vu0112@bingvaxu.cc.binghamton.edu (Cliff Joslyn) (05/07/88)
I have recently been thinking some about formal systems and AI, and have
been prompted by our recent conversations, as well as by an *excellent*
article by Chris Cherniak ("Undebuggability and Cognitive Science",
_Comm. ACM_, 4/88) to make some comments.
It seems patently obvious to me at this point that the following
statements are false:
1) The mind is a formal system.
2) Attempts to construct AI as a formal system can succeed.
These ideas seem to me to be the heart of "classical Cartesian Cognitive
Science" (e.g. Fodor, Chomsky). I assert that these positions are
based on an old, false view of a deterministic, deductive, reducible
world. The relative failure of strong, theoretical AI seems, in
hindsight, terribly obvious.
Cherniak takes the following stance: "A complete computational
approximation of the mind would be a huge, 'branchy,' holistically
structurred, quick-and-dirty (i.e. computationally tractable, but
formally incorrect/incomplete), kludge. . .[as opposed to] a small set
of elegant, powerful, general principles, on the model such as classical
mechanics."
This view is not only common-sensical, but is well-motivated by some
gross approximations about *real* intelligent systems and *real* physics
of information system. For example, let's say that I have a computer so
small that it could calculate a line in a truth-table in the time it
takes for light to cross the diameter of the proton. Cherniak concludes
that there is then an upper bound of ~ 138 independent logical
propositions that can be solved by the truth-table method. A tiny
number! More quotations: "Our basic methodological instinct. . .seems
to be to work out a model of the mind for 'typical' cases - most
importantly, very small cases - and then implicitly to suppose a grand
induction to the full-scale case of a complete human mind."
Instead, we see large software systems (e.g. Star Wars), rather than
being elegant, correct/complete/verifiable formal systems, as being huge
unintelligible bug-ridden masses. It is well known that programmers
quickly lose the ability to understand their own code, let alone verify
it. Visualization past three dimensions is practically impossible, yet
real information systems have thousands of dimensions.
This move away from formalism as a valid paradigm for AI seems perfectly
in step with non-von Neumann arhitecture (i.e. connectionism), as well
other academic trends away from deterministic, deductive, reducible
theories towards the science of fuzzy, uncertain, multi-dimensional
information system.
--
O---------------------------------------------------------------------->
| Cliff Joslyn, Cybernetician at Large
| Systems Science, SUNY Binghamton, vu0112@bingvaxu.cc.binghamton.edu
V All the world is biscuit shaped. . .