AIList-REQUEST@SRI-AI.ARPA (AIList Moderator Kenneth Laws) (09/05/85)
AIList Digest Thursday, 5 Sep 1985 Volume 3 : Issue 118
Today's Topics:
Query - CUSP/MANDALA,
AI Tools - OPS5 in Interlisp & Prolog and Lisp,
Expert Systems - FDA Approval for Expert Systems,
Humor - Re: Good News and Bad News,
Games - Chess Openings & Computers Cheat at Chess?
----------------------------------------------------------------------
Date: 4 Sep 1985 10:19-EST
From: leff%smu.csnet@csnet-relay.arpa
Subject: CUSP/MANDALA
I found this in net.ai. I also have attached a response for him/her.
Can you either get it back to him or put in AIList? (I checked the
curent UUCP map and find no record of this site.)
/* Written 10:38 am Aug 30, 1985 by jjd@oce-rd2.UUCP in smu:net.ai */
/* ---------- "CUSP/MANDALA" ---------- */
Can anybody tell me about CUSP, Mandala and/or other
End User Languages.
CUSP = Customer Programming Language
Mandala = Pictoral Language
Response:
Electronics Week, November 19, 1984
ICOT Details Its Progress. Reports on work done on prolog
machines, a new logic language called Mandala. page 20
%A Robert Haavind
%T Playing to win a New Generation
%J High Technology
%D AUG 1985
%P 63-65
%K prolog lisp machine KL1 mandala ICOT
%X describes new parallel and other novel architectures being developed
in Japan including those for AI
------------------------------
Date: Tue 3 Sep 85 08:18:36-PDT
From: Mark Richer <RICHER@SUMEX-AIM.ARPA>
Subject: OPS5 in Interlisp
Someone asked about OPS5 in Interlisp; I picked up a brochure at IJCAI
on a version that is for sale.
SAIC (Science Applications International Corporation)
First package (1 disk, object code only, and 1 user's manual) $2000
additional packages $500 each
source code $5000 (interlisp)
For more information, write:
saic, p.o. box 2341, La Jolla, CA 92038 [attention: Linda Anderson]
Mark
------------------------------
Date: Mon 26 Aug 85 00:57:43-PDT
From: Andy Freeman <ANDY@SU-SUSHI.ARPA>
Subject: Prolog and Lisp
From: Carl E. Hewitt <HEWITT@MIT-MC.ARPA>
Prolog (like APL before it) will fail as the foundation for Artificial
Intelligence because of competition with Lisp. There are commercially
viable Prolog implementations written in Lisp but not conversely.
From: PEREIRA@SRI-AI.ARPA
1. (the least interesting one) All the so-called commercially viable
Prolog systems in Lisp are not really Prolog systems written IN Lisp,
but rather Prolog systems written FOR Lisp machines. [They are
microcoded and have good support for tagged data types.] Without
those machine-level operations, those Prolog systems would run too
slow and use too much memory to be useful for serious Prolog
programming.
Hewitt's message alludes to an apparent difference between Lisp and
Prolog which Pereira's response ignores. The response also implies
that Prolog in Lisp is inherently slow. A previous message from
Pereira (Prolog Digest Volume 1, Issue 21) correctly emphasizes that
speed is important. (This issue and the ones following are quite
informative.)
From: PEREIRA@SRI-AI.ARPA
What Prolog "does" is not just to make certain deductions in a certain
order, but also MAKE THEM VERY FAST.
One of the differences between Lisp and Prolog is in how efficiently
they support embedded languages. Special purpose languages, like
special purpose hardware, are an important tool. Prolog has not
demonstrated its ability to support them (apart from Logic in Prolog,
of course), Lisp has.
One example of this is the Prolog interpreter Foolog by Martin
Nilsson* in that Prolog Digest issue. Written in 2 pages of Maclisp,
it runs about 75% as fast as interpreted DEC-10 Prolog; a later
version was supposed to be marginally faster. Naturally, Foolog
doesn't include a debugger and other system functions or all of the
user utilities in the Prolog library (but they're written in
Prolog...), but it does support cut, call, bagof, arithmetic, and I/O.
A simple compiler apparently was written later that generated code 25%
as fast as compiled DEC-10 Prolog. After that message, I lost track
of Foolog. Nilsson described it as a "toy"; I doubt much came of it.
Obviously the first 50% is the easiest and one shouldn't ignore the
environment. (Both Prolog and Lisp encourage powerful environments.
At that time, DEC-10 Prolog was fairly mature.) When Prolog is the
appropriate language, then one should use the best Prolog available.
When you need a special language however, Lisp is hard to beat.
If logic is by far the most important language for AI, Lisp may be in
trouble. If it isn't, pure Prolog environments are in trouble.
-andy
ps - The speed advantage of special Lisp hardware is shrinking. I
assume that the same is true for Prolog.
* Nilsson was one of the developers of LM-Prolog.
------------------------------
Date: Tuesday, 3 Sep 1985 11:40:22-PDT
From: billingslea%lite.DEC@decwrl.ARPA (Mark Billingslea `Lock and Load!')
Subject: FDA Approval for Expert Systems
This mail was sent to me, from Germany. Thought I would pass it on to you
for review.
-Mark
From: GYPSC2::ROLLER "Christian E. Roller - PSC Muenchen - RTO"
1-SEP-1985 02:51
To: COORS::BILLINGSLEA
Subj: Some news for AI digest
<><><><><><><><> T h e V O G O N N e w s S e r v i c e <><><><><><><><>
Edition : 897 Friday 30-Aug-1985 Circulation : 365
Send subscription requests, backissue requests and letters to MAGIC::VNS
[TAYLOR ]
[Nashua, NH, USA]
Some AI systems may need FDA approval
Expert systems come within the FDA ambit to the extent that
they supplement doctor's work, according to Richard Beutal, a
Washington D.C. attorney specializing in the legal aspects of
technology.
An expert system may be defined as a computer program that
embodies the expertise of one or more human experts in some
domain and applies this knowledge to provide inferences and
guidance to a user. some of the earliest and most
sophisticated systems were developed for medical diagnosis:
MCYIN, EMCYIN, CADUCEUS AND ATTENDING. [There are several
more in use in Japan. --mjt]
Beutal called attention to proposed FDA regulations that, if
implemented, would require medical expert systems to obtain
FDA pre-marketing approval. Given that FDA approval for what
are class 3 devices could take up to 10 years and that
reclassifying such devices can take almost as long, these FDA
regulations would virtually cause investment to dry up.
{Government Computer News Aug 16, 1985}
[...]
------------------------------
Date: Mon, 26 Aug 85 10:50:13 EDT
From: Cymru am Byth! <bradford@AMSAA.ARPA>
Subject: Re: Good news and bad news, Mr. Wizard...
From: straz@AQUINAS.THINK.COM@MIT-CCC, Steve Strassmann
But the study also showed that TV scientists are killed more often than
soldiers, private eyes, and police officers."
WOW! Just how often is a TV scientist likely to be killed?
Not more than once, I hope!
PJB
"So you think being drunk feels good -- tell that to a glass of water!"
------------------------------
Date: Mon, 26 Aug 85 8:15:37 EDT
From: Bruce Nevin <bnevin@bbncch.ARPA>
Subject: chess openings
It is not clear whether a human expert studying the Caro-Kann defense for five
weeks memorizes 50,000 openings or generalizes in ways that are difficult
in conscious, verbal terms to articulate. Indeed, isn't there evidence that
even human memorizing entails kinds of generalization that are not present in
simple storage of a table for later lookup?
Seems to me this points to the kernel issue of machine learning,
and a hard nut at that!
[For recent progress in chess subpattern learning see Albrecht Heeffer's
"Validating Concepts from Automated Acquisition Systems", IJCAI-85, pp.
613-615. -- KIL]
------------------------------
Date: Wed, 28 Aug 85 09:33:14 pdt
From: Tom Dietterich <tgd%oregon-state.csnet@csnet-relay.arpa>
Subject: Computers Cheat at Chess?
I enjoyed Stuart Cracraft's notes concerning the use of tree searching in
chess programs. Amidst all of the hype about "shallow" versus "deep" expert
systems, it is interesting to note that most computer chess systems should
be classified as "deep". Once they get past the opening moves, they play
every game from "first principles"!
--Tom
------------------------------
End of AIList Digest
********************