steward (12/08/82)
Several articles have past my desk in the past few days which have direct bearing on this discussion of artificial intelligence. First, I must emphasize that I am writing on a subject that is way out of any field of expertise I might profess. I'm attempting to communicate something I only vaguely understand so I'll provide references. (Is net.misc going to be the first net.??? to use references consistently?) 1) "Brain's `Glue' Discovered". New York Times; Tuesday, November 23, 1982, first page of Science Times section. This article describes work done here at Rockefeller University by Dr. Gerald Edelman and his colleagues. They have discovered the protein that joins nerve cells together. It is key to the recognition of nerve cells by other nerve cells and muscles and is also vital to the process of forming links and patterns in the development of the nervous system. The discovery suggests that there is only one or two substances that control cell to cell recognition. (Jim Kempf, do you see anything exciting here?) 2) "Through A Computer Darkly: Group Selection and Higher Brain Function", Bulletin, The American Academy of Arts and Sciences; Vol. 36, No. 1, October 1982, Pages 18 - 49. By Dr. Gerald M. Edelman Now, cursing my ignorance all the way, I'll try to summarize some parts of the paper. First, Dr. Edelman discusses perception and postulates the nature of brain organizations responsible for perceptual functions. Perception is, he feels, at the center of all higher brain function. He presents a theory that the brain is a selective system (Darwinian-like) in which a large pre-existing set of variants of neural networks are selected for and against during the life of an animal. Dr. Edelman briefly describes a selective automaton he made with George N. Reeke, Jr. that is based on his theory of brain organization. (He states that a technical description of the computer, called "Darwin II," can be found in: Edelman, G.M., and Reeke, G.N., Jr. 1982. Proc. Natl. Acad. Sci. [USA]. 79, 2091) Their purpose was "to simulate in a large digital computer a set of selective networks that would represent, classify and associatively recall representations of two dimensional shapes." The restrictions that they made were "in the direct function of the automaton, there would be no instructions, no forced learning, and no program of the kind that is used for computers." (Dwell on that for a while.) Well, they built the computer, presented objects for it to "see", and it selected and developed relationships between individual representations and classifications of two dimensional shapes. He reports that the computer does not work well; but it is very encouraging that it does work at all, especially since there is no semantic instruction in the machine. I think that these two articles are each amazing in themselves, startling when taken together, and absolutely phenomenal when they both describe one person's work. To give the readers of the net a little background, Dr. Edelman won the Nobel Prize in 1972 for his work on the nature and structure of antibodies. He is currently Vincent Astor Professor here at the Rockefeller University and director of Neurosciences Institute of the Neurosciences Research Program which is now located at the University. (Lew Mammel, I hope these articles might guide you closer to a summary of neural simulation work.)