Path: utzoo!utgpu!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!ames!pasteur!ucbvax!hplabs!sri-unix!quintus!ok From: ok@quintus.uucp (Richard A. O'Keefe) Newsgroups: comp.ai Subject: Re: The limitations of logic Message-ID: <819@quintus.UUCP> Date: 7 Dec 88 00:13:25 GMT References: <9020@bcsaic.UUCP> Sender: news@quintus.UUCP Reply-To: ok@quintus.UUCP (Richard A. O'Keefe) Organization: Quintus Computer Systems, Inc. Lines: 55 In article <9020@bcsaic.UUCP> ray@bcsaic.UUCP (Ray Allis) writes: >In <696@quintus.UUCP> ok@quintus.uucp (Richard A. O'Keefe) says: >>In order for a digital system to emulate a neural net adequately, >>it is not necessary to model the entire physical universe, as Ray >>Allis seems to suggest. It only has to emulate the net. > >Emulation, simulation and modelling are all *techniques for analysis* >of the Universe. The trap analysts get into is to forget that the >model is not identical to the thing modelled. Emulating, simulating >or modelling a neural net, however "adequately", does not *duplicate* >a neural net or its behavior. You don't expect Revell models of F-16s >to do much dogfighting, why would you expect a model of a mind to think? By "neural net" I have meant all along the kind of thing AI people call a neural net, that is a computational device of nodes and weighted links. I haven't meant a network of neurons. That may be the cause of the misunderstanding. The point is that a *connectionist* net is not a natural part of the Universe to be modelled, but is *itself* a formal model, and it *is* possible for one formal model to completely duplicate the behaviour of another. I have never alleged that a model of a mind would think (or for that matter, that it would not). All I have claimed is that *one* formal model (a digital system) is capable of emulating *another* formal model (a connectionist net) to the point where the "real" thing (a connectionist net) and its emulation cannot be distinguished. I neither claim nor deny that either can or cannot model collections of biological neurons, minds, human beings, or even politicians. >>>You see, all the ai work being done on digital computers is modelling using >>>formal logic. > >>Depending on what you mean by "formal logic", this is either false or >>vacuous. All the work on neural nets uses formal logic too (whether the >>_nets_ do is another matter). > >Well sheesh! How many interpretations of the phrase "formal logic" ARE >there? I meant "form-al"; of or pertaining to form, disregarding >content. I realize the phrase is redundant, I can only plead seduction by >common usage. How many interpretations? Uncountably many. "Formal logic" is usually taken as meaning some variant of predicate calculus. Seduction by common usage would be an excellent excuse, if only it were true. Differential equations are not normally regarded as "formal logic", for example, though we employ (if we are wise!) formal logic in reasoning about them. The point I was making is that connectionist nets provide no mystical escape from "formal logic" (whatever that is). The claim that "all the AI work being done on digital computers is modelling using formal logic" simply isn't true: as soon as you connect the thing to the real world with cameras, pressure sensors, "arms", and so on, you have a thing which is behaving in the _real_ world, not a model of it. I don't know what a Revell model is, but the Air Force expect the Artificial Wingman project to deliver a computer program which when embodied in a computer in an appropriate aircraft should be able to do real dogfighting.