Re: [agi] Live Forever Machines...
Josh, On 5/30/08, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: You'd get a hell of a lot better resolution with an e-beam blowing up nanometer-sized spots, and feeding the ejecta thru a mass spectrometer. Yes, but all your spectrometer will see is hydrogen, oxygen, and carbon. The e-beam will destroy the very complex molecules that make neurons do what they do, whereas their fluorescence can be observed at UV resolution, even in living tissue. Further, e-beams are WAY too slow. It would take more than a lifetime to scan out an entire brain, because each spot must be done separately and the particles flushed out before the next spot is done, whereas the scanning UV fluorescence microscope can do millions of isolated spots at a time. Note the following complexity constant: Taking a computer capable of simulating a brain, neuron by neuron, synapse by synapse, in real time, and instead loading it with the software needed to reconstruct the logical diagram of a brain given the images from a scanning UV fluorescence microscope, e-beam image, etc., it will take the SAME amount of time to reconstruct the logical diagram, regardless of the complexity of the brain. Why? A brain that is twice as complex will need twice as much computing power to simulate it, and with twice the computing power, the computer can reconstruct twice as complex a brain in the same amount of time. Hence, this constant is in units of time. While it is WAY too early to make any accurate estimate of this constant, it may not be too early to guess its order of magnitude. My own estimates are on the order of one month. Of course, a mass production operation could potentially combine several such computers to accomplish this task in a shorter amount of time. I hereby dub this constant Richfield's Constant. In any case, when it becomes possible to diagram brains, it will NOT take all that long to get the first diagrams. See my talk a couple of years back at Alcor. It's good to see our participants are making it around to other forums. But I would suggest that this is *wy* off-topic for this list... Is it? I guess the REAL question is whether our prime directive here is to 1. create and/or evaluate super intelligences, regardless of what may be needed to achieve this goal (which seems the most logical to me), or 2. hack away at code as AI has been doing for the last 40 years, in the hopes that a breakthrough will be forthcoming without much new information (which seems to be a nearly hopeless venture, and one that no one will even appreciate when the scanning information DOES become available)? This all reminds me of one mathematician's goal of estimating pi by dropping a needle onto a hardwood floor having slats the same width of the needle, and keeping track of the frequency of the needle crossing a crack. He did this many thousands of times, during which time other better ways of calculating pi were developed by other mathematicians. His estimate was about as good as he expected it would be, but was of absolutely no use to anyone other than as a historical note. I suspect that present pre-scanning AGI efforts will suffer exactly the same fate, and for pretty much the same reasons. uploading implications to the contrary notwithstanding. ??? Steve Richfield --- 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] Live Forever Machines...
Ben, I just LOVE your posting, because it asks exactly the right questions. On 5/31/08, Ben Goertzel [EMAIL PROTECTED] wrote: I think that brain scanning is an interesting and important technology/research direction, but I don't see why you think it is easier to create than AGI. I think that we could agree that if we could somehow wring this out to everyone's satisfaction, that two different fields might be combined to everyone's benefit. OK, so I'll try to return your ball here: Of course we have no solid proof as to which is easier, but my sense is that AGI is easier because it's just programming In a very real sense, both efforts are just programming, only the scanning looks (to me) to be a lot easier since it only involves 3D reconstruction based point data. The BIG challenge is recognizing what is significant functionally and what is not. Note that there are presently a number of microscopists working on just this problem, but without (what either of us would consider to be) adequate tools. (and I have a design in hand that I believe will be workable, though many details remain to be resolved), Like all software, it will do what it is designed to do. However, will it then evolve into the great super-intelligence of the future? If you (and others) are capable of recognizing and overcoming EVERY SINGLE ONE of those puzzles, then it may/will. If you fail to recognize and solve JUST ONE of those puzzles, then it won't. I see brain scanning to be cheap insurance that AGI will indeed achieve its wildest dreams, and I suspect that this case could also be made to prospective investors, for YOUR direct and immediate benefit to provide you funding while the scanning equipment is being designed, constructed, and utilized in parallel with your present efforts. I think that this may be one of these rare situations where it may be easier to get two projects funded instead of one. Here, AGI guarantees a market for the scanner, and the scanner guarantees ultimate success for AGI. Either of these failing does NOT spell disaster for the investment. whereas effective brain scanning requires a load of physical engineering work Not that much! Note that semiconductor manufacturers already use e-beam scanning to write one-off semiconductors, and that my scanning UV fluorescence microscope requires NO new science, and could hence be farmed out as an engineering project. Hence, either of these approaches could be in operation within a year or so, as only money is lacking. With NO money, none of either of our efforts stands a chance. With some realistic investment money, scanning would at minimum be cheap insurance that you will be able to overcome ALL of your future problems. I don't believe that we need to understand brain function in detail to make an AGI ... Neither is complete understanding needed for human brain emulation, though whatever understanding there is will sure help debugging. human-brain-emulation is not the only approach to AGI, I agree. However, even without emulation, a roadmap would sure be helpful. and IMO is a very risky approach to take due to the ethically erratic nature of human brain architecture. Again I agree. However, there could literally be BILLIONS of investment dollars to support all of our efforts if immortality were available to the highest bidder. On the other hand, first AGIs will be no better than babies - and we already have too many of them. I am more interested in making AGI systems with action-selection mechanisms more closely tied to logically consistent goal hierarchies, and with greater self-understanding. As I have been posting and explaining, our own very human goal selection sure doesn't work very well, and should NOT be a model for a future AGI, for the SAME ethical/dangerous reasons as simulating human structure. Learning simply doesn't work here, as we can't afford to have enough nuclear exchanges to learn how to do it right. Nor am I convinced that UV fluorescence is necessarily going to be the winning approach to high space-time-res brain scanning ... though it is certainly one potentially viable approach... Only millions, not billions of dollars are needed to evaluate the alternatives and go with the best. Considering that researchers are successfully using non-scanning fluorescence, complete with its visible light resolution, instead of scanning approaches with UV resolution, to do anything at all, suggests that even if the most minimal goals are achieved, that an investment here would at minimum produce some VERY valuable lab equipment. In short, this should be a moneymaker, even if it proves to be a complete failure for AGI. Steve Richfield --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription:
Re: [agi] Live Forever Machines...
I'll respond to other points tomorrow or the day after (am currently on a biz trip through Asia), but just one thing now... You say With NO money, none of either of our efforts stands a chance. With some realistic investment money, scanning would at minimum be cheap insurance that you will be able to overcome ALL of your future problems. but I'm not sure this is true. Linux got a long way with no money, and eventually its freeware success brought investment from various sources. This is part of the inspiration underlying OpenCog Given a viable AGI design (which I have) and an initial, partial codebase (which we have, via opencog), and a population of enthusiastic and qualified OSS contributors, there's no reason that $$ is needed, though it can certainly accelerate things. The big challenge becomes keeping things going on the right course, in the relevant senses of right ben g --- 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] Live Forever Machines...
Steve, Josh, etc. Agree this is off-topic ... it should be posted to [EMAIL PROTECTED] instead, perhaps... so I have cross-posted it there and suggest to continue the discussion there. Steve: I think that brain scanning is an interesting and important technology/research direction, but I don't see why you think it is easier to create than AGI. Of course we have no solid proof as to which is easier, but my sense is that AGI is easier because it's just programming (and I have a design in hand that I believe will be workable, though many details remain to be resolved), whereas effective brain scanning requires a load of physical engineering work I don't believe that we need to understand brain function in detail to make an AGI ... human-brain-emulation is not the only approach to AGI, and IMO is a very risky approach to take due to the ethically erratic nature of human brain architecture. I am more interested in making AGI systems with action-selection mechanisms more closely tied to logically consistent goal hierarchies, and with greater self-understanding. Nor am I convinced that UV fluorescence is necessarily going to be the winning approach to high space-time-res brain scanning ... though it is certainly one potentially viable approach... thx Ben G On Sat, May 31, 2008 at 7:04 AM, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: You'd get a hell of a lot better resolution with an e-beam blowing up nanometer-sized spots, and feeding the ejecta thru a mass spectrometer. See my talk a couple of years back at Alcor. But I would suggest that this is *wy* off-topic for this list... uploading implications to the contrary notwithstanding. On Friday 30 May 2008 04:42:07 pm, Steve Richfield wrote: Ben, et al, I have posted in the past as part of other postings that live forever machines should be at once much easier to build, worth far more than an AGI, and lead directly to an AGI. However, no one has even commented on this. If I am right, then present efforts should shift in that direction. If I am wrong, then efforts directed in that direction should be redirected here. Hence, this IS an important issue to consider here. Live forever Machine: An assembly line that takes dying/dead people, scans out their entire brain structure, and installs that structure into a new android body that is both repairable and capable of periodic checkpoints. In short, to last forever, potentially outliving even our own sun. Of course, additional computing capacity could be installed, thereby making AGIs. Note that the main barrier both to AGI and to live forever machines is the present inability to easily scan out present brain structure. I have explained on other forums how to do this, but it would take some investment in new machinery. The missing device is a scanning UV fluorescence microscope, which would focus isolated spots of UV a few microns into brain tissue and observe the fluorescence and decay characteristics, and then move on to other spots until the top few microns were completely scanned out. Then, a cryostatic microtome would slice off ~4 microns and the process would continue anew, with a computer matching up the newly exposed surface with the deeper part of the structure previously scanned out. Brain tissue is (nearly) transparent and is richly fluorescent, providing a map of chemical structure from its fluorescence. There are experts (like Kathryn Graubard) in this area, but all they have to use are non-scanning fluorescence microscopes that only have visible-light resolution, no ability to read out decay characteristics, and no attached computers to make the best of the available information. From the physical mapping, the computer would then generate a logical mapping, which would then be put into a simulation program to continue the operation of the scanned-out brain. Presuming that the entire AGI concept is indeed correct, this logical mapping could then be added onto, to make an AGI of limitless capability, but with the continuing consciousness of a specific individual human. This provides a smooth path to an AGI but without having to solve the early development puzzles, or having to understand the myriad little details that doubtless stand in the way of people-programmed efforts. Then, one we actually have a functioning AGI whose operations is 100.00% observable, we can work on completely understanding it to make better. It seems like SUCH a waste of time and effort here to be working on theories and ideas, most of which would be made instantly obsolete by the appearance of a scanning UV fluorescence microscope. Wouldn't it be MUCH better to focus on making at least one of these machines to answer substantially all of the outstanding questions regarding how we work? Certainly, any potential investor would have to see the need for such a device to assure future AGI success (not to mention finding the cures to countless diseases), so