X-Message-Number: 24047 Date: Wed, 5 May 2004 08:23:11 -0400 From: Thomas Donaldson <> Subject: CryoNet #24040 - #24046 For Mike Perry: Believe it or not, I'm in the middle of a discussion with someone else on neural nets versus brains versus standard computers. If we were to try to actually build a structure with (to be simple) 1 billion nodes, each one with a connection to every one of the others, then it should be clear that we'd run into serious practical problems. Each node will have 1 billion - 1 connections emanating from it. Either you make the node very large, or you try to make it very small (nano?). If each connection has a diameter of one nanometer, then each node will have 1 billion such connections, which amounts to an outer area of (forgetting pi and the outer area of a sphere) about 1 meter. If the processor inside that node is quite small, you'll have a difficult design problem connecting it to those 1 billion connections. You may as well make it 1 meter square. Basically the connections will make the total size of the actual processors only 10^(-9) of the machine. Then you have to organize a route through all the other connections belonging to different nodes to each of the other nodes. Finally, for most computations only a small subset of those connections would actually be used. This looks to me like a very inefficient use of matter, and one which raises lots of practical problems with its implementation. Our totally connected machine will be roughly 1 km square, for instance, and even to make it we'll need tunnels within it to reach all its parts. And human brains have much more than 1 billion neurons, too. Furthermore, as brains work they always use more than one path at a time. Our problem is to take this system with 1 billion nodes and have more than one path between different neurons, without interference between the paths. (It's important that brains are parallel machines: if they were not, then their operation would take so long that other creatures with parallel brains would have gobbled them up). It's fine to consider such a machine theoretically, but fundamentally we can do much better by having a system able to grow new connections when they are needed rather than carry about almost N! connections which are all turned off. Best wishes and long long life to all, Thomas Donaldson Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=24047