X-Message-Number: 3851 From: (Joseph J. Strout) Subject: Re: CryoNet #3846 Date: Wed, 15 Feb 1995 09:16:07 -0800 (PST) > Message #3846 > From: (Thomas Donaldson) > But a few more words on uploading might help: > > First, it's probably true that we could make a computer capable of reacting > just like a person. In that sense, simulation would be possible. But for > actually copying a real person, you need more than that: you need a system > which works equally fast as a real person. That is where the difficulties > really arise: think of neurons not as single objects which send YES-NO > messages but as objects which themselves are small "computers", and respond > to many different messages both electrical and chemical. And then notice > that we have millions of them, all working in parallel. That is what makes > the problem hard. Thank you for making this important (and too-frequently-neglected) point. Neurons are NOT simple devices; they are horribly complex and still only dimly understood. Some research has suggested that the microtubule system within neurons acts as a sort of cellular automata- style computer, perhaps even with the quantum behavior that Penrose so expects to find (though I'm rather skeptical on this point). The reason neurons have gotten the reputation for being so slow is that in the past, we could not appreciate the amount of processing going on. On the other hand, it is not clear how much of that minute detail is really necessary for the larger-scale behavior of an organism. Much of it, no doubt, is kludy solutions to the problems of working with biological materials. I hope to do research in the next few years which will help pin down exactly how much of what neurons do is relevant to network dynamics (and therefore, thought & behavior). > Second, a need for operation at least as fast as a human being (note that > in many cases computers still can't equal that!) requires parallel computing. > ie neural nets of some kind. These neural nets will, of course, require > physical connections between them ... whereupon we find ourselves with the > problem of making such nets so that new connections (not in computer software > that simulates the system, but in the actual physical system) can be inserted > easily... or otherwise created. Mostly true, though there are work-arounds. If the simulation is sufficiently fast (e.g., using a CM-5 to simulate a nematode), then you can probably get away with software simulation and still operate in real time. Or engineered "neurodes" may be constructed such that every node connects to every other through physical connections of adjustable weight; making a new connection then means simply changing the weight to something other than zero. But, as you point out, this is mere speculation until such devices are on the drawing boards. > Third, there is entirely too much mythology about this uploading. That > mythology (and I'm hardly the first person in this discussion to raise this > issue) makes it hard to argue against uploading for those who believe in it. > They don't want to be uploaded in any present computer: not a Cray, nor an > nCube, nor a Connection Machine, nor any others, even the fastest now > known. They want to be uploaded into some IDEAL computer which by definition > has all the features required: large enough memory, fast enough to make a > real-time simulation, etc etc. True, because such features are necessary. Isn't it just as fair to say that cryonics clients don't want to be revived by current technology; instead, they want to be revived by a technology that can repair their cells and tissues, reverse damage from cracking or cryoprotectants, and cure whatever caused their original death? No technology we know of can do these things, just as no computer can emulate a human brain. The question is: what advances are required, how soon can we expect them, and what can we do to help? > If those in favor of uploading were to start thinking carefully about what > their ideal computer must do, then we may get somewhere. Yes. I think the biggest problem, at current, is our ignorance of the computational properties of the neuron, and small networks of neurons. Once that is understood, it should be fairly easy to extrapolate to more complex systems such as the human brain. (An aside, regarding Mr. Clark's claim that the information signals in an upload would be millions of times faster than in a biological brain: the most basic signals in a neuron spread electrotonically, i.e. at the speed of light. Moreover, if you increased the speed of action potentials while keeping the same speed of light, you'd completely change the dynamics of the system. It may someday be possible to simulate these things faster-than-life by sacrificing some resolution, but I wouldn't count on it any time soon.) ,------------------------------------------------------------------. | Joseph J. Strout Department of Neuroscience, UCSD | | http://sdcc3.ucsd.edu/~jstrout/ | |------------------------------------------------------------------| | Check out the Mind Uploading Home Page: | | http://sunsite.unc.edu/jstrout/uploading/MUHomePage.html | `------------------------------------------------------------------' Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=3851