Copy Link
Add to Bookmark
Report
AIList Digest Volume 1 Issue 007
AIList Digest Sunday, 22 May 1983 Volume 1 : Issue 7
Today's Topics:
LISP for VAX VMS
AI Job
Phil-Sci Mailing List (2)
Computer Resident Intelligent Entity (CRIE) [Long Article]
----------------------------------------------------------------------
Date: 19 May 1983 09:19 cdt
From: Silverman.CST at HI-MULTICS
Subject: lisp for vax vms
We are trying to find out what implementations of lisp exist that we
can run on our vax under vms. Any information about existing systems
and how to get them would be appreciated. Reply to Silverman at
HI-Multics.
------------------------------
Date: Thu 19 May 83 10:41:33-PDT
From: Gordon Novak <NOVAK@SU-SCORE.ARPA>
Subject: AI Job
Two individuals with strong CS background and specific interest in
A.I. sought for development of a modern air traffic control system
for the whole U.S. Position located on East Coast in mid-Atlantic
states. Contact Jay. R. Kronfeld, Kronfeld & Young Inc., 412 Main
St., Ridgefield, Conn. 06877. (203) 438-0478
------------------------------
Date: 9 May 1983 1047-EDT
From: Don <WATROUS@RUTGERS>
Subject: Prolog, Phil-Sci mailing lists
[...]
Also of interest to local readers might be the local Phil-Sci BBoard,
which receives the Philosophy-of-Science mailing list.
Here is its description:
PHILOSOPHY-OF-SCIENCE@MIT-MC
(or PHIL-SCI@MIT-MC)
An immediate redistribution list discussing philosophy of science
with
emphasis on its relevance for Artificial Intelligence.
The list is archived@MIT-OZ in the twenex mail file:
OZ:SRC:<COMMON>PHILOSOPHY-OF-SCIENCE-ARCHIVES.TXT.1
All requests to be added to or deleted from this list, problems,
questions,
etc., should be sent to PHILOSOPHY-OF-SCIENCE-REQUEST@MIT-MC (or
PHIL-SCI-REQUEST@MIT-MC).
Coordinator: John Mallery <JCMa@MIT-MC>
------------------------------
Date: 10 May 1983 01:36 EDT (Tue)
From: _Bob <Carter@RUTGERS>
Subject: Phil-Sci Readers, Please Note
Hi,
Before FTP'ing the archive mentioned by Don's Phil-Sci announcement,
[OZ]SRC:<COMMON>PHILOSOPHY-OF-SCIENCE-ARCHIVE.TXT
please note that this OZ file is written in ZMAIL format, and is not
readable with either MM or BBOARD.EXE. ZMAIL is a LISPMachine mail
reader from MIT. You can TYPE or edit ZMAIL files, but they are
sometimes pretty hard to follow that way.
If you are interested in looking at back issues of this list in a more
civilized fashion, I have been following it from the beginning, and
have a home-built archive archived (howzat again?) on GREEN, as
I-PHIL-SCI.BABYL through VI-PHIL-SCI.BABYL. These files have been
reformatted for convenient reading with the BABYL, an EMACS-based
mail-reader available at Rutgers. Also archived on GREEN is a help
file named
USING-BABYL-TO-READ-PHIL-SCI.HLP.
Please do not attempt to RETRIEVE this stuff; drop me a note instead.
These files total several hundred pages and would swamp my GREEN
directory if restored to disk all at once.
_Bob
------------------------------
Date: Tue, 17 May 83 19:12:45 EDT
From: Mark Weiser <weiser@NRL-CSS>
Subject: Computer Resident Intelligent Entity (CRIE) [Long Article]
1. The Operating System World
An interesting test-bed for Artificial Intelligence (AI) methods
is the world of computer systems. Previous work has focused on
limited particular subdomains, such as digital design [Sussman 77],
computer configuration [McDermott & Steele 81], and programming
knowledge [Waters 82]. Even these restricted domains have proven
themselves very rich areas for AI techniques. However, no one has
(yet) gone far enough in applying Artificial Intelligence techniques
to computer systems. The far out question I'm thinking of is: what
sort of entity would live in the ecological niche supplied by the
computer system environment?
Organisms evolved in the biological world have been shaped
primarily by evolutionary forces. They cannot be wholistically
studied without considering, for instance, their energy intake and
expenditure and their necessity for reproduction [Kormondy 69]. These
particular constraints are biological universals, but are not
necessariy paradigmatic for non-biological intelligent organisms.
Consider human beings, necessarily the prime subjects of those
studying intelligent biological organisms. We* are specifically
attuned to a particular environmental niche by virtue of our sensory
systems, our cognitive processing capabilities, and our motor systems.
Dreyfus [Dreyfus 72], argues from this that machines cannot be
intelligent. Our discussion begins from a view more akin to
Weizenbaum's [Weizenbaum 76]: a machine intelligence is an alien
intelligence. What sort of sensory system is appropriate to this
particular alien intelligence?
2. Traditional perceptual interfaces to the computer world
The usual way of observing a computer system is to take
snapshots. Such a snapshot might be a list of the active jobs on the
system, or the names and sizes of the files, or the contents of a
file. If more information than a snapshot is needed, then many
snapshots are packed together to create a "history" of system
behavior.
Unfortunately a history of snapshots is not a history of the
system. This is well known in performance modeling of computer
systems, where a snapshot of a system every 15 minutes is useless.
Instead an average over the 15 minute interval is the proper level of
information gathering. The problem with snapshots is their time
domain is fixed externally and irrelevantly to the world being
monitored.
It is sometimes possible to recreate the behavior of system
objects by examining a stream of snapshots of the object's states.
But this is the wrong approach to the problem. Rather ask: what sort
of perceptual system would best notice the important objects
(invariants) in a computer system world [Gibson 66]? A snapshot
contains irrelevant information, and is gathered at irrelevant times.
3. New perceptual interfaces
Imagine your favorite computer system. It consists of objects
changing in time: files, programs, processes, descriptions, data
flowing hither and yon--a very active world. A retinal level
description of the biological world would display a similar confusion
of unintegrated sensations. But our retina wins because it is part of
a perceptual system which quickly transforms the input flux to
invariant forms.
Let's ignore the back end (invariant deduction end) of a computer
perceptual system for a moment, and consider just the "retinal" end.
What kind of raw data is available about important system activities?
On the one hand are the contents of files, data structures, program
descriptions, etc. The understanding of these items is relatively
well studied--as a first approximation it is what programs do. The
hard problem is perceiving the information flux. Values in memory and
files are constantly changing and often it is the changes themselves
which are interesting, more than from what the value was changed or to
what it was changed. For instance, noticing someone poking around in
my files is a "who is looking" question rather than a data value
question. Noticing important changes in the system requires an
event-based perceptual system.
Activities occur in widely distributed places in a computer
system. User programs, file systems, system data structures, may all
be relevant to the intelligent computer resident entity. The human
visual system has evolved to make good use of the transparency of our
atmosphere to electromagnetic radiation of a certain wavelength to
allow us to perceive activities in a wide range around us. A great
deal of our intelligence is oriented towards the three dimensional
space which we can survey, because it is here that we have effortless
access to information about the objects which can immediately affect
us [Kaplan 78].
A computer entity must also have effortless access to information
about objects in its area of prime concern. Its perceptual apparatus
should be attuned to changes in those entities so interesting events
are immediately apparent to it. With our current technology** one
solution is to distribute the perceptual apparatus of the entity onto
the objects of concern. This is radically different from any solution
chosen by nature, but the computer system world is radically different
from the biological world. It amounts to daemon-based perception.
The perceptual mechanism of a computer resident intelligent
entity (CRIE) WOULD be similar to production rules [Forgy 81] and
daemons [Rieger 78]. A CRIE retina would have two distinctive
features: (1) it is made up of demons, which are (2) attached to the
objects being observed.
A CRIE perceptual system is quiescent until some event occurs to
which it is attuned. When that happens, a CRIE reacts by invoking
various reasoning and acquisition daemons associated with that event.
These reasoning and acquisition daemons are modular pieces of
information which are the low level meaning of events within a CRIE.
The daemons not only watch for events occurring on the system, but
also can observe larger contexts (such as themselves).
To conclude: Artificial Intelligence research has, as one goal,
understanding how to embed intelligence in a machine. The criticisms
of AI from Dreyfus, Weizenbaum, and others can be used constructively
to design an intelligence appropriate to a machine. This approach to
intelligent system design leads to new kinds of design constraints for
computer perceptual systems, and gives new meaning to the term
"computer vision".
FOOTNOTES
*With apologies to those readers who are not human beings.
**Implementation issues are important for the design of any intelli-
gent entity. Why are our eyes in our head?
REFERENCES
[Dreyfus 72]
Dreyfus, Hubert, What Computers Can't Do, Harper and Row, 1972.
[Forgy 81]
Forgy, C. L., OPS5 User's Manual, Carnegie-Mellon University
CMU-CS-78-116, 1981.
[Gibson 66]
Gibson, James J., The Senses Considered as Perceptual Systems,
Houghton Mifflin Company, 1966.
[Kaplan 78]
Kaplan, R., The green experience, in Humanscape: environments for
people, ed. S. Kaplan and R. Kaplan, Duxbury Press, North
Scituate, Mass., 1978.
[Kormondy 69]
Kormondy, Edward J., Concepts of Ecology, Prentice-Hall, 1969.
[McDermott & Steele 81]
McDermott, J. and Steele, B., Extending a Knowledge-Based System
to Deal with Ad Hoc Constraints, Proc. IJCAI-81, Vancouver, BC,
1981.
[Rieger 78]
Rieger, C., Spontaneous Computation and Its Role in AI Modelling,
in Pattern-Directed Inference Systems, ed. Waterman & Hayes-Roth,
Academic Press, New York, 1978.
[Sussman 77]
Sussman, G., Electrical Design: A Problem for Artificial
Intelligence Research, Proc. IJCAI5, Cambridge, MA, 1977.
[Waters 82]
Waters, R. C., The Programmer's Apprentice: Knowledge Based
Program Editing, IEEE Trans. on Software Eng. SE-8, 1, January
1982.
[Weizenbaum 76]
Weizenbaum, Joseph, Computer Power and Human Reason, W.H. Freeman
and Company, 1976.
[Editors comment:
Mark doesn't seem to be asking about the natural course of evolution
in a digital environment, although that is also an interesting
question. It is not clear to me whether he is proposing a life form
with the usual survival goals, or a monitoring system built by design
and serving some useful purpose. Since it is difficult to discuss
such a thing without knowing its purpose, I suggest that anyone
responding state his own assumptions or teleology.
I think the new LOOPS language/environment at Xerox offers much of the
"instrumentation capability" that Mark's CRIE needs. The software
probes can be attached to any variable a posteriori, in the manner of
a dynamic debugger. This opens up a world of data-based (or dataflow)
techniques integrated with rule-based and other AI techniques.
-- KIL]
------------------------------
End of AIList Digest
********************
