Re: [agi] Reverse Engineering The Brain
J Storrs Hall, PhD wrote: basically on the right track -- except there isn't just one cognitive level. Are you thinking of working out the function of each topographically mapped area a la DNF? Each column in a Darwin machine a la Calvin? Conscious-level symbols a la Minsky? Of course you can make finer distinctions, and different people use the term cognitive in different ways. My usage of the term is coextensive with the usage in cognitive science and cognitive psychology, but that covers a multitude of sins. To the extent that an approach tries to embrace what is known about human cognition it would be cognitive, but if it took little notice of that, it would not. Regular AI does not take much account of human cognition. Neuroscience (even 'cognitive' or 'computational' neuroscience) takes a very superficial attitude toward all things cognitive, even when it says that it is doing otherwise (a sore point in the literature, right now). But anything that takes significant account of cognition is very different from an approach that involves scanning a brain and trying to make a copy without understanding exactly how it works. It is that enormous gap that I was pointing to, and the fact that there are many different ways of taking a significant account of cognition does not make much difference to that gap. Richard Loosemore On Thursday 05 June 2008 09:37:00 pm, Richard Loosemore wrote: There seems to be a good deal of confusion (on this list and also over on the Singularity list) about what people actually mean when they talk about building an AGI by emulating or copying the brain. There are two completely different types of project that seem to get conflated in these discussions: 1) Copying the brain at the neural level, which is usually assumed to be a 'blind' copy - in other words, we will not know how it works, but will just do a complete copy and fire it up. 2) Copying the design of the human brain at the cognitive level. This may involve a certain amount of neuroscience, but mostly it will be at the cognitive system level, and could be done without much reference to neurons at all. Both of these ideas are very different from standard AI, but they are also very different from one another. The criticisms that can be leveled against the neural-copy approach do not apply to the cognitive approach, for example. It is frustrating to see commentaries that drift back and forth between these two. My own position is that a cognitive-level copy is not just feasible but well under way, whereas the idea of duplicating the neural level is just a pie-in-the-sky fantasy at this point in time (it is not possible with current or on-the-horizon technology, and will probably not be possible until after we invent an AGI by some other means and get it to design, build and control a nanotech brain scanning machine). Duplicating a system as complex as that *without* first understanding it at the functional level seems pure folly: one small error in the mapping and the result could be something that simply does not work ... and then, faced with a brain-copy that needs debugging, what would we do? The best we could do is start another scan and hope for better luck next time. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Reverse Engineering The Brain
Steve Richfield wrote: Richard, On 6/5/08, *Richard Loosemore* [EMAIL PROTECTED] mailto:[EMAIL PROTECTED] wrote: There are two completely different types of project that seem to get conflated in these discussions: 1) Copying the brain at the neural level, which is usually assumed to be a 'blind' copy - in other words, we will not know how it works, but will just do a complete copy and fire it up. I suspect that we will have to learn a LOT more to be able to make something like this work, in part because we will need new theory in order to compute parameters that we cannot directly measure. 1.5) Combining scanned information with mathematical constraints to produce diagrams of perfect neurons, even though the precise parameters of the real-world neurons is not fully scannable. What scanned information? What mathematical constraints? What 'perfect' neurons? The problem is that all of this requires you to do work to understand how the system is functioning, because you cannot do something like build a 'perfect' neuron unless you know what its functional role is, and to do that you need to go right up into the high-level description of the system and that means, in the end, that you have to do the entire 'cognitive level' description of the brain *first*, then use it to understand how neurons are being used (what functional role they are playing). For example: does the precise morphology of the dendritic tree matter to the functioning of the neuron? Do you need to scan this information in in complete detail? I don't think you are going to be able to answer this question until after you have understood how the signals exchanged by neurons are being used (high level stuff). Let me try to explain with an analogy. You are duplicating a space shuttle without understanding how it works. You want to know if you can use chewing gum for O-ring seals. Chewing gum is great, although it does become hard and brittle and very brittle in cold weather... but since you do not know what functional role these O-ring seals are playing in the design of the whole system, you decide that maybe it is okay to use chewing gum. So, I don't disagree that there could be a 1.5 approach, but I see now way that it is significantly different from the 2 approach. 2) Copying the design of the human brain at the cognitive level. This may involve a certain amount of neuroscience, but mostly it will be at the cognitive system level, and could be done without much reference to neurons at all. The last 40 years of fruitless AI shows this to be pretty much of a dead end. There is simply too many questions that we don't even know enough to ask. This is not true. The last 40 years of AI have been almost completely unrelated to this 'cognitive' approach. Over the years, the vast majority of AI researchers have subscribed to the following credo: We intend to build an intelligent system, but although we might take some ideas or inspiration from how the human mind works, we feel no obligation to copy the human design because we believe that intelligence does not have to be done that way. I was specifically drawing a distinction between two different ways to build an intelligence in a way that stays close to the human design. The regular AI approach is neither of these two. 2.5) First understanding how we think with neurons, program computers to perform the same or better directly, without reference to neurons or their equivalents. This misses the point. Cognitive level approaches do not have to reduce anything to neurons (at least, not in a significant way), so starting with understanding how we think with neurons doesn't make much sense. If you leave out the specific reference to neurons, what you have is the cognitive level again. Both of these ideas are very different from standard AI, but they are also very different from one another. The criticisms that can be leveled against the neural-copy approach do not apply to the cognitive approach, for example. My more real 1.5 and 2.5 proposals require nearly the same levels of understanding, and ultimately lead to very similar results as simulation gives way via optimization to the same sort of code as direct AGI programming would utilize. In short, I suspect that both paths will ultimately lead to approximately the same final result. Sure we can argue about which path is best, but easiest wins usually rules. You are not addressing the distinction that I made, though. It is frustrating to see commentaries that drift back and forth between these two. My own position is that a cognitive-level copy is not just feasible but well under way, whereas the idea of duplicating the neural level is just a pie-in-the-sky fantasy at this point in time (it is not possible with current or on-the-horizon technology, and will
Re: [agi] Reverse Engineering The Brain
--- On Thu, 6/5/08, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: http://www.spectrum.ieee.org/print/6268 Some rough calculations. A human brain has a volume of 10^24 nm^3. A scan of 5 x 5 x 50 nm voxels requires about 1000 exabytes = 10^21 bytes of storage (1 MB per synapse). A scan would take a 10 GHz SEM 10^11 seconds = 3000 years, or equivalently, 1 year for 3000 scanning electron microscopes running in parallel. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Reverse Engineering The Brain
Or, assuming we decided to spend the same on that as on the Iraq war ($1 trillion: http://www.boston.com/news/nation/articles/2007/08/01/analysis_says_war_could_cost_1_trillion/), at $1 million per scope and associated lab costs, giving a million scopes == 10^5 sec = 28 hours. Which is more important? On Thursday 05 June 2008 03:44:14 pm, Matt Mahoney wrote: --- On Thu, 6/5/08, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: http://www.spectrum.ieee.org/print/6268 Some rough calculations. A human brain has a volume of 10^24 nm^3. A scan of 5 x 5 x 50 nm voxels requires about 1000 exabytes = 10^21 bytes of storage (1 MB per synapse). A scan would take a 10 GHz SEM 10^11 seconds = 3000 years, or equivalently, 1 year for 3000 scanning electron microscopes running in parallel. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
RE: [agi] Reverse Engineering The Brain
A very interesting paper. I am glad they are talking in terms of understanding consciousness by reverse engineering the brain. It supports my belief that consciousness results from and is an essential aspect of the type of computation a human mind does. -Original Message- From: J Storrs Hall, PhD [mailto:[EMAIL PROTECTED] Sent: Thursday, June 05, 2008 3:07 PM To: agi@v2.listbox.com Subject: [agi] Reverse Engineering The Brain http://www.spectrum.ieee.org/print/6268 --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
RE: [agi] Reverse Engineering The Brain
Before we spend the money required to reverse engineer the brain --- we should at least spend the much less amount of money necessary to explore the very promising potential of Novamente-like machines running on the equivalent of about 20 million dollars worth of hardware at today's prices. -Original Message- From: J Storrs Hall, PhD [mailto:[EMAIL PROTECTED] Sent: Thursday, June 05, 2008 5:01 PM To: agi@v2.listbox.com Subject: Re: [agi] Reverse Engineering The Brain Or, assuming we decided to spend the same on that as on the Iraq war ($1 trillion: http://www.boston.com/news/nation/articles/2007/08/01/analysis_says_war_coul d_cost_1_trillion/), at $1 million per scope and associated lab costs, giving a million scopes == 10^5 sec = 28 hours. Which is more important? On Thursday 05 June 2008 03:44:14 pm, Matt Mahoney wrote: --- On Thu, 6/5/08, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: http://www.spectrum.ieee.org/print/6268 Some rough calculations. A human brain has a volume of 10^24 nm^3. A scan of 5 x 5 x 50 nm voxels requires about 1000 exabytes = 10^21 bytes of storage (1 MB per synapse). A scan would take a 10 GHz SEM 10^11 seconds = 3000 years, or equivalently, 1 year for 3000 scanning electron microscopes running in parallel. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Reverse Engineering The Brain
There seems to be a good deal of confusion (on this list and also over on the Singularity list) about what people actually mean when they talk about building an AGI by emulating or copying the brain. There are two completely different types of project that seem to get conflated in these discussions: 1) Copying the brain at the neural level, which is usually assumed to be a 'blind' copy - in other words, we will not know how it works, but will just do a complete copy and fire it up. 2) Copying the design of the human brain at the cognitive level. This may involve a certain amount of neuroscience, but mostly it will be at the cognitive system level, and could be done without much reference to neurons at all. Both of these ideas are very different from standard AI, but they are also very different from one another. The criticisms that can be leveled against the neural-copy approach do not apply to the cognitive approach, for example. It is frustrating to see commentaries that drift back and forth between these two. My own position is that a cognitive-level copy is not just feasible but well under way, whereas the idea of duplicating the neural level is just a pie-in-the-sky fantasy at this point in time (it is not possible with current or on-the-horizon technology, and will probably not be possible until after we invent an AGI by some other means and get it to design, build and control a nanotech brain scanning machine). Duplicating a system as complex as that *without* first understanding it at the functional level seems pure folly: one small error in the mapping and the result could be something that simply does not work ... and then, faced with a brain-copy that needs debugging, what would we do? The best we could do is start another scan and hope for better luck next time. Richard Loosemore --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Reverse Engineering The Brain
basically on the right track -- except there isn't just one cognitive level. Are you thinking of working out the function of each topographically mapped area a la DNF? Each column in a Darwin machine a la Calvin? Conscious-level symbols a la Minsky? On Thursday 05 June 2008 09:37:00 pm, Richard Loosemore wrote: There seems to be a good deal of confusion (on this list and also over on the Singularity list) about what people actually mean when they talk about building an AGI by emulating or copying the brain. There are two completely different types of project that seem to get conflated in these discussions: 1) Copying the brain at the neural level, which is usually assumed to be a 'blind' copy - in other words, we will not know how it works, but will just do a complete copy and fire it up. 2) Copying the design of the human brain at the cognitive level. This may involve a certain amount of neuroscience, but mostly it will be at the cognitive system level, and could be done without much reference to neurons at all. Both of these ideas are very different from standard AI, but they are also very different from one another. The criticisms that can be leveled against the neural-copy approach do not apply to the cognitive approach, for example. It is frustrating to see commentaries that drift back and forth between these two. My own position is that a cognitive-level copy is not just feasible but well under way, whereas the idea of duplicating the neural level is just a pie-in-the-sky fantasy at this point in time (it is not possible with current or on-the-horizon technology, and will probably not be possible until after we invent an AGI by some other means and get it to design, build and control a nanotech brain scanning machine). Duplicating a system as complex as that *without* first understanding it at the functional level seems pure folly: one small error in the mapping and the result could be something that simply does not work ... and then, faced with a brain-copy that needs debugging, what would we do? The best we could do is start another scan and hope for better luck next time. Richard Loosemore --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=103754539-40ed26 Powered by Listbox: http://www.listbox.com
Re: [agi] Reverse Engineering The Brain
Richard, On 6/5/08, Richard Loosemore [EMAIL PROTECTED] wrote: There are two completely different types of project that seem to get conflated in these discussions: 1) Copying the brain at the neural level, which is usually assumed to be a 'blind' copy - in other words, we will not know how it works, but will just do a complete copy and fire it up. I suspect that we will have to learn a LOT more to be able to make something like this work, in part because we will need new theory in order to compute parameters that we cannot directly measure. 1.5) Combining scanned information with mathematical constraints to produce diagrams of perfect neurons, even though the precise parameters of the real-world neurons is not fully scannable. 2) Copying the design of the human brain at the cognitive level. This may involve a certain amount of neuroscience, but mostly it will be at the cognitive system level, and could be done without much reference to neurons at all. The last 40 years of fruitless AI shows this to be pretty much of a dead end. There is simply too many questions that we don't even know enough to ask. 2.5) First understanding how we think with neurons, program computers to perform the same or better directly, without reference to neurons or their equivalents. Both of these ideas are very different from standard AI, but they are also very different from one another. The criticisms that can be leveled against the neural-copy approach do not apply to the cognitive approach, for example. My more real 1.5 and 2.5 proposals require nearly the same levels of understanding, and ultimately lead to very similar results as simulation gives way via optimization to the same sort of code as direct AGI programming would utilize. In short, I suspect that both paths will ultimately lead to approximately the same final result. Sure we can argue about which path is best, but easiest wins usually rules. It is frustrating to see commentaries that drift back and forth between these two. My own position is that a cognitive-level copy is not just feasible but well under way, whereas the idea of duplicating the neural level is just a pie-in-the-sky fantasy at this point in time (it is not possible with current or on-the-horizon technology, and will probably not be possible until after we invent an AGI by some other means and get it to design, build and control a nanotech brain scanning machine). There is nothing in the above sentence that I can agree with, from which to state objections to the remainder! Some of it may turn out to be correct, but too little is known and no one is even building the needed lab equipment to determine just WHAT the situation actually is. However, I believe that the whole thinking thing involves processes that no one here will EVER guess without learning more about biological brains - if nothing more than the mathematics of operation. However, your next paragraph asks some of the right questions, showing that sometimes it is possible to get to the correct place, even though the path to there is severely flawed. Duplicating a system as complex as that *without* first understanding it at the functional level seems pure folly: I absolutely agree. So long as there is any sort of unknown mathematics there is no hope. one small error in the mapping and the result could be something that simply does not work ... No, these MUST be correctable. SEM methods are unworkable because of the high disaster rate as slices are often destroyed. However, my scanning UV fluorescence microscope doesn't have such problems because the scanning is all within unsliced bulk brain, then some is sliced off and discarded and scanning within the unsliced bulk brain continues. Further, there will doubtless be parameters that evade scanning. SEM methods trash the complex molecules that underlie neural function, and so have no hope of success. However, even the UV fluorescence methods may prove to be inadequate to extract everything needed. Hence IMHO there will have to be lots of fudging as the scanner figures out what must have been there to make it all work. This will obviously require better mathematics than we now have. and then, faced with a brain-copy that needs debugging, what would we do? Debugging wetware is much the same as debugging software, only wetware is MUCH more forgiving of errors, since neurons routinely die at a horrendous rate even in healthy people. The best we could do is start another scan and hope for better luck next time. You can NOT rescan. You MUST get it right the first time. Even genetically identical twins raised together have very different brains when you look at the (visible light) microscopic details - as a half-century-old experiment on identical twin lab mice showed. Steve Richfield --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: