RE: duck me!

From: Eugen Leitl (eugen@leitl.org)
Date: Mon Nov 04 2002 - 13:00:27 MST


On Mon, 4 Nov 2002, gts wrote:

> Forgive my ignorance, Eugene, but I have no idea what you mean by a
> "SHRDLU world," much less how it relates to this question of continuous
> experience.

If I don't know, I usually ask Google. And frequently I can even feel
lucky! I just asked Google

http://www.google.com/search?hl=en&ie=UTF-8&oe=UTF-8&q=shrdlu&btnG=Google+Search

and found http://hci.stanford.edu/~winograd/shrdlu/ (I'm remarkably lucky
today). Here's what it says:

SHRDLU

SHRDLU is a program for understanding natural language, written by Terry
Winograd at the M.I.T. Artificial Intelligence Laboratory in 1968-70.
SHRDLU carried on a simple dialog (via teletype) with a user, about a
small world of objects (the BLOCKS world) shown on an early display screen
(DEC-340 attached to a PDP-6 computer).

SHRDLU is described in Winograd's dissertation, which was issued as MIT AI
Technical Report 235, February 1971 with the title Procedures as a
Representation for Data in a Computer Program for Understanding Natural
Language It was published as a full issue of the journal Cognitive
Psychology Vol. 3 No 1, 1972, and as a book, Understanding Natural
Language (Academic Press, 1972).

SHRDLU was written in MacLisp for the ITS system, vintage 1970. The source
code is available at http://hci.stanford.edu/winograd/shrdlu/code and as a
TAR file at ftp://ftp-csli.stanford.edu/pub/shrdlu.tar.gz One project to
port it to more modern machines is described at
http://www.ont.com/~keldon/shrdlu.html with pointers to the converted code
files.A discussion of various efforts is on the SHRDLU is at
http://www.semaphorecorp.com/misc/shrdlu.html

You can download a Windows text-only console version of SHRDLU implemented
in Common Lisp, or a graphical 3-D version implemented with an extra Java
layer. Source code is included. These files were supplied by Greg Sharp,
and were produced by the UMR student project to resurrect SHRDLU.
Double-click the SHRDLU.BAT file in either version to start running. No
guarantees. For questions about the code, contact Greg Sharp.
mike8s2@hotmail.com

If you have other questions, contact Terry Winograd
<winograd@cs.stanford.edu>. You can also see the story of how SHRDLU got
its name.
 
> Please explain the neurological mechanisms by which you believe you are
> awakened by your alarm clock if you are not continuously aware of your
> environment.

Of course you have a clock ticking. However, I'm assuming this is still
the identity thread. If it isn't, I have at all no point to make.

If this is still the identity thread, I'm suggesting to use SHRDLU as a
model for discussing identity in an upload (or machine intelligence, in
fact this is exactly what SHRDLU is), with the following modifications:
hooking up a reality simulator which renders the world, and a machine
vision package which interprets the input of a frame buffer rendered by
the reality simulator (including a physical model of robotic arm dynamics,
or we can just make it slow enough that we can assume the arm is in a
predictable position), or input of a CCD camera and the according robotic
arm operating in a blocks world. We're assuming two identical,
synchronizable (though not necessarily always synchronized) computers
running the program. The internal state of each computer at each clock
cycle is being dumped in a very large nonvolatile storage, constituting
the system trajectory, which we can look at after the experiment. We can
hook up both machines to reality simulators, to robotic arms/camera/blocks
world, or a mix thereof. We also have a program which measures the state
of the arm and the frame buffer, and its internal representation, and can
make adjustments.

It's a bit more fancy, but this is easily doable as a Ph.D. thesis, though
no one would want to do it, because it is entirely predictable. Zero
novelty.

Does above description make sense? It's a bit short, but I can elaborate
if/when needed. If we can agree on this setup, I'm ready to translate
anything you would want to figure out about real uploads, or meat people
in above framework. I suggest we start with uploads first, because it's so
much easier.

That is, if anybody still feels like beating this very dead horse.



This archive was generated by hypermail 2.1.5 : Wed Jan 15 2003 - 17:57:57 MST