AIList-REQUEST@SRI-AI.ARPA (AIList Moderator Kenneth Laws) (10/09/85)
AIList Digest Wednesday, 9 Oct 1985 Volume 3 : Issue 139
Today's Topics:
Corrections - AI at GE & Attenber's Research Field,
Opinion - AI Hype
----------------------------------------------------------------------
Date: 7 Oct 85 11:29 EDT
From: WAnderson.wbst@Xerox.ARPA
Subject: Correction: Intellectual Honesty & the SDI
In a message posted in AIList Digest on Friday, 20 Sep 1985, Volume 3,
Number 125, I commented about statements made in two papers I picked up
at the GE exhibit at IJCAI-85.
In view of subsequent discussions with people at GE I wish to state that
the authors of these papers are NOT in any way connected with GE. It
was not my intent to cast aspersions on work done by GE. I wish to
correct any misconceptions people may have about the type and quality of
research at GE from my message.
Bill Anderson
People interested in GE AI R&D, may contact
Dr. Larry Sweet
K1-5C13
General Electric Company
Corporate Research and Development
Schenectady, NY 12301
Dr. Sweet is the manager of the AI group here.
------------------------------
Date: Fri, 4 Oct 85 12:24 EDT
From: Attenber%ORN.MFENET@LLL-MFE.ARPA
Subject: typographic error
In case anyone cares, the second paragraph of my unintelligible
note of Sep.25 should have started "as a researcher in plasma physics"
not "as a researcher in particle physics". Sorry.
------------------------------
Date: Fri, 4 Oct 85 08:57:33 EDT
From: George J. Carrette <GJC@MIT-MC.ARPA>
Subject: ai hype vs profitable use
Observations I've been able to make (from inside and out) of the
activities of LMI's process control division may be interesting here.
(1) They call it PICON(tm) for Process Intelligent Control.
The world "artificial" hardly ever comes up. The purpose is to
more intelligently control the industrial process.
(2) The first installations at Exxon and Texaco were in place for
months before they even told anyone.
(3) Promise of profits were never made. The marketing was as an
ALARM condition detector/advisor. More subtle and possibly dangerous
conditions could be detected than with existing technology, and
more ALARM conditions could be handled at once, with more intelligent
selection of danger priorities. The purpose is of course to avoid
the 3-MILE-ISLAND-EFFECT.
(4) Once in place it was the customers that realized for themselves
that with the more intelligent modeling and flexibility in PICON
they could optimize the control of the process more closely,
and could start to save a percent or two in cost or get a percent or two
higher yield. Obviously in oil refineries this can translate into
big paydirt far above the cost of a few lispmachines and development costs.
The "downplaying" the PICON people have had to do is of course caused
by all the previous and continuing ai hype. This is probably why
they dont use the term AI very often.
------------------------------
Date: 4 Oct 85 09:39 PDT
From: allmer.pasa@Xerox.ARPA
Subject: AI Hype
Does anyone really expect truth in advertising?? Whenever I read a blurb
about some new "fully compatible" DBMS I always hear that click, you
know, what do they mean by "fully-compatible". So I don't see the point
for ranting on about the "AI Hype". It seems to me that if there was
nothing more than hype, there wouldn't be an AIList, or AIList
Moderator, no one would want to learn any "AI techniques" or take any "AI
courses", no one would spend millions of $$$$ to get "AI Systems", etc.
There's got to be more to AI technology than the hype, or this is the
greatest scam in history (Mr. Guiness, where are you?). At the Expert
Systems panel for IJCAI85, Terry Winograd, (how's that for
name-dropping), made mention of the "illusion of the label 'expert
system'", that because we choose to call them that, they should be
as-good-as/better than 'experts' when they are finished, sort of a
"magical black-box" mentality. To sell the concept, one does not call
their proposed system "idiot savant", or "limited-domain model", even
though that is what the buyer ends up with. You just have to understand
what is meant by the terminology of the "hype".
Doug Allmer
------------------------------
Date: 4 Oct 1985 1203-PDT (Friday)
From: jeff@isi-vaxa.ARPA (Jeffery A. Cavallaro)
Subject: AI, SCI, and SDI
After observing the various discussions regarding the place or state
in today's SCI/SDI-hyped environment, I think that it might be wise to
mention a chain of events that I believe has gone virtually unnoticed
in the AI community. The opinions expressed here are based on my
connections within the REAL research world (academia, not the defense
contractor flops referred to as IR&D), and the defense/aerospace sector
over the past 5 or so years. (Limited compared to some I admit)
About 5 years ago, when DARPA was still being referred to as ARPA, DARPA's
commonly stated goal was to increase America's strength by promoting
raw research that would trickle into the ECONOMIC sector. That may seem
like a rather empty statement based on today's activities, but that
was really the way it was. (Enter Dick Cavett).
ACADEMIA (5 YEARS AGO):
Around the same time (5 years ago), the various VLSI research groups around the
nation had their own mini-SCI, it was called the Silicon Compiler Project.
It may not have gotten the same attention, because the funding stakes
were probably not as high as AI today. It can easily be stated that
SCP was a success.
Meanwhile, the research AI groups of the day were constantly complaining that
they were spinning their wheels. The results of their work were constantly
being blocked from entering the marketplace for a variety of rights and
economic reasons.
DEFENSE/AEROSPACE (5 YEARS AGO):
In defense/aerospace, VLSI technology is basically unheard of. Companies
like TRW (Torrance Rubber Works), HUGHES, and the like, were more interested
in off-the-shelf solutions from large vendors such as DEC, IBM, etc.
These vendors, in turn, had representatives participating in SCP at
various research institutions.
AI was a different story. Defense contractors had AI projects (and funds)
coming out of their ears. One of the largest such project was BETA-LOCE,
an in-the-field battle management-type system. All such projects, without
many exceptions, were unqualified FAILURES.
TODAY (Exit Dick Cavett):
Sitting in meetings at ISI today, where the goals of SCI/SDI are being
stated, is like sitting in a status meeting at TRW as little as 2
years ago.
VLSI, having been successful, has now obtained a firm place in defense/
aerospace. Due to the strong success and firm base achieved while it
was in the researcher's hands, VLSI is proving to be an excellent tool
now available to the real world.
AI projects in defense/aerospace have hit hard times. SCI is now on hand.
I believe that one of the unspoken goals of SCI (and SDI) is to seriously
shift emphasis of AI projects back to the research institutions, hoping
that they can achieve success similiar to VLSI.
But, a funny thing is happening. The barriers to their work have been
lifted, but the AI world is still complaining. They are disatisfied
with the end user (as if Defense wasn't actually the end user all along).
The VLSI researches didn't seem to have the same qualms. But then again,
maybe they weren't faced with SDI-type goals.
In conclusion (finally), the AI research community is currently in an
EXCELLENT position. They now have the attention (which, in a way, is just as
important as funds) that they have desired for a long time. Of course,
careful consideration is needed, but continued complaining and temper
tantrums of "We can't do that in the foreseeable future, so I don't even
want to try!!" will simply be detrimental to ALL parties involved.
(Oh, am I going to get jumped on for this one!!!)
Jeff
------------------------------
Date: Fri, 4 Oct 85 13:18 EDT
From: Jeffrey R Kell <JEFF%UTCVM.BITNET@WISCVM.ARPA>
Subject: Hype and success/failure
The 'hype' of AI in terms of systems to carry on conversations or drive
cars or whatever is largely based on an idealistic projection of what we
have achieved in the directions we want to follow. It began with the
stereotyped image of a computer as a thinking being when in fact it was
a primitive set of vacuum tubes. The AI image suggests that we can do
just about anything; that we can eventually achieve that sort of goal.
AI researchers realize their current successes, albeit not on such a
grand scale, with a guarded skepticism. If the shortcomings of any AI
project are stressed, the image sways in the other direction - maybe AI
can do nothing.
The relative success/failure of such projects seems to revolve around a
gap between the ideal and the practical. A 'pure' AI system built with
'pure' AI processes is almost doomed to certain shortcomings whether in
speed, size, usability, correctness, completeness, or what have you.
Some practical (traditional) methodologies must be employed to narrow
the scope of the project. Universal problem solvers were certainly not
a practical success, but they did lay the groundwork for expert systems
which have been successful. But the balance between AI and traditional
methods varies, and disciples of 'pure' AI will argue that many expert
systems are not AI at all. Perhaps not, from a purist view, but they
would not have been possible without the conceptual contributions from
the AI field. It is THESE forms of contributions that will be of the
most lasting value, whether you view them as an idealistic success or
not.
------------------------------
Date: Sat, 5 Oct 1985 23:55 EDT
From: MINSKY%MIT-OZ@MIT-MC.ARPA
Subject: AIList Digest V3 #132
I was not planning to prolong this discussion, but I can't resist
pointing out the disgraceful lengths that Gary Martins will go to
prove himself right. If you will examine my statement and then
Martins', you'll [note] an astonishing performance: he takes two
of my sentences, ONE OF WHICH CAREFULLY QUALIFIES THE OTHER,
breaks each of them into clauses and than attacks each clause by
itself! I see no room in a professional discussion for that degree of
intellectual and rhetorical dishonesty. Talk about "hype!"
Then he has the bad taste to talk about
"utterly non-'AI' software" that keeps track of payrolls, arranges
airline reservations, manages power distribution grids, guides
missiles, allocates resources, monitors inventories, analyzes radar
signals, does computer animation, assists in mechanical design and
fabrication, manipulates spreadsheets, controls space vehicles, drives
robots, integrates CAT scans, and performs lots of other mundane
tasks. and about nicely engineered non-"AI" systems that play
world-class chess.
The latter "non-AI" chess programs are, of course, essentially the AI
chess programs of the 1960's, based on Shannon. Samuel, and McCarthy's
tree-pruning heuristics and plausible move generators. The robot
drivers are mostly based on the early MIT, SRI, and Stanford
prototypes. Many of the aircraft control systems are based on the
adaptive algorithms developed in that general community in the same
period, and everyone knows the origins of much of computer graphics in
the early work of Sutherland, Knowlton, and many others in the AI
community. As for those missile guiders, the roots of that whole
field now called "pattern recognition" have similar origins. And I'm
pretty sure that the first practical airline reservation was designed
by Danny Bobrow of the BBN AI group around 1966.!
I'm not claiming that AI set the stage for accounting programs, and
some of the others. But don't you agree that Martins could have made
himself a better case by mentioning a few first-rate programs that
didn't have substantial roots in the AI of 15 to 20 years ago. If
there's anything wrong with the present-day AI hype, it's simply that
some people may be led to expect various goodies in 3 years instead of
15 -- and perhaps that's what we ought to tell people.
Time to take a course in the history of AI, Gary.
------------------------------
Date: Mon, 7 Oct 85 15:54:22 mdt
From: ted%nmsu.csnet@CSNET-RELAY.ARPA
Subject: ai/military flame
When did the military manage to take credit for automobiles????
Or for that matter how do they manage to take credit for debugging
and field testing the interstate highway system when there was a
civilian highway system in place before the interstates and as
far as I know, the only connection was that internal defence is
a standard rationale for better internal transportation?
Minsky's comments about what ai can do and other software
can't are very illuminating when compared with the real
world in the form of the first milestone test of arpa's
autonomous land vehicle in denver this spring. one might
think that this would have provided a perfect example of
the way that ``ai software can ... drive a car''.
unfortunately for true fans, the ai approach to driving the
vehicle (line extraction, motion field analysis and so on)
turned out to be very difficult (read as late) to implement
so that the prime contractor (who is very capable in conventional
software) implemented a VERY conventional system which selected
``gray'' pixels from the television image and managed to steer
the vehicle toward the center of mass of the gray. doesn't
this sound more like an example of conventional software doing
something (not very well, but literally good enough for government
work) that ai type software failed to do???
------------------------------
Date: Thu, 3 Oct 85 14:09:20 GMT
From: gcj%qmc-ori.uucp@ucl-cs.arpa
Subject: Mega-Hype
A comment from Vol 3 # 128:-
``Since AI, by definition, seeks to replicate areas of human cognitive
competence...''
This should perhaps be read in the context of the general discussion which
has been taking place about `hype'. But it is still slightly off the mark
in my opinion.
I suppose this all rests on what one means but human cognitive competence.
The thought processes which make us human are far removed from the cold
logic of algorithms which are the basis for *all* computer software, AI or
otherwise. There is an element in all human cognitive processes which
derives from the emotional part of our psyche. We reach decisions not only
because we `know' that they are right, but also because we `feel' them to
correct. I think really that AI must be seen as an important extension to
the thinking process, as a way of augmenting an expert's scope.
Gordon Joly (now gcj%qmc-ori@ucl-cs.arpa
(formerly gcj%edxa@ucl-cs.arpa
------------------------------
End of AIList Digest
********************