Xref: utzoo comp.ai:5322 talk.philosophy.misc:3388 sci.philosophy.tech:1840 Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!samsung!shadooby!sharkey!tygra!jpp From: jpp@tygra.UUCP (John Palmer) Newsgroups: comp.ai,talk.philosophy.misc,sci.philosophy.tech Subject: Re: Can Machines Think? Keywords: Neural Networks and Computation Message-ID: <35@tygra.UUCP> Date: 31 Dec 89 11:31:19 GMT References: <31821@iuvax.cs.indiana.edu> <32029@iuvax.cs.indiana.edu> <1037@ra.stsci.edu> <85217@linus.UUCP> Reply-To: jpp@tygra.UUCP (John Palmer) Followup-To: comp.ai Organization: ThunderCat Systems Inc, Detroit, MI Lines: 81 }In article <85217@linus.UUCP> bwk@mbunix.mitre.org (Barry Kort) writes: }In article <1037@ra.stsci.edu} bsimon@stsci.EDU (Bernie Simon) writes: } } } 6) While there are good reasons to believe that thinking is a physical } } activity, there are no good reasons for believing that thinking is the } } execution of a computer program. Nothing revealed either through } } introspection or the examination of the anatomy of the brain leads to } } the conclusion that the brain is operating as a computer. If someone } } claims that it is, the burden of proof is on that person to justify that } } claim. Such proof must be base on analysis of the brain's structure and } } not on logical, mathematical, or philosophical grounds. Since even the } } physical basis of memory is poorly understood at present, any claim that } } the brain is a computer is at best an unproven hypothesis. } }The brain is a collection of about 400 anatomically identifiable }neural networks, interconnected by trunk circuits called nerve bundles, }and connected to the outside world by sensory organs (eyes, ears, nose, }tactile sensors) and effectors (muscles, vocal cords). Neural networks }are programmable computational devices, capable of categorizing stimuli }into cases, and capable of instantiating any computable function (some }more easily than others). Artificial neural networks are used today }for classifying applicants for credit or insurance. They have also }been used to read ASCII text and drive a speech synthesizer, thereby }demonstrating one aspect of language processing. As to memory, you }might want to explore recent research on the Hebb's synapse. } }--Barry Kort But the brain is not structurally programmable. The tradeoff principle states that no system can have structural programmability, evolutionary adaptability and efficiency at the same time. Digital computers are programmable, but lack efficiency (I may post more on this later) and evolutionary adaptability. The brain (ie: humans) has evolutionary adaptability and (relative) efficiency. Biological neurons are much more complex than their weak cousins (artificial neurons) and contain internal dynamics which play a very important role in their function. Things like second messenger systems and protein/substrate interactions are important. Internal dynamics rely heavily on the laws of physics and we cannot determine what "function" a neuron "computes" unless we do a physics experiment first. Computer engineers work very hard to mask off the effects of the laws of physics (ie: by eliminating the effects of background noise) in order to produce a device which is structurally programmable. Biological neurons, on the other hand, RELY on the laws of physics to do their work. The basic computing element of biological systems, the protein, operates by recognizing a substrate. This is accomplished by Brownian Motion and depends on weak bonds (VanderWaals interactions, etc). Thus, there is a structure/function relationship which is essential. Artificial neural nets will still be unable to solve hard problems (patttern recognition, REAL language processing, etc) because they are implemented in silicon (usually as a virtual machine on top of a standard digital computer) and are therefore inherently inefficient. In theory (Church-Turing Thesis) it is possible for such problems to be solved by digital computers, but most of the hard problems are intractable. We are very quickly reaching the limits of speed of silicon devices. The only hope of solving these hard problems is by developing devices which take advantage of the laws of physics and that have a very strong structure/function relationship. Of course, these devices will not be structurally programmable, but will have to be developed by an evolutionary process. My point: We are not going to solve the hard problems of AI by simply developing programs for our digital computers. We have to develope hardware that has a strong structure/function relationship. Sorry if this posting seems a little incoherent. Its 5am and I just woke up. I'll post more on this later. Most of these ideas are to be attributed to Dr. Michael Conrad, Wayne State University, Detroit, MI. -- = CAT-TALK Conferencing Network, Prototype Computer Conferencing System = - 1-800-446-4698, 300/1200/2400 baud, 8/N/1. New users use 'new' - = as a login id. E-Mail Address: ...!uunet!samsung!sharkey!tygra!jpp = - <<>>> -