Newsgroups: comp.ai.philosophy
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!uhog.mit.edu!news.media.mit.edu!minsky
From: minsky@media.mit.edu (Marvin Minsky)
Subject: Re: When is a simulation of a Y a Y? (Was Bag the Turing
Message-ID: <1995Jan8.063326.5690@news.media.mit.edu>
Sender: news@news.media.mit.edu (USENET News System)
Cc: minsky
Organization: MIT Media Laboratory
References: <3ekae6$t7p@mp.cs.niu.edu> <3enci8$9kt@tadpole.fc.hp.com> <3enie0$php@mp.cs.niu.edu>
Date: Sun, 8 Jan 1995 06:33:26 GMT
Lines: 37

In article <3enie0$php@mp.cs.niu.edu> rickert@cs.niu.edu (Neil Rickert) writes:
>In <3enci8$9kt@tadpole.fc.hp.com> allsop@fc.hp.com (Brent Allsop) writes:

>
>>	Hmmm.  All robots that I know of, like "Shakey" at Stanford,
>>convert camera and other sensing data into 3D models.
>
>That may be the case.  I interpret it differently.
>
>It is my impression that simple robots maintain a 3D internal model,
>and attempt to update it from camera and other sensing data.  They
>do this because they have a relatively inept sensory capability, and
>can trust the internal model more than they can trust sensory
>information.
>
>One consequence is that they suffer from the frame problem.  That is,
>the internal model is sometimes not correctly updated.

Reminds me of the joke about when someone said 'look at that black
cow' and the respondent replied 'well, it's black on this side anyway.
The situated action idea suffers from not even being able to assume
that it's a cow on the other side.  The solution to that "frame
problem" is to use (internal model) frames for reasoning, so that you
can make pretty good default assumptions about what you can't see.
How do you know there's still a floor in back of you after you took a
step.  You don't 'know'.  You don't even need to remember it, if you
keep your room-frame active for a while.

It is we who have a "relatively inept sensory capability", and must
use our models all the time.  

(This discussion would be more sound if there were some evidence that
situated robots could actually be made to be good at things.  So far,
they don't seem to work well enough.)

