Copy Link
Add to Bookmark
Report
AIList Digest Volume 5 Issue 097
AIList Digest Tuesday, 14 Apr 1987 Volume 5 : Issue 97
Today's Topics:
Administrivia - Recent Delivery Problems,
Policy - Source Code Postings,
Seminars - The Anatomy of AI Tarpits (CMU) &
Analogical Transformation Extension (CMU) &
Leaning on the World (CMU) &
Parallelism for KR Languages (SRI) &
Understanding and Personality (SUNY Buffalo) &
An Integrated Framework for Factory Scheduling (CMU)
----------------------------------------------------------------------
Date: Mon 13 Apr 87 10:28:38-PDT
From: Ken Laws <Laws@STRIPE.SRI.COM>
Subject: Recent Delivery Problems
AIList distribution has been delayed for about a week due to 1) mailer
problems caused by the recent change in Arpanet host names, 2) the time
it has taken me to put in a garden, and 3) the birth of my third child,
Devon Lee Laws. The mailer and garden problems seem to be fixed now,
but please excuse my continued poor response time -- I have to spend a
greater percentage of my evenings and weekends looking after my family now.
-- Ken
------------------------------
From: "Norbert E. Fuchs" <fuchs%ifi.unizh.chunet@RELAY.CS.NET>
Subject: Source Code Postings
Using FTP for source code downloading may be fine for those on the Arpanet, but
there are others like myself outside the USA who have no access to an Arpanet
site. I support posting source code in AIList - or an equivalent solution - so
that everybody has the possibility of downloading.
--- nef
------------------------------
Date: 1 Apr 87 15:01:35 EST
From: Patricia.Mackiewicz@isl1.ri.cmu.edu
Subject: Seminar - The Anatomy of AI Tarpits (CMU)
SPECIAL AI SEMINAR
TOPIC: "The Anatomy of AI Tarpits"
SPEAKER: Phil Agre, MIT
WHEN: Monday, April 6, 1987, 1:00 pm
WHERE: Doherty Hall 3313
ABSTRACT
This is a talk I gave at a recent workshop on Meta-Level Architectures.
I originally wrote it to blow off steam at the last dozen AI papers I
had read, but I have come to think that it provides a clean, simple
explanation of why AI (among other fields) is wedged.
My thesis is that AI, as a field, has a pathological attitude toward
language. AI repeatedly gets itself into tarpits based on
pseudo-technical words like, for example, "planning." The community
(or, lately, some sub-field of it) cultivates a habit of seeing "planning"
in an activity's slightest intentionality, regularity, deliberateness,
or planfulness -- and marginalizing or ignoring anything else. Then it
writes (destruct plan ...). I will describe an anatomy of tarpits that
lets us predict in remarkable detail how their victims will get stuck.
Then I will discuss five examples: the mind, planning, knowledge,
variables, and the meta level. Along the way I will suggest some ways out.
**************************************************************************
If you are interested in an appointment with Phil Agre please contact
Patty at extension 8818 or pah@d.
**************************************************************************
------------------------------
Date: 1 Apr 87 19:51:18 EST
From: Steven.Minton@cad.cs.cmu.edu
Subject: Seminar - Analogical Transformation Extension (CMU)
Wei-Min Shen is giving this week's seminar. As usual, we will meet
in 7220 at 3:15 on Friday. Here's the abstract:
Analogical Transformation Extention
and its Applications
One of the aspects of learning by analogy is concerned with constructing and
generalizing a transformation in the source domain and productively using it
in the target domain. In this talk, we will discuss a preliminary approach,
ATE, to the problem and its applications to: (1) creating new operators
(more general than Macro-Operators) in AI discovery systems; and (2) solving
problems in Geometric-Analogy Intelligence-Tests.
For the first application, we will discuss in detail an implemented system,
ARE. It starts with a small set of creative operations and a small set of
heuristics, and uses ATE to create all the concepts attained by Lenat's AM
system, and others as well. Besides showing a way to meet the criticisms of
lack of parsimony that have been leveled against AM, the ARE system provides
a route to discovery systems that are capable of "refreshing" themselves
indefinitely by continually creating new operators.
For the second application, we will compare the ATE approach with the method
used by Evans in his program for solving problems in Geometric-Analogy
Intelligence-Tests, and show that the ATE approach can solve the problems
more efficiently.
This discussion is a report on an ongoing project. We will appreciate any
suggestions and comments. In case I cannot answer your hard questions, I
will bring some delicious chinese rice pudding as my defence.
------------------------------
Date: 3 Apr 87 08:36:22 EST
From: Patricia.Mackiewicz@isl1.ri.cmu.edu
Subject: Seminar - Leaning on the World (CMU)
TOPIC: "Leaning on the World"
SPEAKER: Phil Agre, MIT
WHEN: Tuesday, April 7, 1987, 3:30pm
WHERE: Wean Hall 5409
David Chapman and I have been studying the organization of everyday
routine activity (things like making breakfast and driving to work)
with an eye to understanding the human cognitive architecture.
In trying to explain what we've observed, we've been lead away from
mentalistic metaphors emphasizing containment and boundary (perception,
behavior, programs and processes, content-bearing datastructure-like
representations) and toward metaphors emphasizing agents' interactions
with their worlds.
Our central distinction is between an agent's "machinery" and the
"dynamics" of its activity. We have found that, for the broad range of
routine activity we have studied, a very simple architecture suffices.
It consists of an innate "periphery" (along the lines of Marr and
Ullman) and a constructed "center". Careful analysis of the reliable
patterns of interaction in the agent's world allows the center to be
made out of very simple hardware, in fact combinational logic.
This simplicity derives largely from a new theory of representation.
Where traditional representation schemes posit objectively defined
"individuals" in the world, our scheme of "indexical-functional
aspects" (or "aspects" for short) parses the nearby materials
according to their relationship to the agent's person (i.e.,
indexically) and purposes (i.e., functionally). Such a scheme
generalizes its understanding without putting variables in for
constants, so it does not need any hardware for matching, binding, and
substitution.
Chapman is almost done implementing an instance of this architecture.
Pengi is a program that plays the video game Pengo. Pengi's periphery
simulates a person looking at a video game monitor. Its center is a
fixed combinational network derived from a specification of the salient
aspects of the recurring game situations. With luck, a demo will be
available.
Strongly suggested reading (copies may be available):
Chapman and Agre, Penti: An Implementation of a Theory of Situated
Activity, submitted to AAAI-87.
Chapman and Agre, Abstract Reasoning as Emergent from Concrete
Activity, Workshop on Reasoning About Action, 1986.
Shimon Ullman, Visual Routines, MIT AI Lab Memo 723, June 1983.
**************************************************************************
If you are interested in an appointment with Phil Agre please contact
Patty at extension 8818 or pah@d.
**************************************************************************
------------------------------
Date: Fri, 10 Apr 87 12:23:48 PDT
From: Amy Lansky <lansky@venice.ai.sri.com>
Subject: Seminar - Parallelism for KR Languages (SRI)
PARALLELISM IN INTERPRETERS FOR KNOWLEDGE REPRESENTATION LANGUAGES
Henry Lieberman (HENRY@OZ.AI.MIT.EDU)
MIT
11:00 AM, MONDAY, April 13
SRI International, Building E, Room EJ228
While there has been considerable interest in applying parallelism to
problems of search in knowledge representation languages, lingering
assumptions of sequentiality in the interpreters for such languages still
stand in the way of making effective use of parallelism. Most knowledge
representation languages have a sequential QUERY-SEARCH-ANSWER loop, the
analog of the READ-EVAL-PRINT loop of Lisp, and employ parallelism only in
the SEARCH phase, if at all. I will discuss parallel alternatives to
sequential interpreters for knowledge representation languages, and new
approaches to constructing user interfaces for these languages. These
observations arise out of experience with the representation language Omega
of Attardi, Simi, and Hewitt. The approach is motivated by a desire to
respond to Hewitt's "open systems" critique of logic-based systems, which
strives for systems that can deal with inconsistent beliefs, dynamically
revise beliefs, and are sensitive to allocation of resources.
VISITORS: Please arrive 5 minutes early so that you can be escorted up
from the E-building receptionist's desk. Thanks!
------------------------------
Date: 10 Apr 87 22:24:44 GMT
From: rocksvax!rocksanne!sunybcs!rapaport@CS.ROCHESTER.EDU (William
J. Rapaport)
Subject: Seminar - Understanding and Personality (SUNY Buffalo)
philosophy of science
STATE UNIVERSITY OF NEW YORK AT BUFFALO
GRADUATE GROUP IN COGNITIVE SCIENCE
JOHN HAUGELAND
Department of Philosophy
University of Pittsburgh
UNDERSTANDING AND PERSONALITY
Artificial Intelligence (AI) has inherited a conception of pure under-
standing from modern philosophy, especially Descartes and Kant. How-
ever, developments within AI, specifically with regard to knowledge
representation, have partially undermined this conception. It will be
argued that they have not gone far enough in this. In particular,
``impurities'' like ego and affects must be included as well.
Thursday, April 23, 1987
4:00 P.M.
Knox 4, Amherst Campus
Co-sponsored by:
Department of Computer Science
and
Colloquium in the History and Philosophy of Science
Informal discussion at 8:00 P.M. at Stuart Shapiro's house, 112 Park-
ledge Drive, Snyder, NY. Call Bill Rapaport (Dept. of Computer Science,
636-3181), Gail Bruder (Dept. of Psychology, 636-3676), or Zeno Swijtink
(Dept. of Philosophy, 636-2444) for further information.
------------------------------
Date: 10 Apr 87 13:05:44 EDT
From: Patricia.Mackiewicz@isl1.ri.cmu.edu
Subject: Seminar - An Integrated Framework for Factory Scheduling
(CMU)
AI SEMINAR
TOPIC: Toward An Integrated Framework For Factory Scheduling
SPEAKER: Steve Smith, CMU
WHEN: Tuesday, April 14, 1987, 3:30 p.m.
WHERE: Wean Hall 5409
ABSTRACT:
In this talk we present work aimed at providing an integrated framework for
coordinating factory production. An integrated framework is defined as one
that merges predictive generation/expansion of the production schedule with
reactive schedule management in response to the dynamics of factory
operation. We describe OPIS, a knowledge-based scheduling system that
advocates a common view of predictive and reactive scheduling as an
opportunistic problem solving process. This view is realized by a system
architecture that combines constraint propagation and consistency
maintenance techniques with heuristics for dynamically focusing the
scheduler according to characteristics of current solution constraints. A
collection of scheduling methods, varying in the decomposition of the
problem that is assumed and the types of constraints and objectives that are
emphasized, are defined to provide strategic alternatives. We present
experimental evidence of the effectiveness of this approach in generating
schedules and give examples of its use in reactively revising them as the
situation warrants. We then turn attention to the central assumption of an
incrementally maintained schedule as the basis for factory floor
decision-making and consider its computational implications. Current work
directed toward improving the robustness of predictive schedules and
hierarchically distributing the scheduling effort is described.
------------------------------
End of AIList Digest
********************