From Aaron Sloman Wed Jul 23 23:31:49 BST 1997
To: JRStern@gte.net,rickert@cs.niu.edu
Subject: continuity

I've just posted this. Don't know if it would ever reach you. News
seems to be increasingly erratic.

Posted 23 Jul 1997
Newsgroups: comp.ai.philosophy
References: <5r00dl$3af@ux.cs.niu.edu> <5r07e4$86c$1@gte1.gte.net> <5r0bme$3j3@ux.cs.niu.edu> <5r0rc5$99u$2@gte1.gte.net>
From: AaronSloman@cs.bham.ac.nospam (Aaron Sloman See text for reply address)
Subject: Re: "Stuck" research (AI isn't stuck, What is AI's Vision?)

[Apologies for anti-spam address]

JRStern@gte.net (JRStern) quotes Neil

> On 21 Jul 1997 14:01:34 -0500, rickert@cs.niu.edu (Neil Rickert)
> wrote:
> >For simplicity, I predict that strong AI will never succeed, if AI
> >practitioners persist in their denial of a distinction between
> >discrete and continuous systems.

Er, who denies this exactly?

It's so obvious I can't imagine who would deny it, even if not
everyone studies it.

The question arises: where does it impact on the aims, methods,
theories of AI? Of course not everyone can study everything and
there are lots of people who focus on problems where continuity is
irrelevant.

But there have always been people concerned with other areas of AI,
where the distinction between discrete and continuous systems cannot
be ignored. e.g. vision, speech, robotics, some types of neural
nets, various kinds of fuzzy or probabilistic inference, and of
course explicit reasoning about continuous systems (e.g. the work of
Tony Cohn and colleagues reasoning about spatial and topological
relationships, and much work on qualitative physics),

So why proclaim this distinction as if it were some major obstacle
nobody has noticed?

Just to illustrate, I recall that some time in the 70s robot
designers discovered that whereas it is very hard to program a robot
with a rigid grasp of a rod so that it is pushed into a
tight-fitting cylinder, the task becomes much easier if, like
biological systems, you combine the software with a "compliant
wrist" mechanism whose geometry and sloppiness combine to make it
behave as if it were *pulling* the rod into the hole instead of
pushing it in.

I think issues like this are commonplace in various robotic labs,
e.g. the Brooks Lab at MIT.

In vision it is commplace to use mathematics which treats an image
as if it were continuous even if it is sampled discretely. (Have you
heard of techniques for achieving sub-pixel accuracy?)

Don't think that AI is defined by your favourite text book
introduction to expert systems or blocks world planning, or parsing,
or by what a Turing machine can do.

That's just the narrow minded view of AI promoted by people who work
in restricted sub-areas and don't read journals, conference
proceedings, etc. which go beyond their narrow interests.

A good way to find out what AI actually is is to look at IJCAI
(International Joint Conference) Proceedings over the last 28 years
or so. The conferences are held every two years and are very wide
ranging.

This year in Japan: see

    http://www.etl.go.jp/etl/suiron/ijcai-97/  (most up-to-date info)
    http://ijcai.org/ijcai-97/   (official page)

I would like to be there, but can't! But I expect that a significant
subset of the workshops and papers will address topics involving
continuity. Probably a small subset (10% ???), but real enough.

People who don't work on problems involving continuity don't all
necessarily deny that it exists or is important.

(There's a lot of relevant discussion in Brian C Smith's 1996 book
On the Origin of Objects, MIT Press. Not an easy read.)

Cheers.
Aaron
===