mckee@corwin.ccs.northeastern.EDU (George McKee) (03/02/88)
I think the recent speculations about learning about natural intelligence by simulating the brain in a nanotechnological device aren't looking carefully enough at the problem. If the brain is anything like the immune system, the source of the changes in neural structure that lead to learning, thought, and behavior are in the genome itself. If you recall the biomedical "breakthroughs" of just a few months ago, you'll know that the source of the immune system's ability to recognize new information is in "variable sequences" in part of the genome that codes for antibodies. Given the way genetic crossover can transfer enzymatic networks from one system to another, there's little reason to believe something similar doesn't work in the nervous system. It would help explain why the brain has a higher rate of protein synthesis than almost anywhere else in the body, and that blocking protein synthesis blocks some kinds of learning (G.Ungar's work years ago). You might argue that the protein synthesis is just making new synapses, but that fails to explain why more than 40% of the genome is expressed in the brain. There's alot more than just modification of synaptic efficiency (connection weights) going on there. If modifications in DNA sequences, or for that matter any molecular structure, play any significant role in the function of the nervous system in vivo, then those nanotechnologists that think they can do AI by building complete human brains "in calculo" are working at the wrong level of detail. Not only will they have to simulate neurons and synapses, but they'll have to simulate the molecules that control and form the structure of those synapses. What they ought to be doing is worrying about creating devices functionally equivalent to macromolecules in which the components have the same stability properties as real molecules, but the components are smaller, faster, and less noisy. Any technology that can get inside of molecules will of course be called "picotechnology." One way towards this is to use matter composed of other subatomic particles than electrons, protons, and neutrons. Looking at the subject line above, you can see where this leads... If you were trying to make devices out of positronium, you might attempt to stabilize them with a ceramic matrix like that of the high- temperature superconductors, but carrying electron-positron pairs rather then the superconductors' electron-electron pairs. It's true that the platinum-iridium sponge that forms the matrix for the creation and destruction of positrons in the "positronic brains" that power Asimov's robot stories contains rare-earth metals just like the high-temp superconductors, but I think that was just luck, and that Dr.A. chose that alloy simply because it was shiny and expensive. There can be no doubt about the success of R.Daneel Olivaw as an AI artifact, but of course this whole microtechnological exercise ignores the really interesting part of AI, namely the programming, not to mention the robopsychology. - George McKee College of Computer Science Northeastern University, Boston 02115 CSnet: mckee@Corwin.CCS.Northeastern.EDU Phone: (617) 437-5204 Usenet: in New England, it's not unusual to have to say "can't get there from here." p.s. I should add that I happen to be on David Baltimore's side of the debate whether or not to have a big project to sequence the entire human genome. Without going into a long discussion that's really irrelevant here, I think that a "big science" sequencing project will lead to a myopic focusing of attention on the mere task of sequencing, rather than the broader and harder to predict/manage task of understanding how 1-dimensional sequences become 3-dimensional proteins and organisms. Alas, it's characteristic of the adversarial nature of the political process to end up with only one golden egg in the funding basket. I wouldn't like to see a genome sequencing project end up like the Apollo or Space Shuttle projects have, but I'd bet that the probablity of such an outcome is directly proportional to the size of the project budget.