itself (reminds me of the old SNL sketch: first, get a million dollars...). Also, my hunch is that there is quite a bit of overlap between the solutions to the two problems. Rob Disclaimer: I'm discussing things above that I'm not and don't claim to be an expert in, but from what I have seen so far on this list, that should be alright. AGI is by its nature very multidisciplinary which necessitates often being breadth-first, and therefore shallow in some areas. On Sun, Jun 20, 2010 at 2:06 AM, Steve Richfield steve.richfi...@gmail.com wrote: No, I haven't been smokin' any wacky tobacy. Instead, I was having a long talk with my son Eddie, about self-organization theory. This is his proposal: He suggested that I construct a simple NN that couldn't work without self organizing, and make dozens/hundreds of different neuron and synapse operational characteristics selectable ala genetic programming, put it on the fastest computer I could get my hands on, turn it loose trying arbitrary combinations of characteristics, and see what the winning combination turns out to be. Then, armed with that knowledge, refine the genetic characteristics and do it again, and iterate until it efficiently self organizes. This might go on for months, but self-organization theory might just emerge from such an effort. I had a bunch of objections to his approach, e.g. Q. What if it needs something REALLY strange to work? A. Who better than you to come up with a long list of really strange functionality? Q. There are at least hundreds of bits in the genome. A. Try combinations in pseudo-random order, with each bit getting asserted in ~half of the tests. If/when you stumble onto a combination that sort of works, switch to varying the bits one-at-a-time, and iterate in this way until the best combination is found. Q. Where are we if this just burns electricity for a few months and finds nothing? A. Print out the best combination, break out the wacky tobacy, and come up with even better/crazier parameters to test. I have never written a line of genetic programming, but I know that others here have. Perhaps you could bring some rationality to this discussion? What would be a simple NN that needs self-organization? Maybe a small pot of neurons that could only work if they were organized into layers, e.g. a simple 64-neuron system that would work as a 4x4x4-layer visual recognition system, given the input that I fed it? Any thoughts on how to score partial successes? Has anyone tried anything like this in the past? Is anyone here crazy enough to want to help with such an effort? This Monte Carlo approach might just be simple enough to work, and simple enough that it just HAS to be tried. All thoughts, stones, and rotten fruit will be gratefully appreciated. Thanks in advance. Steve agi | Archives | Modify Your Subscription agi | Archives | Modify Your Subscription --- agi
RE: [agi] A fundamental limit on intelligence?!
-Original Message- From: Steve Richfield [mailto:steve.richfi...@gmail.com] John, Your comments appear to be addressing reliability, rather than stability... Both can be very interrelated. It can be an oversimplification to separate them, or too impractical/theoretical. On Mon, Jun 21, 2010 at 9:12 AM, John G. Rose johnr...@polyplexic.com wrote: -Original Message- From: Steve Richfield [mailto:steve.richfi...@gmail.com] My underlying thought here is that we may all be working on the wrong problems. Instead of working on the particular analysis methods (AGI) or self-organization theory (NN), perhaps if someone found a solution to large- network stability, then THAT would show everyone the ways to their respective goals. For a distributed AGI this is a fundamental problem. Difference is that a power grid is such a fixed network. Not really. Switches may connect or disconnect Canada, equipment is constantly failing and being repaired, etc. In any case, this doesn't seem to be related to stability, other than it being a lot easier to analyze a fixed network rather than a variable network. There are a fixed amount of copper wires going into a node. The network is usually a hierarchy of networks. Fixed may be more limiting, sophisticated and kludged rendering it more difficult to deal with so don't assume. A distributed AGI need not be that fixed, it could lose chunks of itself but grow them out somewhere else. Though a distributed AGI could be required to run as a fixed network. Some traditional telecommunications networks are power grid like. They have a drastic amount of stability and healing functions built-in as have been added over time. However, there is no feedback, so stability isn't even a potential issue. No feedback? Remember some traditional telecommunications networks run over copper with power, and are analog; there are huge feedback issues of which many taken care of at a lower signaling level or with external equipment such as echo-cancellers. Again though, there is a hierarchy and mesh of various networks here. I've suggested traditional telecommunications since they are vastly more complex, real-time and many other networks have learned from it. Solutions for large-scale network stabilities would vary per network topology, function, etc.. However, there ARE some universal rules, like the 12db/octave requirement. Really? Do networks such as botnets really care about this? Or does it apply? Virtual networks play a large part, this would be related to the network's ability to reconstruct itself meaning knowing how to heal, reroute, optimize and grow.. Again, this doesn't seem to relate to millisecond-by-millisecond stability. It could be as the virtual network might contain images of the actual network, as an internal model and use this for changing the network structure for a more stable one if there were timing issues... Just some thoughts... John --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
[agi] Formulaic vs. Equation AGI
One constant in ALL proposed methods leading to computational intelligence is formulaic operation, where agents, elements, neurons, etc., process inputs to produce outputs. There is scant biological evidence for this, and plenty of evidence for a balanced equation operation. Note that unbalancing one side, e.g. by injecting current, would result in a responding imbalance on the other side, so that synapses might (erroneously) appear to be one-way. However, there is plenty of evidence that information flows both ways, e.g. retrograde flow of information to support learning. Even looking at seemingly one-way things like the olfactory nerve, there are axons going both ways. No, I don't have any sort of comprehensive balanced-equation theory of intelligent operation, but I can see the interesting possibility. Suppose that the key to life is not competition, but rather is fitting into the world. Perhaps we don't so much observe things as orchestrate them to our needs. Hence, we and our world are in a gigantic loop, adjusting our outputs to achieve balancing characteristics in our inputs. Imbalances precipitate changes in action to achieve balance. The only difference between us and our world is implementation detail. We do our part, and it does its part. I'm sure that there are Zen Buddhists out there who would just LOVE this yin-yang view of things. Any thoughts? Steve --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] A fundamental limit on intelligence?!
John, On Mon, Jun 21, 2010 at 10:06 AM, John G. Rose johnr...@polyplexic.comwrote: Solutions for large-scale network stabilities would vary per network topology, function, etc.. However, there ARE some universal rules, like the 12db/octave requirement. Really? Do networks such as botnets really care about this? Or does it apply? Anytime negative feedback can become positive feedback because of delays or phase shifts, this becomes an issue. Many competent EE people fail to see the phase shifting that many decision processes can introduce, e.g. by responding as quickly as possible, finite speed makes finite delays and sharp frequency cutoffs, resulting in instabilities at those frequency cutoff points because of violation of the 12db/octave rule. Of course, this ONLY applies in feedback systems and NOT in forward-only systems, except at the real-world point of feedback, e.g. the bots themselves. Of course, there is the big question of just what it is that is being attenuated in the bowels of an intelligent system. Usually, it is computational delays making sharp frequency-limited attenuation at their response speeds. Every gamer is well aware of the oscillations that long ping times can introduce in people's (and intelligent bot's) behavior. Again, this is basically the same 12db/octave phenomenon. Steve --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] A fundamental limit on intelligence?!
Steve: For example, based on ability to follow instruction, cats must be REALLY stupid. Either that or really smart. Who wants to obey some dumb human's instructions? --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] A fundamental limit on intelligence?!
Isn't this the argument for GAs running on multicored processors? Now each organism has one core/fraction of a core. The brain will then evaluate * fitness* having a fitness criterion. The fact they can be run efficiently in parallel is one of the advantages of GAs. Let us look at this another way, when an intelligent person thinks about a problem, they will think about it in terms of a set of alternatives. This could be said to be the start of genetic reasoning. So it does in fact take place now. A GA is the simplest parallel system which you can think of for purposes of illustration. However when we answer *Jeopardy* type questions parallelism is involved. This becomes clear when we look at how Watson actually works.http://www.nytimes.com/2010/06/20/magazine/20Computer-t.html It works in parallel and then finds the most probable answer. - Ian Parker - Ian Parker On 21 June 2010 16:38, Abram Demski abramdem...@gmail.com wrote: Steve, You didn't mention this, so I guess I will: larger animals do generally have larger brains, coming close to a fixed brain/body ratio. Smarter animals appear to be the ones with a higher brain/body ratio rather than simply a larger brain. This to me suggests that the amount of sensory information and muscle coordination necessary is the most important determiner of the amount of processing power needed. There could be other interpretations, however. It's also pretty important to say that brains are expensive to fuel. It's probably the case that other animals didn't get as smart as us because the additional food they could get per ounce brain was less than the additional food needed to support an ounce of brain. Humans were in a situation in which it was more. So, I don't think your argument from other animals supports your hypothesis terribly well. One way around your instability if it exists would be (similar to your hemisphere suggestion) split the network into a number of individuals which cooperate through very low-bandwidth connections. This would be like an organization of humans working together. Hence, multiagent systems would have a higher stability limit. However, it is still the case that we hit a serious diminishing-returns scenario once we needed to start doing this (since the low-bandwidth connections convey so much less info, we need waaay more processing power for every IQ point or whatever). And, once these organizations got really big, it's quite plausible that they'd have their own stability issues. --Abram On Mon, Jun 21, 2010 at 11:19 AM, Steve Richfield steve.richfi...@gmail.com wrote: There has been an ongoing presumption that more brain (or computer) means more intelligence. I would like to question that underlying presumption. That being the case, why don't elephants and other large creatures have really gigantic brains? This seems to be SUCH an obvious evolutionary step. There are all sorts of network-destroying phenomena that rise from complex networks, e.g. phase shift oscillators there circular analysis paths enforce themselves, computational noise is endlessly analyzed, etc. We know that our own brains are just barely stable, as flashing lights throw some people into epileptic attacks, etc. Perhaps network stability is the intelligence limiter? If so, then we aren't going to get anywhere without first fully understanding it. Suppose for a moment that theoretically perfect neurons could work in a brain of limitless size, but their imperfections accumulate (or multiply) to destroy network operation when you get enough of them together. Brains have grown larger because neurons have evolved to become more nearly perfect, without having yet (or ever) reaching perfection. Hence, evolution may have struck a balance, where less intelligence directly impairs survivability, and greater intelligence impairs network stability, and hence indirectly impairs survivability. If the above is indeed the case, then AGI and related efforts don't stand a snowball's chance in hell of ever outperforming humans, UNTIL the underlying network stability theory is well enough understood to perform perfectly to digital precision. This wouldn't necessarily have to address all aspects of intelligence, but would at minimum have to address large-scale network stability. One possibility is chopping large networks into pieces, e.g. the hemispheres of our own brains. However, like multi-core CPUs, there is work for only so many CPUs/hemispheres. There are some medium-scale network similes in the world, e.g. the power grid. However, there they have high-level central control and lots of crashes, so there may not be much to learn from them. Note in passing that I am working with some non-AGIers on power grid stability issues. While not fully understood, the primary challenge appears (to me) to be that the various control mechanisms (that includes humans in the loop) violate a basic requirement
Re: [agi] High Frame Rates Reduce Uncertainty
My comment is this. The brain in fact takes whatever speed it needs. For simple processing it takes the full speed. More complex processing does not require the same speed and so is taken more slowly. This is really an extension of what DESTIN does spatially. - Ian Parker On 21 June 2010 15:30, deepakjnath deepakjn...@gmail.com wrote: The brain does not get the high frame rate signals as the eye itself only gives brain images at 24 frames per second. Else u wouldn't be able to watch a movie. Any comments? --- 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/?; Powered by Listbox: http://www.listbox.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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
RE: [agi] A fundamental limit on intelligence?!
-Original Message- From: Steve Richfield [mailto:steve.richfi...@gmail.com] Really? Do networks such as botnets really care about this? Or does it apply? Anytime negative feedback can become positive feedback because of delays or phase shifts, this becomes an issue. Many competent EE people fail to see the phase shifting that many decision processes can introduce, e.g. by responding as quickly as possible, finite speed makes finite delays and sharp frequency cutoffs, resulting in instabilities at those frequency cutoff points because of violation of the 12db/octave rule. Of course, this ONLY applies in feedback systems and NOT in forward-only systems, except at the real-world point of feedback, e.g. the bots themselves. Of course, there is the big question of just what it is that is being attenuated in the bowels of an intelligent system. Usually, it is computational delays making sharp frequency-limited attenuation at their response speeds. Every gamer is well aware of the oscillations that long ping times can introduce in people's (and intelligent bot's) behavior. Again, this is basically the same 12db/octave phenomenon. OK, excuse my ignorance on this - a design issue in distributed intelligence is how to split up things amongst the agents. I see it as a hierarchy of virtual networks, with the lowest level being the substrate like IP sockets or something else but most commonly TCP/UDP. The protocols above that need to break up the work, and the knowledge distribution, so the 12db/octave phenomenon must apply there too. I assume any intelligence processing engine must include a harmonic mathematical component since ALL things are basically network, especially intelligence. This might be an overly aggressive assumption but it seems from observance that intelligence/consciousness exhibits some sort of harmonic property, or levels. John --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] An alternative plan to discover self-organization theory
Matt, I'm not sure I buy that argument for the simple reason that we have massive cheap processing now and pretty good knowledge of the initial conditions of life on our planet (if we are going literal here and not EC in the abstract), but it's definitely a possible answer. Perhaps not enough people have attempted to run evolutionary computation experiments at these massive scales either. Rob On Mon, Jun 21, 2010 at 12:59 PM, Matt Mahoney matmaho...@yahoo.com wrote: rob levy wrote: On a related note, what is everyone's opinion on why evolutionary algorithms are such a miserable failure as creative machines, despite their successes in narrow optimization problems? Lack of computing power. How much computation would you need to simulate the 3 billion years of evolution that created human intelligence? -- Matt Mahoney, matmaho...@yahoo.com -- *From:* rob levy r.p.l...@gmail.com *To:* agi agi@v2.listbox.com *Sent:* Mon, June 21, 2010 11:56:53 AM *Subject:* Re: [agi] An alternative plan to discover self-organization theory (I'm a little late in this conversation. I tried to send this message the other day but I had my list membership configured wrong. -Rob) -- Forwarded message -- From: rob levy r.p.l...@gmail.com Date: Sun, Jun 20, 2010 at 5:48 PM Subject: Re: [agi] An alternative plan to discover self-organization theory To: agi@v2.listbox.com On a related note, what is everyone's opinion on why evolutionary algorithms are such a miserable failure as creative machines, despite their successes in narrow optimization problems? I don't want to conflate the possibly separable problems of biological development and evolution, though they are interrelated. There are various approaches to evolutionary theory such as Lima de Faria's evolution without selection ideas and Reid's evolution by natural experiment that suggest natural selection is not all it's cracked up to be, and that the step of generating, (mutating, combining, ) is where the more interesting stuff happens. Most of the alternatives to Neodarwinian Synthesis I have seen are based in dynamic models of emergence in complex systems. The upshot is, you don't get creativity for free, you actually still need to solve a problem that is as hard as AGI in order to get creativity for free. So, you would need to solve the AGI-hard problem of evolution and development of life, in order to then solve AGI itself (reminds me of the old SNL sketch: first, get a million dollars...). Also, my hunch is that there is quite a bit of overlap between the solutions to the two problems. Rob Disclaimer: I'm discussing things above that I'm not and don't claim to be an expert in, but from what I have seen so far on this list, that should be alright. AGI is by its nature very multidisciplinary which necessitates often being breadth-first, and therefore shallow in some areas. On Sun, Jun 20, 2010 at 2:06 AM, Steve Richfield steve.richfi...@gmail.com wrote: No, I haven't been smokin' any wacky tobacy. Instead, I was having a long talk with my son Eddie, about self-organization theory. This is *his*proposal: He suggested that I construct a simple NN that couldn't work without self organizing, and make dozens/hundreds of different neuron and synapse operational characteristics selectable ala genetic programming, put it on the fastest computer I could get my hands on, turn it loose trying arbitrary combinations of characteristics, and see what the winning combination turns out to be. Then, armed with that knowledge, refine the genetic characteristics and do it again, and iterate until it *efficiently* self organizes. This might go on for months, but self-organization theory might just emerge from such an effort. I had a bunch of objections to his approach, e.g. Q. What if it needs something REALLY strange to work? A. Who better than you to come up with a long list of really strange functionality? Q. There are at least hundreds of bits in the genome. A. Try combinations in pseudo-random order, with each bit getting asserted in ~half of the tests. If/when you stumble onto a combination that sort of works, switch to varying the bits one-at-a-time, and iterate in this way until the best combination is found. Q. Where are we if this just burns electricity for a few months and finds nothing? A. Print out the best combination, break out the wacky tobacy, and come up with even better/crazier parameters to test. I have never written a line of genetic programming, but I know that others here have. Perhaps you could bring some rationality to this discussion? What would be a simple NN that needs self-organization? Maybe a small pot of neurons that could only work if they were organized into layers, e.g. a simple 64-neuron system that would work as a 4x4x4-layer visual recognition system, given the input that I fed it? Any thoughts
Re: [agi] An alternative plan to discover self-organization theory
Rob, Real evolution had full freedom to evolve. Genetic algorithms usually don't. If they did, the number of calculations it would have to make to really simulate evolution on the scale that created us would be so astronomical, it would not be possible. So, what matt said is absolutely correct. There probably isn't enough processing power in the world to do what real evolution did and there probably never will be. So, lets say that you restrict the genetic algorithm. Well, now it doesn't have the freedom to find the right solution. You may think you've given it enough freedom, but most likely you have not. If you do give it enough freedom, its likely to take all eternity to find a solution to many, if not most, real life problems. Dave On Mon, Jun 21, 2010 at 3:15 PM, rob levy r.p.l...@gmail.com wrote: Matt, I'm not sure I buy that argument for the simple reason that we have massive cheap processing now and pretty good knowledge of the initial conditions of life on our planet (if we are going literal here and not EC in the abstract), but it's definitely a possible answer. Perhaps not enough people have attempted to run evolutionary computation experiments at these massive scales either. Rob On Mon, Jun 21, 2010 at 12:59 PM, Matt Mahoney matmaho...@yahoo.comwrote: rob levy wrote: On a related note, what is everyone's opinion on why evolutionary algorithms are such a miserable failure as creative machines, despite their successes in narrow optimization problems? Lack of computing power. How much computation would you need to simulate the 3 billion years of evolution that created human intelligence? -- Matt Mahoney, matmaho...@yahoo.com -- *From:* rob levy r.p.l...@gmail.com *To:* agi agi@v2.listbox.com *Sent:* Mon, June 21, 2010 11:56:53 AM *Subject:* Re: [agi] An alternative plan to discover self-organization theory (I'm a little late in this conversation. I tried to send this message the other day but I had my list membership configured wrong. -Rob) -- Forwarded message -- From: rob levy r.p.l...@gmail.com Date: Sun, Jun 20, 2010 at 5:48 PM Subject: Re: [agi] An alternative plan to discover self-organization theory To: agi@v2.listbox.com On a related note, what is everyone's opinion on why evolutionary algorithms are such a miserable failure as creative machines, despite their successes in narrow optimization problems? I don't want to conflate the possibly separable problems of biological development and evolution, though they are interrelated. There are various approaches to evolutionary theory such as Lima de Faria's evolution without selection ideas and Reid's evolution by natural experiment that suggest natural selection is not all it's cracked up to be, and that the step of generating, (mutating, combining, ) is where the more interesting stuff happens. Most of the alternatives to Neodarwinian Synthesis I have seen are based in dynamic models of emergence in complex systems. The upshot is, you don't get creativity for free, you actually still need to solve a problem that is as hard as AGI in order to get creativity for free. So, you would need to solve the AGI-hard problem of evolution and development of life, in order to then solve AGI itself (reminds me of the old SNL sketch: first, get a million dollars...). Also, my hunch is that there is quite a bit of overlap between the solutions to the two problems. Rob Disclaimer: I'm discussing things above that I'm not and don't claim to be an expert in, but from what I have seen so far on this list, that should be alright. AGI is by its nature very multidisciplinary which necessitates often being breadth-first, and therefore shallow in some areas. On Sun, Jun 20, 2010 at 2:06 AM, Steve Richfield steve.richfi...@gmail.com wrote: No, I haven't been smokin' any wacky tobacy. Instead, I was having a long talk with my son Eddie, about self-organization theory. This is *his*proposal: He suggested that I construct a simple NN that couldn't work without self organizing, and make dozens/hundreds of different neuron and synapse operational characteristics selectable ala genetic programming, put it on the fastest computer I could get my hands on, turn it loose trying arbitrary combinations of characteristics, and see what the winning combination turns out to be. Then, armed with that knowledge, refine the genetic characteristics and do it again, and iterate until it *efficiently* self organizes. This might go on for months, but self-organization theory might just emerge from such an effort. I had a bunch of objections to his approach, e.g. Q. What if it needs something REALLY strange to work? A. Who better than you to come up with a long list of really strange functionality? Q. There are at least hundreds of bits in the genome. A. Try combinations in pseudo-random order, with each bit getting asserted in
Re: [agi] A fundamental limit on intelligence?!
On Mon, Jun 21, 2010 at 4:19 PM, Steve Richfield steve.richfi...@gmail.com wrote: That being the case, why don't elephants and other large creatures have really gigantic brains? This seems to be SUCH an obvious evolutionary step. Personally I've always wondered how elephants managed to evolve brains as large as they currently have. How much intelligence does it take to sneak up on a leaf? (Granted, intraspecies social interactions seem to provide at least part of the answer.) There are all sorts of network-destroying phenomena that rise from complex networks, e.g. phase shift oscillators there circular analysis paths enforce themselves, computational noise is endlessly analyzed, etc. We know that our own brains are just barely stable, as flashing lights throw some people into epileptic attacks, etc. Perhaps network stability is the intelligence limiter? Empirically, it isn't. Suppose for a moment that theoretically perfect neurons could work in a brain of limitless size, but their imperfections accumulate (or multiply) to destroy network operation when you get enough of them together. Brains have grown larger because neurons have evolved to become more nearly perfect Actually it's the other way around. Brains compensate for imperfections (both transient error and permanent failure) in neurons by using more of them. Note that, as the number of transistors on a silicon chip increases, the extent to which our chip designs do the same thing also increases. There are some medium-scale network similes in the world, e.g. the power grid. However, there they have high-level central control and lots of crashes The power in my neighborhood fails once every few years (and that's from all causes, including 'the cable guys working up the street put a JCB through the line', not just network crashes). If you're getting lots of power failures in your neighborhood, your electricity supply company is doing something wrong. I wonder, does the very-large-scale network problem even have a prospective solution? Is there any sort of existence proof of this? Yes, our repeated successes in simultaneously improving both the size and stability of very large scale networks (trade, postage, telegraph, electricity, road, telephone, Internet) serve as very nice existence proofs. --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
[agi] Read Fast, Trade Fast
http://www.zerohedge.com/article/fast-reading-computers-are-about-drink-your-trading-milkshake --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] A fundamental limit on intelligence?!
Russell, On Mon, Jun 21, 2010 at 1:29 PM, Russell Wallace russell.wall...@gmail.comwrote: On Mon, Jun 21, 2010 at 4:19 PM, Steve Richfield steve.richfi...@gmail.com wrote: That being the case, why don't elephants and other large creatures have really gigantic brains? This seems to be SUCH an obvious evolutionary step. Personally I've always wondered how elephants managed to evolve brains as large as they currently have. How much intelligence does it take to sneak up on a leaf? (Granted, intraspecies social interactions seem to provide at least part of the answer.) I suspect that intra-specie social behavior will expand to utilize all available intelligence. There are all sorts of network-destroying phenomena that rise from complex networks, e.g. phase shift oscillators there circular analysis paths enforce themselves, computational noise is endlessly analyzed, etc. We know that our own brains are just barely stable, as flashing lights throw some people into epileptic attacks, etc. Perhaps network stability is the intelligence limiter? Empirically, it isn't. I see what you are saying, but I don't think you have made your case... Suppose for a moment that theoretically perfect neurons could work in a brain of limitless size, but their imperfections accumulate (or multiply) to destroy network operation when you get enough of them together. Brains have grown larger because neurons have evolved to become more nearly perfect Actually it's the other way around. Brains compensate for imperfections (both transient error and permanent failure) in neurons by using more of them. William Calvin, the author who is most credited with making and spreading this view, and I had a discussion on his Seattle rooftop, while throwing pea gravel at a target planter. His assertion was that we utilize many parallel circuits to achieve accuracy, and mine was that it was something else, e.g. successive approximation. I pointed out that if one person tossed the pea gravel by putting it on their open hand and pushing it at a target, and the other person blocked their arm, that the relationship between how much of the stroke was truncated and how great the error was would disclose the method of calculation. The question boils down to the question of whether the error grows drastically even with small truncation of movement (because a prototypical throw is used, as might be expected from a parallel approach), or grows exponentially because error correcting steps have been lost. We observed apparent exponential growth, much smaller than would be expected from parallel computation, though no one was keeping score. In summary, having performed the above experiment, I reject this common view. Note that, as the number of transistors on a silicon chip increases, the extent to which our chip designs do the same thing also increases. Another pet peeve of mine. They could/should do MUCH more fault tolerance than they now are. Present puny efforts are completely ignorant of past developments, e.g. Tandem Nonstop computers. There are some medium-scale network similes in the world, e.g. the power grid. However, there they have high-level central control and lots of crashes The power in my neighborhood fails once every few years (and that's from all causes, including 'the cable guys working up the street put a JCB through the line', not just network crashes). If you're getting lots of power failures in your neighborhood, your electricity supply company is doing something wrong. If you look at the failures/bandwidth, it is pretty high. The point is that the information bandwidth of the power grid is EXTREMELY low, so it shouldn't fail at all, at least not more than maybe once per century. However, just like the May 6 problem, it sometimes gets itself into trouble of its own making. Any overload SHOULD simply result in shutting down some low-priority load, like the heaters in steel plants, and this usually works as planned. However, it sometimes fails for VERY complex reasons - so complex that PhD engineers are unable to put it into words, despite having millisecond-by-millisecond histories to work from. I wonder, does the very-large-scale network problem even have a prospective solution? Is there any sort of existence proof of this? Yes, our repeated successes in simultaneously improving both the size and stability of very large scale networks (trade, NOT stable at all. Just look at the condition of the world's economy. postage, telegraph, electricity, road, telephone, Internet) None of these involve feedback, the fundamental requirement to be a network rather than a simple tree structure. This despite common misuse of the term network to cover everything with lots of interconnections. serve as very nice existence proofs. I'm still looking. Steve --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed:
Re: [agi] A fundamental limit on intelligence?!
On Mon, Jun 21, 2010 at 11:05 PM, Steve Richfield steve.richfi...@gmail.com wrote: Another pet peeve of mine. They could/should do MUCH more fault tolerance than they now are. Present puny efforts are completely ignorant of past developments, e.g. Tandem Nonstop computers. Or perhaps they just figure once the mean time between failure is on the order of, say, a year, customers aren't willing to pay much for further improvement. (Note that things like financial databases which still have difficulty scaling horizontally, do get more fault tolerance than an ordinary PC. Note also that they pay a hefty premium for this, more than you or I would be willing or able to pay.) The power in my neighborhood fails once every few years (and that's from all causes, including 'the cable guys working up the street put a JCB through the line', not just network crashes). If you're getting lots of power failures in your neighborhood, your electricity supply company is doing something wrong. If you look at the failures/bandwidth, it is pretty high. So what? Nobody except you cares about that metric. Anyway, the phone system is in the same league, and the Internet is a lot closer to it than it was in the past, and those have vastly higher bandwidth. Yes, our repeated successes in simultaneously improving both the size and stability of very large scale networks (trade, NOT stable at all. Just look at the condition of the world's economy. Better than it was in the 1930s, despite a lot greater complexity. postage, telegraph, electricity, road, telephone, Internet) None of these involve feedback, the fundamental requirement to be a network rather than a simple tree structure. This despite common misuse of the term network to cover everything with lots of interconnections. All of them involve massive amounts of feedback. Unless you're adopting a private definition of the word feedback, in which case by your private definition, if it is to be at all consistent, neither brains nor computers running AI programs will involve feedback either, so it's immaterial. --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] A fundamental limit on intelligence?!
John, Hmmm, I though that with your EE background, that the 12db/octave would bring back old sophomore-level course work. OK, so you were sick that day. I'll try to fill in the blanks here... On Mon, Jun 21, 2010 at 11:16 AM, John G. Rose johnr...@polyplexic.comwrote: Of course, there is the big question of just what it is that is being attenuated in the bowels of an intelligent system. Usually, it is computational delays making sharp frequency-limited attenuation at their response speeds. Every gamer is well aware of the oscillations that long ping times can introduce in people's (and intelligent bot's) behavior. Again, this is basically the same 12db/octave phenomenon. OK, excuse my ignorance on this - a design issue in distributed intelligence is how to split up things amongst the agents. I see it as a hierarchy of virtual networks, with the lowest level being the substrate like IP sockets or something else but most commonly TCP/UDP. The protocols above that need to break up the work, and the knowledge distribution, so the 12db/octave phenomenon must apply there too. RC low-pass circuits exhibit 6db/octave rolloff and 90 degree phase shifts. 12db/octave corresponds to a 180 degree phase shift. More than 180 degrees and you are into positive feedback. At 24db/octave, you are at maximum * positive* feedback, which makes great oscillators. The 12 db/octave limit applies to entire loops of components, and not to the individual components. This means that you can put a lot of 1db/octave components together in a big loop and get into trouble. This is commonly encountered in complex analog filter circuits that incorporate 2 or more op-amps in a single feedback loop. Op amps are commonly compensated to have 6db/octave rolloff. Put 2 of them together and you right at the precipice of 12db/octave. Add some passive components that have their own rolloffs, and you are over the edge of stability, and the circuit sits there and oscillates on its own. The usual cure is to replace one of the op-amps with an *un*compensated op-amp with ~0db/octave rolloff, until it gets to its maximum frequency, whereupon it has an astronomical rolloff. However, that astronomical rolloff works BECAUSE the loop gain at that frequency is less than 1, so the circuit cannot self-regenerate and oscillate at that frequency. Considering the above and the complexity of neural circuits, it would seem that neural circuits would have to have absolutely flat responses and some central rolloff mechanism, maybe one of the ~200 different types of neurons, or alternatively, would have to be able to custom-tailor their responses to work in concert to roll off at a reasonable rate. A third alternative is discussed below, where you let them go unstable, and actually utilize the instability to achieve some incredible results. I assume any intelligence processing engine must include a harmonic mathematical component I'm not sure I understand what you are saying here. Perhaps you have discovered the recipe for the secret sauce? since ALL things are basically network, especially intelligence. Most of the things we call networks really just pass information along and do NOT have feedback mechanisms. Power control is an interesting exception, but most of those guys are unable to even carry on an intelligent conversation about the subject. No wonder the power networks have problems. This might be an overly aggressive assumption but it seems from observance that intelligence/consciousness exhibits some sort of harmonic property, or levels. You apparently grok something about harmonics that I don't (yet) grok. Please enlighten me. Are you familiar with regenerative receiver operation where operation is on the knife-edge of instability, or super-regenerative receiver operation, wherein an intentionally UNstable circuit is operated to achieve phenomenal gain and specifically narrow bandwidth? These were common designs back in the early vacuum tube era, when active components cost a day's wages. Given all of the observed frequency components coming from neural circuits, perhaps neurons do something similar to actually USE instability to their benefit?! Is this related to your harmonic thoughts? Thanks. Steve --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] High Frame Rates Reduce Uncertainty
My view on the efficiency of the brain's learning has to do with low latency communications in general, which is a similar concept to high frame rates, but not limited to the visual senses. Low latency produces rapid feedback. Rapid feedback produces rapid adaptation, and the reduced weight of the time axis in the formula results in a reduction of short-term memory loss and thus more resources for the brain to work with. --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] High Frame Rates Reduce Uncertainty
That should be a reduction of the penalty caused from short-term memory loss. On Mon, Jun 21, 2010 at 4:20 PM, Mark Nuzzolilo nuzz...@gmail.com wrote: My view on the efficiency of the brain's learning has to do with low latency communications in general, which is a similar concept to high frame rates, but not limited to the visual senses. Low latency produces rapid feedback. Rapid feedback produces rapid adaptation, and the reduced weight of the time axis in the formula results in a reduction of short-term memory loss and thus more resources for the brain to work with. --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
Re: [agi] High Frame Rates Reduce Uncertainty
Hi, * AGI should be scalable - More data just mean the potential for more accurate results. * More data can chew up more computation time without a benefit. ie If all you want to do is identify a bird, it's still a bird at 1 fps and 1000 fps. * Don't aim for precision, aim for generality. Eg. AGI KNOWS 1000 objects. If you test to see if your object is a bird, and it is not, you still have 999 possible objects. If you test if it is an animal, you can split your search space in half - you've reduce the possibilities to 500. Successive generalisation produce accuracy, sometimes referred as a hierarchical approach. On Fri, 2010-06-18 at 14:19 -0400, David Jones wrote: I just came up with an awesome idea. I just realized that the brain takes advantage of high frame rates to reduce uncertainty when it is estimating motion. The slower the frame rate, the more uncertainty there is because objects may have traveled too far between images to match with high certainty using simple techniques. So, this made me think, what if the secret to the brain's ability to learn generally stems from this high frame rate trick. What if we made a system that could process even high frame rates than the brain can. By doing this you can reduce the uncertainty of matches very very low (well in my theory so far). If you can do that, then you can learn about the objects in a video, how they move together or separately with very high certainty. You see, matching is the main barrier when learning about objects. But with a very high frame rate, we can use a fast algorithm and could potentially reduce the uncertainty to almost nothing. Once we learn about objects, matching gets easier because now we have training data and experience to take advantage of. In addition, you can also gain knowledge about lighting, color variation, noise, etc. With that knowledge, you can then automatically create a model of the object with extremely high confidence. You will also be able to determine the effects of light and noise on the object's appearance, which will help match the object invariantly in the future. It allows you to determine what is expected and unexpected for the object's appearance with much higher confidence. Pretty cool idea huh? Dave agi | Archives | Modify Your Subscription --- 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=8660244-6e7fb59c Powered by Listbox: http://www.listbox.com
RE: [agi] A fundamental limit on intelligence?!
-Original Message- From: Steve Richfield [mailto:steve.richfi...@gmail.com] John, Hmmm, I though that with your EE background, that the 12db/octave would bring back old sophomore-level course work. OK, so you were sick that day. I'll try to fill in the blanks here... Thanks man. Appreciate it. What little EE training I did undergo was brief and painful :) On Mon, Jun 21, 2010 at 11:16 AM, John G. Rose johnr...@polyplexic.com wrote: Of course, there is the big question of just what it is that is being attenuated in the bowels of an intelligent system. Usually, it is computational delays making sharp frequency-limited attenuation at their response speeds. Every gamer is well aware of the oscillations that long ping times can introduce in people's (and intelligent bot's) behavior. Again, this is basically the same 12db/octave phenomenon. OK, excuse my ignorance on this - a design issue in distributed intelligence is how to split up things amongst the agents. I see it as a hierarchy of virtual networks, with the lowest level being the substrate like IP sockets or something else but most commonly TCP/UDP. The protocols above that need to break up the work, and the knowledge distribution, so the 12db/octave phenomenon must apply there too. RC low-pass circuits exhibit 6db/octave rolloff and 90 degree phase shifts. 12db/octave corresponds to a 180 degree phase shift. More than 180 degrees and you are into positive feedback. At 24db/octave, you are at maximum positive feedback, which makes great oscillators. The 12 db/octave limit applies to entire loops of components, and not to the individual components. This means that you can put a lot of 1db/octave components together in a big loop and get into trouble. This is commonly encountered in complex analog filter circuits that incorporate 2 or more op- amps in a single feedback loop. Op amps are commonly compensated to have 6db/octave rolloff. Put 2 of them together and you right at the precipice of 12db/octave. Add some passive components that have their own rolloffs, and you are over the edge of stability, and the circuit sits there and oscillates on its own. The usual cure is to replace one of the op-amps with an uncompensated op-amp with ~0db/octave rolloff, until it gets to its maximum frequency, whereupon it has an astronomical rolloff. However, that astronomical rolloff works BECAUSE the loop gain at that frequency is less than 1, so the circuit cannot self-regenerate and oscillate at that frequency. Considering the above and the complexity of neural circuits, it would seem that neural circuits would have to have absolutely flat responses and some central rolloff mechanism, maybe one of the ~200 different types of neurons, or alternatively, would have to be able to custom-tailor their responses to work in concert to roll off at a reasonable rate. A third alternative is discussed below, where you let them go unstable, and actually utilize the instability to achieve some incredible results. I assume any intelligence processing engine must include a harmonic mathematical component I'm not sure I understand what you are saying here. Perhaps you have discovered the recipe for the secret sauce? Uhm, no I was merely asking your opinion if the 12db/octave phenomena applies to a non-EE based intelligence system. If it could be lifted off of its EE nativeness and applied to ANY network since there are latencies in ALL networks. BUT it sounds as if it is heavily analog circuit based, though there may be some *analogue in an informational network. And this would be represented under a different technical name or formula most likely. since ALL things are basically network, especially intelligence. Most of the things we call networks really just pass information along and do NOT have feedback mechanisms. Power control is an interesting exception, but most of those guys are unable to even carry on an intelligent conversation about the subject. No wonder the power networks have problems. Steve - I actually did work in nuclear power engineering many years ago and remember the Neanderthals involved in that situation believe it or not. But I will say they strongly emphasized practicality and safety verses theoretics and academics. And especially trial and error was something to be frowned upon ... for obvious reasons. IOW, do not rock the boat since there are real reasons for them being that way! This might be an overly aggressive assumption but it seems from observance that intelligence/consciousness exhibits some sort of harmonic property, or levels. You apparently grok something about harmonics that I don't (yet) grok. Please enlighten me. I was wondering if YOU could envision a harmonic correlation between certain electrical circuit phenomenon and intelligence. I've just suspected that there are harmonic properties in intelligence/consciousness. IOW there