LAWS@SRI-AI.ARPA (05/13/85)
From: AIList Moderator Kenneth Laws <AIList-REQUEST@SRI-AI>
AIList Digest Monday, 13 May 1985 Volume 3 : Issue 63
Today's Topics:
Queries - Requirements Decomposition &
Connectionism and Parallel Distributed Processing,
Binding - Walter Reitman,
Games - Nim,
Expert Systems - Prospector on a PC,
Psychology - Emotional Attachment & Reason and Emotion &
Emotions and Memory & Simulation of Human Understanding
----------------------------------------------------------------------
Date: Thu, 9 May 85 09:21 CDT
From: David_Lagrone <lagrone%ti-eg.csnet@csnet-relay.arpa>
Subject: Requirements Decomposition: Correctness and Consistency
Can someone help me with a solution or even an approach to a solution to
the following problem?
Given a set of requirements; e.g., a requirements specification,
and a "first-level expansion" of those requirements into a top-
level design. Can one "prove" or otherwise demonstrate whether
the lower-level, more detailed requirements are a "necessary and
sufficient" statement of the related higher-level requirement?
It has been pointed out that this problem may mean "are you trying
to find out if all requirements mentioned in higher levels are talked
about in lower levels (by actual presence)" or "if things mentioned
in lower levels contain the right content or meaning to fulfill higher
level requirements."
I am more interested in a solution to the "content analysis" problem; how-
ever, the "presence" problem needs resolution as well.
I would appreciate responses being sent to me at:
LAGRONE%TI-EG@CSNET-RELAY
Thank you, very much, for your time and help with this.
...Regards...David LaGrone
------------------------------
Date: Fri, 10 May 85 16:14:56 est
From: "Marek W. Lugowski" <marek%indiana.csnet@csnet-relay.arpa>
Subject: Info-request: Connectionism and Parallel Distr. Processing
This is a general info request concerning the design of a graduate AI course.
Over here at Indiana University CS Dept. we are working on the updating our
AI curriculum with a graduate-level course to be called "Connectionism and
Parallel Distributed Processing". A prerequisite for this course would be our
standard 2-semester sequence of AI courses taken both by undergraduates and
graduates. It involves a hefty amount of programming in Lisp. The new course
would have a programming project, too. And now, the questions:
(1) Whose AI work do you feel is most appropriate for this course?
(We're thinking along the lines of the Hinton/McClelland/Rumelhart/Sejnowski/
Smolensky axis, and Hofstadter. Others?)
(2) What textbooks, if any, would you recommend? Or should the course
be based on papers? Which ones?
(3) Could you recommend a course along those lines already taught
elsewhere? Perhaps you could send us a course outline?
Thank you. Please reply to me. Will summarize if desired. -- Marek Lugowski
IU CS Department
marek@indiana.csnet
------------------------------
Date: Wed, 8 May 85 18:36 EDT
From: Gibbons@MIT-MULTICS.ARPA
Subject: Binding: Walter Reitman
A friend of mine noticed an attempt to bind Walter Reitman in Volume 3,
Issue 47 of the AI digest. I can provide current information for you.
Walter spent some time at New York University and then, in the fall of
1983, became the department manager for the Artificial Intelligence
Department at Bolt Beranek and Newman in Cambridge, MA where he replaced
Bill Woods. In the fall of 1984, he left BBN to become the Vice President
of Palladian Software, Inc., also in Cambridge, where he has been, and is
responsible for, assembling the technical staff. Several other BBN
employees, some from the AI department (including myself), have since
joined Palladian as have others in the community.
For your information, Palladian Software is a well funded and fast growing
startup developing very sophisticated applications in the domains of
finance and manufacturing. We are constructing hybrid systems using
appropriate mixtures of both conventional and artificial intelligence
technologies. We are currently developing on Symbolics 3640's (every
developer has his/her own machine - and office).
If you wish to contact Walter, or Palladian, messages can be sent to me,
Jeff Gibbons. My arpanet address is Gibbons at MIT-MULTICS.
jeff
------------------------------
Date: Thu, 9 May 85 12:16:06 edt
From: Dana S. Nau <dsn@tove>
Subject: Nim
I know of two different games that are sometimes called "nim". The first
one (the _real_ nim) involves splitting piles of sticks into smaller piles
of sticks in such a way the the smaller piles contain differing numbers of
sticks.
The second game involves removing various quantities of sticks from a pile of
sticks until none are left. This game is called Bachet's game, but it has
sometimes incorrectly been called nim--even in such well-known books as
Horowitz and Sahni's "Fundamentals of Data Structures Using Pascal".
I imagine the request for information on nim programs was for the real nim.
But in the case of Bachet's game, it's trivial to tell whether a game
position is a forced win or forced loss (for example, see "Mathematical
Games and Pastimes" by Domoryad), and thus it's very easy to write a
computer program to play the game perfectly.
------------------------------
Date: Thu 9 May 85 10:57:28-PDT
From: Joe Karnicky <KARNICKY@SU-SCORE.ARPA>
Subject: Prospector on a PC
The April 29, 1985 issue of COMPUTER ENGINEERING contains the following
statement in an article entitled AI MOVES FROM LABS TO PERSONAL COMPUTERS
"The power of AI is illustrated by Stanford Research International
Inc.'s Prospector, a so-called knowledge-based system. Written in
Lisp, it is considered a classic example of AI.
This program, which integrated the knowledge of many US Geological
Survey experts, was responsible for suggesting the existence of
a molybdenum deposit in Cascades, Wash., a find estimated to be
worth more than $100 million. It also exemplifies AI's future
direction. The system can now be implemented on a personal computer
for less than $3000 with available software."
This rather startling statement deals with two issues about which I would
very much appreciate receiving information from readers of the AIlist:
(1) What exactly was the role of Prospector in this discovery?
Would the discovery have been made without the program? How much of the
discovery was made by the program, and how much by the programmers?
If Prospector is so competent, are other geologists using it? If not, why
not? I've never seen a thorough discussion of these issues, and
responses by AIlist readers who are familiar with the project would be very
welcome.
(2) How good an expert system can be run on a pc using one of the
many commercial tools available? What practical expert systems
have been created that run on pc's? I've seen a lot of articles describing
expert system software for pc's, but what useful systems have been/can be
contructed? It's a long jump from choosing a red wine with fish to
discovering a $100 million ore deposit. Again, any information would be
appreciated.
[I can provide a little info on the first query, although I
have not been closely connected with Prospector. First, SRI
International has not been Stanford Research Institute since
May 16, 1977. Prospector was primarily developed by John
Gashnig, Peter Hart, Dick Duda, and Rene Reboh, although others
have contributed (including M. Einaudi, a USGS geologist).
John died of lung cancer; Peter, Dick, and Rene have moved on to
found Syntelligence. Prospector lives on, but is not under active
development; the original code has suffered from hardware changes
and bit rot, but versions have been ported to many systems.
I don't know whether USGS is adding additional mineral
models to its repertoire. The geologists working on the
project seemed to feel that this was a useful exercise,
whether or not the program ever became an expert geologist.
As for it fitting on a PC, I'm not too surprised -- the core
of the program is a [compiled] inference net that does not grow
during execution. About the only thing that does grow is the
history list used for the explanatory capability. I would assume
that most of the difficulty in porting Prospector is in providing
the software tools for editing graphs and knowledge bases.
The story I heard about the molybdenum strike was that this
particular site was fed into Prospector because geologists
were already convinced that molybdenum should be there -- they
just hadn't been able to find it over several decades of
exploratory digging. Humans were thus "responsible for suggesting
the existence" of the deposit, as well as for loading in the domain
models, probability functions, and field data. What Prospector did
was to highlight a spot under a large pile of tailings as being
the best place to dig; sure enough, that's where the molybdenum was.
A strike like that more than pays for the system's entire development
cost, a fact which was not lost on the oil industry -- Schlumberger
and others soon started large in-house projects. The majority
of people interested in Prospector, however, have been looking for
an off-the-shelf expert system capable of reasoning in any
domain. ("Sure, we'll just rip out all that stuff about uranium
and load in some knowedge about rubber tires. Anything else?")
There has been surprisingly little interest in supporting the
development of new uncertain-reasoning techniques appropriate for
other problem domains. That is one reason Prospector has not
been overtaken by a new generation of expert systems.
Can anyone answer Joe's question about how sophisticated a PC
expert system can be? -- KIL]
------------------------------
Date: 5 May 85 13:30 EDT (Sun)
From: _Bob <Carter@RUTGERS.ARPA>
Subject: Emotional Attachment
From: MINSKY%MIT-OZ at MIT-MC.ARPA
Bob Carter has answered Batali's objections
to my "assault" theory better than I would have. Thanks.
You are more than welcome. The theory has a verisimilitude that
attracts me, and I look forward to reading your book when it is
published.
But I am bowing out of this discussion and leaving it to those who
prefer to see the world in clearer tones of black and white than
I do.
They are right. Ideas @i(are) dangerous.
_B
------------------------------
Date: Mon, 6 May 85 8:57:20 EDT
From: Bruce Nevin <bnevin@BBNCCH.ARPA>
Subject: reason and emotion
Reason is a good servant but a lousy master.
Benjamin Franklin it was who said
What a wonderful thing it is to be a rational being: for we can
make up a reason for whatever we have decided to do!
We have this truly marvellous ability to reason, and we use it most of
the time to rationalize. We freely invent plausible premisses, complete
with supporting histories, to make our cherished beliefs into reasoned
conclusions. We can do this only by maintaining selective ignorance, by
ignoring data that do not fit.
My own impression is that artificial intelligence systems mirror the
choices made by many persons who enter fields of science and
engineering, in that they model only the intellectual fraction of human
reasoning ability.
Human belief-disbelief systems depend upon a variety of logics. (I am
thinking of research of Milton Rokeach.) Emotions have their own logic,
and this logic drives our ratiocination far more than many of us are
comfortable acknowledging.
At the risk of being trite, let me affirm that there is nothing wrong
with emotional motivations or emotionally-directed reasoning. It is
when we keep the emotional richness of our reasoning out of conscious
awareness that something is wrong, because we have hobbled ourselves.
Then there can be a great deal wrong with our reasoning and with our
responses to experience. We cripple ourselves by cutting off a limb of
our reasoning ability, and then attribute our distress and dysfunction
to the alleged unruliness of the very limb we have tried to amputate.
Integration is not a matter of reconciling contrary things. It is not
nearly so complicated. Integration is rather a matter of recognizing
unity from multiple perspectives. Binocular vision provides a
simplistic model. With both eyes you see, not two conflicting images
somehow reconciled, but one image with depth added. The third dimension
is not predictable from either single image by itself. Allowing the
full breadth of our premisses and our logic brings the equivalent of
depth perception to our conclusions. And to our belief-disbelief
systems.
The self-insulated intellect can be very alarmed by this kind of talk.
Quite emotional, in fact, in its cold, steely way. How difficult it can
be to realize that this is not a Darwinian competition (poor Charles
Darwin! He said that cooperation was far more important than
competition in evolution, but his audience heard what they wanted to
hear . . . ), not an either/or choice between intellect and that
nameless other, but the removing of a patch over one eye!
(Does the artificially intelligent computer look like a tyrannical
pirate to the nontechnical majority of the world?)
This relates very much to the `false sense of power' noted by M.
Schoppers (marcel@SRI-AI) in his note on `living programs' (3.45), to
our commonly shared Pollyana-ish beliefs about self and the world noted
by jmyers (3.56), and to the call for some reality-checking issued by
the rape survivor/counselor from Berkeley (3.56).
It is difficult to go through childhood without experiencing deep
humiliation--sometimes physical rape--at the hands of ignorant,
half-blind adults. Because of its power, and especially because it
keeps reality at arms length, we often sieze upon intellectual prowess
as a means never again ever to be so humiliated. Altogether too often
we carry with us a need to humiliate others `first'. This may be
difficult to bring to conscious awareness in ourselves. Until we do it
consciously, however, we cannot choose whether to do it or not.
And this ignorance is not bliss.
Something to consider during the discussion period after a paper or
research proposal has been presented!
------------------------------
Date: 7-May-85
From: Wolf-Dieter Batz <L12%DHDURZ2.BITNET@WISCVM.ARPA>
Subject: emotions & memory - public discussion
Hello,
Thanks a lot to all the folks who replied to my last
advertisement in this place.
To everyone who's interested but still didn't write:
Please don't write to my personal address, but to the
Digest - I'm convinced that there is enough stuff in our
minds to be discussed in public. - ok?
Next point:
My thesis is written in German and is about 150 pages long.
Anyway it is stored on disk so that it can easily be
transmitted by means of network exchange. Such requests
will be answered as fast as possible, but in general I won't
transmit it all - so tell me about special questions.
So long Folks & kind regards *** Wodi <l12@dhdurz2.bitnet>
------------------------------
Date: 8-May-85
From: Wolf-Dieter Batz <L12%DHDURZ2.BITNET@WISCVM.ARPA>
Subject: Re: C. Mason on Simulation of Human Understanding
While working in the field 'History of Memory Research'
I discovered the writings of J.F. HERBART, a German philosopher
living from 1776-1841.
Writing about 'Psychologie als Wissenschaft neu gegruendet
auf Erfahrung, Metaphysik und Mathematik' he didn't use
the term 'memory', although he dealt with various topics
of information processing.
My very personal opinion is now:
HERBART's theory is a theory of memory. He just uses various
synonyms for this term, like ideas, images and so on.
If one substitutes this synonyms with 'memory' he got it made!
This indicated to me that there's no possibility theorizing
about HUMAN Information Processing without refering to memory
mechanisms.
Anyway this is a debatable point, and I'm looking forward
to public replies to this proposal.
So long folks & kind regards *** Wodi <l12@dhdurz2.bitnet>
------------------------------
End of AIList Digest
********************