perm filename UICC[RDG,DBL] blob
sn#543037 filedate 1980-11-05 generic text, type C, neo UTF8
COMMENT ⊗ VALID 00006 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00002 00002 Mailed to THORPE@CMUA 13:58 31-Oct
C00005 00003 Mailed (by Post Office) 3 November
C00017 00004 ∂02-Nov-80 2014 CSD.LENAT at SU-SCORE Re: Comments before I mail this off?
C00018 00005 ∂05-Nov-80 1020 CSD.SMITH at SU-SCORE Re: FYI
C00019 00006 Out-takes
C00021 ENDMK
C⊗;
Mailed to THORPE@CMUA 13:58 31-Oct
U of I @ CC
Jack Mostow suggested you might be able to answer a question I have about
the University of Illinois at Chicago Circle. Is it on any computer
network? In particular, is it on any of the networks which (you know to)
include Stanford - such as the ARPAnet, or SUMEX. (I checked out the
ARPAnet, but could not find it there - does it have perhaps some cryptic
designator?)
Thank you.
Russ Greiner
(I am trying to send a message to Prof Minkowycz and A. Sharma.)
∂31-Oct-80 1518 Charles.Thorpe at CMU-10A (C410CT60) Re: U of I @ CC
Date: 31 October 1980 1750-EST (Friday)
From: Charles.Thorpe at CMU-10A (C410CT60)
To: Russell Greiner <RDG at SU-AI>
Subject: Re: U of I @ CC
In-Reply-To: Russell Greiner's message of 31 Oct 80 16:58-EST
Message-Id: <31Oct80 175047 CT60@CMU-10A>
The only Chicago organization I could get to admit to having ARPAnet connections
was Argonne National Labs, and they're a long ways away from Circle campus. I
don't know about the other nets. Sorry I couldn't be of more help.
Chuck Thorpe
PS--greet Jack for me.
Mailed to THORPE@CMUA 15:50 31-Oct
Thanks anyway.
Do you know anything about the research Minkowycz et al are pursuing?
Russ
∂01-Nov-80 1210 Charles.Thorpe at CMU-10A (C410CT60) Re: Thanks anyway.
Date: 1 November 1980 1507-EST (Saturday)
From: Charles.Thorpe at CMU-10A (C410CT60)
To: Russell Greiner <RDG at SU-AI>
Subject: Re: Thanks anyway.
In-Reply-To: Russell Greiner's message of 31 Oct 80 18:50-EST
Message-Id: <01Nov80 150726 CT60@CMU-10A>
Sorry, I don't know anything about Minkowycz. My only dealings with the U of
Chicago were in trying to get an ARPA link to CMU; I never actually attended
there.
Chuck
Mailed to THORPE@CMUA 13:13 1-Nov
OK
Oh, Jack wasn't sure what your affiliation with UofI@CC was.
Russ
Mailed (by Post Office) 3 November
1 November 1980
Professor W.J. Minkowycz
Dr. A. Sharma
University of Illinois at Chicago Circle
College of Engineering
Box 4348
Chicago, Illinois 90980
Dear Sirs:
I am pleased to learn of your interest in the RLL-1 system, as expressed
in your letter of 10 October. I am sending, under seperate cover, the
memos you (explicitly) requested, together with two other potentially
relevant ones. The "Details of RLL-1" report includes many low-level
specifics of our current system.
While we do plan to release RLL-1 eventually, we are cautious about
releasing it too soon. Our reasons stem from various causes:
(1) The particulars of the starting RLL-1 system (i.e the "bootstrap") are
still being hammered out. The difficulties of this design are in two
camps: First we want the system to be as unbiased as possible. Given
that this starting system will necesarily employ some conventions we want
to insure each such convention is "minimal" and necessary -- that is, they
should not force you, the user, into difficult situations or awkward
coding. The neo-natal MRS (described in more detail below) is much closer
to realizing this goal that RLL-1 will ever be.
Secondly, RLL-1 should come equipped with enought powerful, general
constructs that the user can readily do a great many useful things, over a
wide range of tasks. Bifucating again, many of RLL-1 current components
have not yet reached the generality we think possible, and furthermore,
there are many areas which have not even been considered, and so the
modules capable of performing this type of task have not yet been built.
This is not surprising: By design, RLL-1 is a continuously growing and
evolving system -- one capable of adding on new components as the need
arises. The concern here is the large number of known ommissions.
(2) Another major problem is the bugs which are present in the current
RLL-1 system. We feel it will take about a month to correct the ones we
now know about.
(3) The other major issue stems from research directions. Prof Lenat and
I developed RLL-1 as a tool, to be used to build the EURISKO system. [See
Appendix B.2 for an overview of RLL-1's role as foundation for this
system. There were two major reasons why we encouraged others to use
RLL-1: First, these other applications will push on the set of features
RLL-1 will have to provide; the modules built to handle such situations
will expand RLL-1's capabilites, making it a more general tool for our
uses as well. Second, RLL-1 will provide a Lingua Franca for
EURISKO-related knowledge bases. To function, the EURISKO system must
first include a large collection of diverse knowledge bases. Rather than
inputing these ourselves, we would rather develope a symbiotic relation
with other researchers -- they will be given the underlying EURISKO
system, as a particular representational/control scheme for their data,
and in exchange, we get to peak over their shoulders, and collect their
set of heuristics. As such, it is strongly in our interest to continue
supporting EURISKO. We have further decided to support/improve RLL-1 only
where such modifications are crucial to its application to this primary
line of research -- which is concerned more with heuristics than with
representations, per se.
The research ideas of a representation language language are still being
pursued here at Stanford, by Professor Micheal Genesereth and David Smith.
They are now developing the MRS (for Modifiable Representation System)
mentioned above. They have guaranteed continued support of its kernel,
and intend to soon develop a full user community, in which implementations
of diverse representational systems can be coopertively constructed,
collected and dispersed. After the (both conceptual and implementation)
bugs have been purged from both of our respective systems, RLL-1 will
actually be encoded in MRS. At this point RLL-1 will be (viewable as) one
of many "plug-in" modules, which "twiddles" that copy of MRS into a system
which follows a particular set of conventions.
With these caveats in mind, here are the options for using RLL-1, as I see
them:
(1) We could export to you the RLL-1 system, as it is now. We would, of
course, make no commitment to support this pre-release version. It is,
however, thoroughly documented, and does do a great number of things
correctly.
(2) In about another month (after removing those major known bugs) we plan
to freeze the then-existent system. This system will be fairly static --
the only updates will be repairs to bugs subsequently found, as opposed to
extensions. Note that we will fix just those problems which seem to lie
on our critical path.
(3) The EURISKO network is just starting up -- within a few months after
getting a working version of RLL-1, we feel it will be distributable. If
your task area is conducive to a EURISKO type of exploration, you will be
welcome to use that system. Recall that this network, which will be built
on top of RLL-1, will continue to be supported from here.
(4) If you wish to pursue representation issues, qua research topic, you
may wish to join the growing RLL-1/MRS community. "Good citizens" of this
group are expected to contribute both ideas and code to an evolving,
improving system.
Let me know which of these options seems most appropriate -- or perhaps
you can envision another possibility.
I have a quick list of remaining issues/questions:
1. Your letter refered to a description of your task. Could you perhaps
send me another copy of this article, as I never found it.
2. Is UofI@CC on any computer network which also includes Stanford? Do
you know of any (even indirect) electronic path from there to here? I can
be reached at RDG@SU-AI, CSD.GREINER@SCORE or GREINER@SUMEX; and much
prefer this faster communication to the slower postal mail.
3. I am enclosing a copy of the first (and only) message posted on the
(Virtual) RLL Bulletin Board. If you wish, I will see that you receive
further messages as well. (Note this bulletin board is intended to
discuss theoretical issues of representation language langauges, as
opposed to our implementation, RLL-1, per se. Notications of
updates/modifications/bugs/etc to RLL-1 will appear elsewhere.)
4. RLL-1, as it now stands, does require InterLisp. MRS, however, will
have parallel implementations in both InterLisp and MacLisp. Thus once
RLL-1 has been written in MRS it will be MacLisp compatible.
This has been the first formal request for RLL-1, from outside of
Stanford. As such the policies concerning how to distribute this system
are quite flexible. If you have any suggestions on how we should
facilitate releasing our system, please let me know. Please let me know,
also, if there is any way I may be of further help with your project.
Sincerely,
Russell Greiner
Computer Science Department
Stanford University
Stanford, CA 94305
CC: Professor Douglas B. Lenat, Professor Micheal R Genesereth, David E
Smith
∂02-Nov-80 2014 CSD.LENAT at SU-SCORE Re: Comments before I mail this off?
Date: 2 Nov 1980 2011-PST
From: CSD.LENAT at SU-SCORE
Subject: Re: Comments before I mail this off?
To: RDG at SU-AI
In-Reply-To: Your message of 1-Nov-80 1437-PST
An ecellent answer; much bette than I'm sure they're expecting. Thanks.
Doug
-------
∂05-Nov-80 1020 CSD.SMITH at SU-SCORE Re: FYI
Date: 5 Nov 1980 1017-PST
From: CSD.SMITH at SU-SCORE
Subject: Re: FYI
To: RDG at SU-AI
In-Reply-To: Your message of 4-Nov-80 1806-PST
You should probably ask Ed or Bruce about distribution. I think that Stanford
requires that a copyright notice go out with the software, and may also require
that a liability release form be signed.
-------
Out-takes
First, many design related issues are just now being decided --
many of these may have considerable ramifications on the ...
In addition to these areas which we know ... are areas which have not yet
been included - simple omissions.
In addition, there are still a non-trivial number of simple bugs, not yet
tracked down.
The current RLL-1 system (and MRS as well) is basically a general (hopefully
universal) nucleus, and a collection of particulars.
Defn of system - i.e. RLL more like an idea; implementation is just
certain bootstrapping...
RLL, in theory, is a composition of a large number of modules,
each of which encodes some body of knowledge.
For many applications may be essential to have things like EURISKO, a KB
which includes some idea of a control system (here Agenda mechanism)
I'm not familiar with hardware, and other implementation sorts of things.