Next August, I’m presenting a paper entitled “Connectionist Computation” at the International Joint Conference on Neural Networks. The paper, which will be published in the conference proceedings, is an account of why (some) connectionist systems perform computations even though they don’t execute programs.
(Background: In 2005, Martin Roth published a paper in Mind and Language arguing that (some?) connectionist systems compute because, contrary to what you might expect, they execute programs. Indeed, many philosophers hold that computation requires executing programs; but then, either connectionist systems execute programs (as Roth maintains) or they don’t compute. By employing my general account of computation, I show why we need not impale ourselves on either horn of this dilemma.)
The organizers of IJCNN invited me to submit an expanded version of my paper to a special issue of the journal Neural Networks. So I have written a largely new paper, entitled “Some Connectionist Systems Compute, Others Don’t.” Here is the abstract:
<<I address whether connectionist systems perform computations in the sense of ‘computation’ used in computability theory and computer science. I explicate and defend the following theses. (1) Many connectionist systems compute–they perform computations. (2) Some connectionist systems compute in a classical way. Digital computers, which are very large networks of logic gates, belong to this class of connectionist systems. (3) Other connectionist systems compute in a non-classical way. (4) Yet other connectionist systems do not perform computations. Brains may well fall into this last class.>>
My paper is due by August 1 (about two weeks from now). I would be very interested in receiving comments before the paper is due (and would, of course, acknowledge them in the final version). If anyone would like to comment on the paper before it’s due, please contact me and I’ll email it to you.