Copy Link
Add to Bookmark
Report
AIList Digest Volume 2 Issue 093
AIList Digest Thursday, 19 Jul 1984 Volume 2 : Issue 93
Today's Topics:
Bindings & Humor - Small Computer Lisps,
Programming Languages - AI Language for Parallel Machine,
Expert Systems - Commodity Experts,
Commonsense Reasoning - Cultural Influences,
AI Jargon & Philosophy - Definitions,
Intelligence - Measurement by Logical Inferences,
Administrivia - Advertisements,
Demonstration - GMR DATALOG Demo at AAAI,
Games - Chess Experiment
----------------------------------------------------------------------
Date: Tue, 10 Jul 84 12:39:55 pdt
From: Shimon Cohen <Shimon>
Subject: Yet another new language ?
[Forwarded from the Prolog Digest by Laws@SRI-AI.]
We, at Fairchild AI laboratory, are in the process of designing a
super parallel computer for AI applications (called FAIM). In the
process we have to define the language that will be used for this
machine. We intend to summarize our initial ideas in a working paper
and distribute it to interested people.
If you (or someone you know) are interested in a copy of this working
paper please mail your name and full address to: Shimon@SRI-KL.
Thank you.
-- Shimon Cohen
------------------------------
Date: 16 Jul 84 21:22:13-PDT (Mon)
From: ihnp4!mhuxl!ulysses!gamma!pyuxww!pyuxa!diamant @ Ucb-Vax.arpa
Subject: Re: Small Computer Lisps?
Article-I.D.: pyuxa.882
For anyone interested, the net addresses of the authors of the
previously mentioned BYTE review of muLISP and IQLISP:
Jordan Bortz: decvax!cbosgd!osu-dbs!gang!jordan
and
John Diamant: ihnp4!pyuxgg!diamant (until Aug. 10th)
decvax!cwruecmp!diamant (after that)
I apologize for posting this to the net, but it was originally in the
article and was accidentally edited out.
John Diamant
ihnp4!pyuxgg!diamant
------------------------------
Date: 13 Jul 84 11:29:54-PDT (Fri)
From: hplabs!sdcrdcf!sdcsvax!akgua!mcnc!unc!ulysses!mhuxl!mhuxm!sfjec!
sfmag!eagle!prem @ Ucb-Vax.arpa
Subject: Re: Small Computer Lisps?
Article-I.D.: eagle.1180
I certainly wouldn't expect a grown up computer to.
------------------------------
Date: 18 Jul 84 10:28:01 PDT (Wednesday)
From: Cherry.es@XEROX.ARPA
Subject: Commodity Experts
Jan, The input of the wide areas of data (fundamental and technicial
plus current new items) into a AI enviromnent for Stock/Commodity
trading may be a little too much to start with. You can however, input
only technicial data into the database and use a wide variety of
technicial indicators to evaluate the trend for short term, long term,
etc. This will also aid in making market projections using technicial
indicators such as on-balance-volume ior OBV (Granville),
Parabolic-Time-Price Curves (Wilder), etc. Due to the large number of
different technicial indicators for the various markets (OBV is fine for
Gold but not for Silver), AI could be very well suited for this kind of
application. This is primarily due to hard data for technicial analysis
(Open, Hi, Lo, Close, Settle, Volume, Open Interest, Interest rates,
etc.) This information is also available from many dial-up data bases
for automatic entry for the particular market. By utilizing an
algorithm which would first use all methods programmed for each analysis
and then take the results and utilize them for further analysis, a
reasonably accurate projection may be possible based on only technicial
indicators.
Fundamental information is harder to deal with. Analysis based on
fundamental knowledge requires that some effect either positive or
negative be applied to a repeatable phenomenia. Since news items often
do not repeat, the AI software would first have to separate the valid
fundamental indicators from the other fundamental noise. Over a period
of time, certain phenomenia may be evaluated as being pertinent to the
trend of a given market. At this point, any tool could be written to
search out this "signal" from the "noise". In order for any tool to be
effective, the tool would be required to have a high signal to noise
ratio. The tool must be able to ascertain, for example, if a member of
the FRB says he thinks the discount rate will decline,what will be the
immediate and future impoact on the price of soybeans? Since
historically statments of this type have caused both higher and lower
prices due to other fundamental variables, a fundamental formula would
have to be derived which works with total unknown abstract concepts.
The Technicial Analysis Group (TAG) has a very complete package of
technicial software however, due to the immediacy of the markets, and
the small systems (Apple, IBM-PC) which the programs run on, it is not
feasible to do a complete analysis on every market on a daily basis. A
multi-tasking AI environment would be able to take the results of each
of these TAG tools and then work with those results.
I would recommend that any attempt to utilize AI for the purpose of
market analysis should start with only technicial analysis and that the
AI environment gradually be fed only certain types of Fundamental data.
Bob
------------------------------
Date: Wed, 18 Jul 1984 11:18 EDT
From: REID%MIT-OZ@MIT-MC.ARPA
Subject: Commonsense Reasoning
Regarding the Fahrenheit/Celcius problem (if 32 is 0 and 212 is 100 ...).
Even there, the "obvious" answer is only obvious due to cultural
biases. A computer might indeed solve the problem without "blinking an
LED," but the answer it would likely to come up with is NOT what we
think of as obvious. Simply put, the "data" given fits equally well
with the hypothesis that the mapping is just the lower 3 (or 4) order
bits of the binary representation of the first number. It would seem
to me that a computer would be more likely to hit upon this mapping,
unless it were endowed with a lot of "common sense" and (human)
cultural information.
--- Reid Simmons ---
------------------------------
Date: 17 Jul 84 1157 PDT
From: John McCarthy <JMC@SU-AI.ARPA>
Subject: re: AI-speak ?? (from SAIL's BBOARD)
[Forwarded from the Stanford bboard by Laws@SRI-AI. This reply
is in response to a request from David Cheriton@Navajo.]
jmc - As one of the importers of the first two terms from philosophy into
AI, I will say what I mean.
ontology - The dictionaries define it as the branch of philosophy that
studies what exists. In the bad old days, they argued about whether
physical objects, disembodied spirits, God, exist. Quine (1940s or 1950s?)
modernized the idea by saying that the ontology of a theory is the set
over which the bound variables range. As a nominalist he favored an
impoverished ontology, e.g. just because you want to predicate red(x),
doesn't mean that you need red or redness as an object. The AI
usage is derived from Quine's and remains quite close to it. The programs
or logical sentences have variables, and the ontology of the program
includes the sets from which these variables take values. For example,
Mycin includes bacteria in its ontology, because some of its variables
range over bacteria (the kinds of bacteria, not individual bacteriums),
but doesn't have doctors. It actually doesn't have patients either.
epistemology - In philosophy it means the study of knowledge, its sources
and limits. Again AI usage is derived from that and remains fairly
close. AI is more concerned than most philosophers with how the
knowledge is represented. AI is concerned with "epistemologically
adequate" internal languages for programs, i.e. languages that are
adequate for representing the knowledge that can actually be obtained
with given opportunities to observe and experiment. See McCarthy and
Hayes "Some philosophical problems from the standpoint of artificial
intelligence", Machine Intelligence 4, 1969.
teleology - I haven't used it in AI, so I can't speak precisely about
AI usage. In philosophy it means explaining things by ascribing
purpose to them. Extreme examples are, "The purpose of the rainbow
is to teach us that the next time God destroys the world it will be
by fire and not by water" and "The purpose of the ant is to teach us
not to be lazy". Teleological explanations were driven out of
biology accompanied by considerable squabbling. In AI the term
might be used to refer to goal-driven programs, but then it would
seem that the usage is further from the philosophical usage.
------------------------------
Date: Tue 17 Jul 84 11:45:17-PDT
From: Bruce Buchanan <BUCHANAN@SUMEX-AIM.ARPA>
Subject: AI jargon
[Forwarded from the Stanford bboard by Laws@SRI-AI.]
David,
[...] Let me try to give you a straight
answer on terminology. I spent five years in hard-core philosophy
and never felt that the terms were well-defined there, so it is no
wonder that AIers who have adopted the terms from philosophy also
have no consistent definitions. Dictionary defns are probably not
very illuminating on these things, so I haven't looked at what you
might have found there.
ONTOLOGY -- a conceptual map, a systematic description of the
objects in the world, a study of "what is"
[or the discipline of creating an ontology]
EPISTEMOLOGY -- a study of what we know & how we know it, usually
broken into a priori and a posteriori (or empirical) knowledge
TELEOLOGY -- a study of purposeful behavior (often, though not always,
defined wrt God's purpose).
This is oversimplified, of course. I would recommend Plato's
Timaeus and Theaetetus on the first two, and Aristotle's Metaphysics
on the last. By the time St.Thomas began writing about these things,
their definitions are not so clear as in Plato & Aristotle.
Epistemology is the most relevant to AI in its emphasis on
knowledge -- what it is, where it comes from, etc.
[...]
bgb
------------------------------
Date: Wed 18 Jul 84 14:35:14-EDT
From: David Rogers <DRogers%MIT-OZ@MIT-MC.ARPA>
Subject: an interesting implicit definition of intelligence
Can you spot the fallacy in this implicit definition of intelligence?
"As the programs become more refined and the network of paths and
boxes grow more complex, it becomes increasingly difficult to predict
what a computer will decide. In one second, it can process between 10
and 100 thousand logical inferences, or syllogisms. In 1981, the Japanese
government announced that it would provide almost a half a billion dollars
in seed money over the next decade to produce machines that will be able
to draw as many as 1 billion logical inferences per second.
If that goal is achieved, a computer could make, in one second, a decision
so complex that it would take a human 30 years to unravel it, assuming that
he or she could think constantly at the superhuman speed of 1 syllogism
per second. Given 10 seconds to ponder a problem, a computer's decision
would have to be taken on faith. By human standards it would be unfathomable.
When computers can have thoughts that would take more than a human lifetime
to understand, it is tempting to consider them smarter than their makers."
From "The Lure of Artificial Intelligence", by George Johnson, in the
APF reporter, Vol 7, No. 3.
(In a box at the bottom of the page, one reads "George Johnson, a freelance
writer, is reporting on the quest to build computers smarter than humans.")
------------------------------
Date: 16 Jul 1984 1329 PDT
From: Larry Carroll <LARRY@JPL-VLSI.ARPA>
Reply-to: LARRY@JPL-VLSI.ARPA
Subject: adverts
I don't mind messages like the one (very indirectly) from IBM. I'm quite
capable of recognizing even very covert bias and would discount it auto-
matically. I'd much rather do my own filtering than have a moderator do it.
larry @ jpl-vlsi
------------------------------
Date: Tue 17 Jul 84 20:29:07-PDT
From: Ken Laws <Laws@SRI-AI.ARPA>
Reply-to: AIList-Request@SRI-AI
Subject: GMR DATALOG Demo at AAAI
Kurt Godden informs me that General Motors Res. Labs. will have a live
demo of their DATALOG natural language query system at their AAAI
Villa Capri hospitality suite August 7-8. Vistors can also discuss
GMR projects in expert systems, natural language, computer vision,
robotics, etc.
-- Ken Laws
------------------------------
Date: 16 Jul 84 18:35-PDT
From: mclure @ Sri-Unix.arpa
Subject: Delphi Experiment: move 2 please?
The Vote Tally
--------------
Folks, the moves are in and have been tallied. The winner is: 1. ... c5.
A total of 21 votes were cast. Originally there was a tie between 1.
... c5 and 1. ... e5 with 7 votes each. I cast the deciding vote in
favor of the Sicilian because machines play optimally in Classical
double-KP positions and we want to avoid positions the machine likes.
The Machine Moves
-----------------
The Prestige 8-ply replied 2. Nf3 from book in 0 seconds.
Humans Move # Votes
BR BN BB BQ BK BB BN BR 1 ... c5 8
BP BP ** BP BP BP BP BP 1 ... e5 7
-- ** -- ** -- ** -- ** 1 ... e6 2
** -- BP -- ** -- ** -- 1 ... d5 1
-- ** -- ** WP ** -- ** 1 ... d6 1
** -- ** -- ** WN ** -- 1 ... f5 1
WP WP WP WP -- WP WP WP 1 ... Nc6 1
WR WN WB WQ WK WB ** WR
Prestige 8-ply
The Game So Far
---------------
1. e4 c5
2. Nf3
Your move, please?
Replies to Arpanet: mclure@sri-unix or Usenet: sri-unix!mclure.
DO NOT SEND REPLIES TO THE ENTIRE LIST! Just send them to one of
the above addresses.
Addendum
--------
For readers who don't understand all of this, I am conducting a Delphi
experiment wherein a large network-based readership can send moves in
for a chess game. Each reader's move is a vote that is combined with
other readers' votes. The move with the most votes is played against
the Prestige chess machine searching a minimum of 8 full ply deep. At
this level it is probably playing around the ELO 2200 level. The results
will eventually be published in a journal along with an analysis of the
experiment.
[In view of the limited number of respondents, I shall have to discontinue
publishing the play-by-play in this digest. Please contact mclure@sri-unix
if you wish to follow the game. -- KIL]
------------------------------
End of AIList Digest
********************
