Re: Computation

From: Harnad, Stevan (
Date: Sun Feb 04 1996 - 16:03:47 GMT

> From: "Parker, Chris" <>
> Date: Sun, 4 Feb 1996 16:02:55 +0000
> Computation is the manipulation of interpretable symbols. An example is
> that symbols interpretable by accountants as numerical values of
> incomes and expenditures (input) are fed into a computer running a
> financial application or software package. The computer manipulates and
> possibly transforms these symbols according to rules (computation) and
> produces further symbols interpretable by accountants as a numerical
> cash balance (output).

Very good, though you need to include some of the other features of
computation in your definition: (1) the arbitrariness of the shape of the
symbols (they neither "resemble" nor are causally connected with what
they are about); (2) the implementation-independence of computation;
(3) the fact that the symbol manipulation is "syntactic" -- i.e., based
only on the shape of the symbols -- and not "semantic" (based on

> Positive evidence:
> 1. C=C provided a solution to the dualist mind/body problem.

Not a solution, perhaps, but if it were correct, it would help
understand why we have trouble equating mental states with physical
states. Answer: because mental states are computational states, and
computational states are independent of physical states in the sense
that the hardware details are irrelevant; only the computational
properties matter.

> I > mind/body = program/computer, then the gap between the mental mind and
> the physical brain disappeared in the same way that programs run on
> computers. The mind, like a program, was implementation independent,
> and the brain, like the computer, is irrelevant to understanding
> function.


> 2. Artificial Intelligence (AI) has been very successful in emulating
> intelligent human activities such as: question and answer programs
> which appear to learn as they run, as in the Turing Test situation
> where a computer passes the (linguistic) test if an expert could not
> tell whether its answers to questions came from a computer or human;
> and programs that can emulate chess experts.

But of course no programme can pass the Turing Test (lifelong penpal
capacity) yet, by a long shot! Only brief fragments of it.

You have to point out that psychology's responsibility had been to
explain how people could do what they can do. Behaviourism had failed.
AI was the first to show how ANY system could do what people can do.

> 3. Computation has been successful in human type functions such as
> formal logic and calculations, exceeding human ability in speed,
> complexity and accuracy.

Sounds like a variant on 2 above; and the exceeding is of course not a

> 4. The successful simulation or modeling capacity of computation, from
> simple thermostatically controlled furnaces to aircraft behaviour in
> wind tunnels and traffic congestion on motorways, has also been an
> achievement which has tempted some to see computation's ability to
> simulate eventually extended to the mind. Pylyshyn said that cognition
> should be seen as literal computation, not merely simulated by it.

A bit of confusion here. The power of computation is a plus for
computation, partly because it makes computation a very general kind of
activity, capable of doing and hence explaining many things, partly
also because the mind is rather powerful too, and even computer-like in
some respects. The issue of computer simulation (as in a computer
simulating an airplane or a planetary system or a thermostat) versus
a system that literally computes is Searle's Weak AI vs. Strong AI
(computationalism), and has to be discussed carefully and separately,
each on its own terms.

> 5. There is also fabricated evidence that neural nets in computation
> are functionally equivalent to those in brains by virtue of their
> architecture.

It is unclear what "fabricated evidence" means; but what is the evidence
that serial computational simulations of neural nets (or real parallel
distributed networks?) are equivalent to those in brains? And what is
the point about function and about architecture?

Functional equivalence can mean input/output equivalence (the penpal
test is like that: send it the letters, and it sends back the replies,
functionally indistinguishable form a penpal); or it can be "strong
equivalence" -- running the same programme. But what has this to do with
the brain, about which we don't even know whether it is computing at

Architecture can mean hardware; or it can mean "virtual hardware," in
which one kind of machine runs a programme that makes it pretend it's
another kind of machine (like a PC pretending to be a Mac); the user
cannot tell the difference.

But you have to sort these out if you want to make a point about them.
Some of your positive points about computation are redundant with one

> 6. The failure of common sense (grandmother) objections may also be
> seen as evidence: that unlike us computers can't do anything new, be
> creative, make mistakes, exercise choice, have feelings; unlike us
> computers are inflexible, have no real history, are just machines and
> are isolated from the world. All these are either plain wrong or
> unproven.

Good idea to bring these in, but need to be fleshed out to make sense to

> Whether or not cognition = computation, it is quite clear that AI can
> perform some of the same functions as cognition, and this has helped us
> to understand our own cognition in new and novel ways.

This similarity has been taken as another point in computation's

This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:58 GMT