Posted Sun Jul 28 15:49:42 BST 1996
Newsgroups: comp.ai.philosophy
References: <4tb841$dji@usenet.srv.cis.pitt.edu> <hubey.838436988@pegasus.montclair.edu> <4td5ub$5oe@ux.cs.niu.edu>
Subject: Re: Note on design stance (Re: Sloman (Part 5))

rickert@cs.niu.edu (Neil Rickert) writes:

> Date: 27 Jul 1996 08:35:07 -0500
> Organization: Northern Illinois University
>
> hubey@pegasus.montclair.edu (H. M. Hubey) writes:
> > ... then
> >you can skip the intentional states, and make behavior a deterministic
> >function of inputs. This is what behaviorism, even the radical ones,
> >essentially did.

[NR] (with typo corrected)
> On the surface, this sounds fair enough.  Now I have a computer on my
> desk.  Most people would say that the behavior (outut) is a
> deterministic function of the inputs.  So why not dispense with all
> of the talk of software, data structure, virtual machine, process
> contexts, etc?  The fact is, if we tried to describe the operation of
> the computer without reference to this terminology, we would find it
> extremely difficult, or even impossible.

Yes.

More importantly, if we tried to formulate a GENERAL science that's
applicable to MANY kinds of computing systems, by talking only about
observable input-output relations and the underlying physical
mechanisms, then we'd lack the ability to formulate theories
explaining similarities and differences between this computer and
many others, some using the same software on different hardware,
some using partly similar software, some including different
combinations of subsets of the software that's on this computer
alongside other software items that are not on this computer, etc.

It would be partly like trying to do the whole of chemistry
(including organic chemistry) without ever talking about any
unobservable physical structures larger than sub-atomic particles.

The same goes for attempts to do human and animal psychology without
talking about anything between observable behaviour of whole
organisms and physical processes.

Many people agree with this, and even Anders sometimes seems to.

What is often missed is that it is useful to use intentional
concepts for some of the intermediate levels. E.g.
    This process contains information about the state of that one.
    This structure describes the differences between those two.
    This table records information for the operating system.
    etc. etc.

Sometimes the usefulness of this way of talking is acknowledged, but
it is then hedged with qualifications:
    The machine does not really understand,
    The structures do not refer for the machine, only for us,
    It's all *derivative* intentionality,
    etc.

This is usually based on the notion that understanding is an
all-or-nothing matter, and since the machine and its subsystems do
not have full human-like understanding of the internal structures,
they therefore have no understanding, and no intentional states.

The answer to this is to examine patiently all the different sorts
of understanding that there are, in order to see what they have in
common and how they differ, and how they fit into a larger framework
of varieties of intentionality, e.g. some of which do not require a
social system with public language, etc.

But there is enormously strong resistance to work which requires
deep revision of some of our basic conceptual frameworks. (On this
Quine was right).

The resistance is even stronger when doing the work requires
breaking disciplinary barriers, e.g. between philosophy and science.

Aaron