Copy Link
Add to Bookmark
Report
AIList Digest Volume 2 Issue 037
AIList Digest Friday, 30 Mar 1984 Volume 2 : Issue 37
Today's Topics:
Expert Systems - Judicial Expert Systems & Radiography,
Linguistics - Use of 'And',
Bibliography - Fuzzy Set Papers,
Proposals - AI Teaching & Hierarchical System Research,
Seminars - Objects and Parts & Pattern Recognition & Databases
----------------------------------------------------------------------
Date: Thu, 29 Mar 84 07:49 PST
From: DSchmitz.es@Xerox.ARPA
Subject: Judicial Expert Systems?
I'd like to know if there's any work going on out there toward the
development of expert systems (or other AI-type systems) designed to
assist in making legal decisions. Such systems as I have in mind would
be used by judges, lawyers, legal theorists, perhaps even international
courts.
Please reply to DSchmitz.es@PARC-MAXC
Thank you
[I believe there has been work at Stanford and at Yale. I also remember
reading some newspaper account of a man who wishes to market an
automated jury: each side types in its legal precedents and
the computer decides which side wins. AIList carried a seminar
notice the Stanford work last year. Can anyone give more specific
information? -- KIL]
------------------------------
Date: 26 Mar 84 11:28:10-PST (Mon)
From: decvax!linus!vaxine!debl @ Ucb-Vax
Subject: Help on Radiography Discussion
I have been told that a discussion of expert systems to read radiographs
occured on the net recently. Any information or references from this
discussion would be appreciated. Thank you.
David Lees
[There was a message by Dr. Tsotsos about his group's work at U.
of Toronto on the ALVEN system for interpreting heart images.
You might also inquire on Vision-List (Kahn@UCLA-CS); it has not
discussed this topic, but you might get a discussion started.
Dr. Jack Sklansky and associates have been developing systems
to find tumors in chest radiographs; they might be considered
"expert systems" in the sense that their performance is very
good. Chris Brown, Dana Ballard, and others at the U. of Rochester
have been using hypothesize-and-test and other AI techniques in the
analysis of chest radiographs and ultrasound heart images. -- KIL]
------------------------------
Date: 24 Mar 84 2:49:00-PST (Sat)
From: decvax!cca!ima!inmet!andrew @ Ucb-Vax
Subject: Re: Use of 'and'
Article-I.D.: inmet.1150
I haven't heard of that one, but there was an article recently (in
Datamation?) about a natural language processing system which
repeatedly gave no results when asked for "all customers in Ohio
and Indiana". Of course, no customer can be in both states
at once; the question should have been phrased as ".. Ohio *or*
Indiana". When this was pointed out, the person using the
program commented something to the effect of "Don't tell *me*
how to think!"
------------------------------
Date: Tue, 27 Mar 1984 11:13:27 EST
From: David M. Axler <AXLER%upenn-1100.csnet@csnet-relay.arpa>
Subject: Fuzzy Set Papers
Some very interesting early work on the applications of fuzzy set
theory to language behavior was done at the Language Behavior Research
Laboratory out at U. Cal - Berkeley. Much of this was later available
via the Lab's series of Working Papers and Monographs. Of interest to
AI researchers concerned w/language processing and/or fuzzy sets are:
Monograph #3, "Natural Information Processing Rules: Formal Theory and
Applications to Ethnography", William H. Geoghegan, 2/73.
Working Paper #43, "Basic Objects in Natural Categories", Eleanor Rosch,
Carolyn B. Mervis, Wayne Gray, David Johnson, and Penny Boyes-Braem, 1975.
Working Paper #44, "Color Categories as Fuzzy Sets", Paul Kay and Chad
McDaniel, 1975.
My list of the available papers is severely out of date, and I strongly
suspect that there's a fair amount of later work also available. Those
interested should write to the lab, as follows:
University of California
Language Behavior Research Laboratory
2220 Piedmont Avenue
Berkeley, CA 94720
(If anyone out at Berkeley would like to fill the list in on more recent
and relevant work from the lab, great...)
--Dave Axler
------------------------------
Date: 26 Mar 84 12:12:42-PST (Mon)
From: harpo!ulysses!burl!clyde!akgua!psuvax!bobgian @ Ucb-Vax
Subject: Teaching Proposal
Article-I.D.: psuvax.919
My main interest is Artificial Intelligence, although I define that rather
broadly: to me, AI is the field which unifies all others. Philosophy,
psychology, mathematics, compilers, languages (both programming and natural),
"systems" (both of the CS and of the EE variety), data structures, machine
architecture, discrete representations, continuous representations, cognitive
science, art, history, music, business administration, political science,
etc, etc, etc, are ALL subfields of AI to me. They all represent specific
domains in which intelligent activity is studied and/or mechanized.
I'm sure many agree with me that what the American educational system
needs is a program integrating computer "literacy" with critical thinking
abilities in many other domains. I do not mean "literacy" in the "Oh yes,
I can run statistical packages" sense. I mean an approach to critical
thinking built on the foundations of the computational paradigm -- the view
that knowledge and understanding can be represented explicitly, and that one
can discover procedures for manipulating those representations in order to
solve real problems. Such a program could form the backbone of a very
stimulating university-wide undergraduate "core" program integrating not
only mathematics and the physical sciences but communications skills and
all the "liberal arts" as well.
I visualize such a program as presenting a coherent and integrated approach
to the cognitive skills most important for healthy and productive functioning
in the modern world. It would present the major principles of cognition as
seen through the organizing principles of information processing.
This is more than an approach to teaching. To me, it is also the seed of new
approaches to machine learning and cognitive modeling. It uses undergraduate
education as an experimental testbed for research in AI, psychology,
linguistics, and social systems. That "cutting edge" fervor alone should
make it very interesting to students.
Bob Giansiracusa
Computer Science Dept, Penn State U, 814-865-9507 (ofc), 814-234-4375 (home)
Arpa: bobgian%PSUVAX1.BITNET@Berkeley
UUCP: bobgian@psuvax.UUCP -or- ..!allegra!psuvax!bobgian
Bitnet: bobgian@PSUVAX1.BITNET CSnet: bobgian@penn-state.CSNET
USmail: PO Box 10164, Calder Square Branch, State College, PA 16805
------------------------------
Date: 26 Mar 84 11:38:27-PST (Mon)
From: harpo!ulysses!burl!clyde!akgua!psuvax!bobgian @ Ucb-Vax
Subject: Proposed Research Description
Article-I.D.: psuvax.918
ADAPTIVE COGNITIVE MODEL FORMATION
The goal of this work is the automatic construction of models which can
predict and characterize the behavior of dynamical systems at multiple
levels of abstraction.
Numeric models used in simulation studies can PREDICT system behavior
but cannot EXPLAIN those predictions. Traditional expert systems can
explain certain predictions, but their accuracy is usually limited to
qualitative ("symbolic") statements. This research effort attempts to
couple the explanatory power of symbolic representations with the
precision and testability of numeric models.
Additionally, the computational burden implicit in the use of numeric
simulation models rapidly becomes astronomical when accurate performance
is needed over large domains (fine sampling density).
The solution my work explores consists of developing AUTOMATICALLY
a hierarchical sequence of SYMBOLIC models which convey QUALITATIVE
information of the sort that a human analyst generates when interpreting
numeric simulations. These symbolic models portray system behavior at
multiple levels of abstraction, allowing symbolic simulation and inference
procedures to optimize the "run time" versus "accuracy" tradeoff.
I profess the philosophical bias that the study of learning and modeling
mechanisms can proceed productively in a relatively domain-independent
manner. Obviously, domain-specific knowledge will speed the solution search
process. Such constraints can be regarded as "seeds" for search in a process
whose algorithm is largely domain-independent. Anecdotal support for this
hypothesis comes from the observation that HUMANS can become expert at theory
and model formation in a wide variety of different domains.
Bob Giansiracusa
Computer Science Dept, Penn State U, 814-865-9507 (ofc), 814-234-4375 (home)
Arpa: bobgian%PSUVAX1.BITNET@Berkeley
UUCP: bobgian@psuvax.UUCP -or- ..!allegra!psuvax!bobgian
Bitnet: bobgian@PSUVAX1.BITNET CSnet: bobgian@penn-state.CSNET
USmail: PO Box 10164, Calder Square Branch, State College, PA 16805
------------------------------
Date: Wed, 28 Mar 84 10:09:15 pst
From: chertok%ucbkim@Berkeley (Paula Chertok)
Subject: UCB Cognitive Science Seminar--April 3
[Forwarded from the SRI-AI bboard by Laws@SRI-AI.]
BERKELEY COGNITIVE SCIENCE PROGRAM
Spring 1984
IDS 237B - Cognitive Science Seminar
Time: Tuesday, April 3, 1984, 11-12:30pm
Location: 240 Bechtel
OBJECTS, PARTS AND CATEGORIES
Barbara Tversky, Dept. of Psychology, Stanford
Many psychological, linguistic and anthropological measures
converge to a preferred level of reference, or BASIC LEVEL,
for common categories; for example, TABLE, in lieu of FURNI-
TURE or KITCHEN TABLE. Here we demonstrate that knowledge
of categories at that level (and only that level) of
abstraction is dominated by knowledge of parts. Basic level
categories are perceived to share parts and to differ from
one another on the basis of other features. We argue that
knowledge of part configuration underlies the convergence of
perceptual, behavioral and linguistic measures because part
configuration plays a large role in both appearance and
function. Basic level categories are especially informative
because structure is linked to function via parts at this
level.
***** Followed by a lunchbag discussion with speaker *****
*** in the IHL Library (Second Floor, Bldg. T-4) from 12:30-2 ***
------------------------------
Date: 28 Mar 1984 09:28:05-PST (Wednesday)
From: Guy M. Lohman <LOHMAN%ibm-sj.csnet@csnet-relay.arpa>
Reply-to: IBM-SJ Calendar <CALENDAR.IBM-SJ@csnet-relay.arpa>
Subject: IBM San Jose Research Laboratory calendar of Computer
Science seminars 2-6 April 84
[Forwarded from the SRI-AI bboard by Laws@SRI-AI.]
IBM San Jose Research Lab
5600 Cottle Road
San Jose, CA 95193
Thurs., April 5 Computer Science Colloquium
3:00 P.M. MINIMUM DESCRIPTION LENGTH PRINCIPLE IN MODELING
Auditorium Traditionally, statistical estimation and modeling
involve besides certain well established procedures,
such as the celebrated maximum likelihood technique,
a substantial amount of judgment. The latter is
typically needed in deciding upon the right model
complexity. In this talk we present a recently
developed principle for modeling and statistical
inference, which to a considerable extent allows
reduction of the judgment portion in estimation.
This so-called MDL-principle is based on a purely
information theoretic idea. It selects that model in
a parametric class which permits the shortest coding
of the data. The coding, of which we only need the
length in terms of, say, binary digits, must,
however, be self-containing in the sense that the
description of the parameters themselves needed in
the imagined encoding are included. For this reason,
the optimum model cannot possibly be very complex
unless the data sample is very large. A fundamental
theorem gives an asymptotically valid formula for the
shortest possible code length as well as for the
optimum model complexity in a large class of models.
For short samples no simple formula exists, but the
optimum complexity can be estimated numerically and
taken advantage of. Finally, the principle is
generalized so as to allow any measure for a model's
performance such as its ability to predict.
J. Rissanen, San Jose Research
Host: P. Mantey
Fri., April 6 Computer Science Seminars
Auditorium
KNOWLEDGE AND DATABASES (11:15)
We define a knowledge based approach to database
problems. Using a classification of application from
the enterprise to the system level we can give
examples of the variety of knowledge which can be
used. Most of the examples are drawn from work at
the KBMS Project in Stanford. The objective of the
presentation is to illustrate the power but also the
high payoff of quite straightforward artificial
intelligence applications in databases.
Implementation choices will also be evaluated.
G. Wiederhold, Stanford University
Host: J. Halpern
---------------------------------------------------------------
Visitors, please arrive 15 mins. early. IBM is located on U.S. 101
7 miles south of Interstate 280. Exit at Ford Road and follow the signs
for Cottle Road. The Research Laboratory is IBM Building 028.
For more detailed directions, please phone the Research Lab receptionist
at (408) 256-3028. For further information on individual talks,
please phone the host listed above.
IBM San Jose Research mails out both the complete research calendar
and a computer science subset calendar. Send requests for inclusion
in either mailing list to CALENDAR.IBM-SJ at RAND-RELAY.
------------------------------
End of AIList Digest
********************
