What happens when one picks up a brick? Any child knows that the brick is now held in the air, there is one fewer object on the ground, and THAT'S ALL. For over twenty years it has been astonishingly hard to make a computer draw this conclusion. This "frame problem" is symptomatic of a host of problems in how to properly represent common knowledge about everyday actions. The papers in this volume discuss some of these problems, develop approaches to solving them, or draw philosophical conclusions from them.
Note: This is a reposting of PJ Hayes's summary, which was not archived previously. PSYCOLOQUY's revised policy is now to archive book summaries to serve as the first in the thread of multiple book reviews and authors' responses. A review (by J. van Brakel) will follow in the next number of PSYCOLOQUY (psycoloquy.92.3.60.frame-problem.2.vanbrakel).
CALL FOR BOOK REVIEWS, AUTHOR SUMMARIES & NOMINATIONS OF BOOKS FOR REVIEW: PSYCOLOQUY is calling for multiple electronic book reviews. Qualified professionals from the spectrum of disciplines covered by PSYCOLOQUY (psychology, the cognitive sciences, neuroscience, behavioral and evolutionary biology, computer science, linguisitics and philosophy) are invited to submit reviews of the Ford/Hayes volume (or of other books you would like to see reviewed in PSYCOLOQUY). All submissions are subject to editorial review.
If book authors wish to submit a summary of their recent books and to send review copies to the PSYCOLOQUY editorial office, a CALL FOR BOOK REVIEWS like this one will be posted to the PSYCOLOQUY readership, the copies of the book will be distributed to selected reviewers, and the reviews will be published in PSYCOLOQUY. Authors will also have an opportunity to publish responses to the reviews.
Note that what appears below is just a brief synopsis of an edited volume; MONOGRAPHS are preferred to edited volumes for PSYCOLOQUY multiple review. The monograph's authors should submit an article-length author's Precis (~500 lines); it will be refereed and, if accepted, it will appear in the archival journal portion of PSYCOLOQUY, as the first in the series of multiple reviews.
1.0 What happens when one picks up a brick? Any child knows that the brick is now held in the air, there is one fewer object on the ground, and THAT'S ALL. For over twenty years it has been astonishingly hard to make a computer draw this last conclusion reliably and convincingly, or even to say precisely what it means. This "frame problem" is symptomatic of a host of problems in how to properly represent common knowledge about everyday actions. The papers in this collection (originally presented at a 1989 workshop) discuss some of these problems, develop approaches to solving them, or draw philosophical conclusions from them.
2.0 Some approaches to action reasoning assume that what is needed is more information for the programs to use. Others focus rather on the reasoning methods they use. In the former spirit, Haugh develops an axiomatic theory of causal relationships, and Weld argues that system dynamics are needed. Tenenberg, however, urges the use of a probabilistic approach to knowledge representation, and Brown develops a new quantified modal logic which contains the seeds of its own description.
3.0 Some of the papers develop or criticise themes that are familiar in AI. Etherington, Kraus and Perlis explain and extend McCarthy's technique of circumscriptive reasoning. Goodwin and Trudel relate the idea of "persistence" - that things should stay true unless there is a reason to falsify them - to that of time as a continuum (a surprisingly underdeveloped idea in this area). Weber criticises persistence as an unrealistically strong assumption and shows that it has some unintuitive consequences.
4.0 Erik Sandewall makes a revisionist survey of the classical AI literature and suggests that what has always been regarded as an unfortunate pun - the frame problem and Minsky's representational notation of frames - should in fact be taken seriously.
5.0 Some of the chapters discuss broader questions. Perlis relates this whole complex of representational difficulties to the fact that the agents we are trying to imitate have evolved with only very partial knowledge of their worlds. Stein generalises the frame problem, as usually described, to the point where it seems to be the general problem of counterfactual reasoning: an alarming and controversial conclusion. Fetzer, a philosopher of science, and Hayes, one of the book's editors, give us a debate on the relevance of philosophy to AI. Fetzer argues that the frame problem is really the problem of scientific induction, so a successful robot must have solved Hume's problem. Hayes argues that Fetzer misunderstands the nature of the AI enterprise and Fetzer replies with a suggestion that a new kind of programming might be the answer.
6.0 Most of the papers are self-contained and some provide excellent introductions to the frame problem and the history of attempts to solve it.
Table of Contents
Framing The Problem
Kenneth M. Ford, University of West Florida Patrick J. Hayes, Xerox PARC
The Modal Quantificational Logic Z Applied to the Frame Problem
Frank M. Brown, University of Kansas
Limited Scope and Circumscriptive Reasoning
David Etherington, AT&T Bell Laboratories Sarit Kraus, Hebrew University - Israel Donald Perlis, University of Maryland
The Frame Problem: Artificial Intelligence Meets David Hume
James H. Fetzer, University of Minnesota
Commentary on: "The Frame Problem: Artificial Intelligence Meets David Hume"
Pat Hayes, Xerox PARC
A Response to Pat Hayes
J. H. Fetzer
Persistence in Continuous First Order Temporal Logics
Scott D. Goodwin, University of Waterloo Andre Trudel, Acadia University
Omniscience Isn't Needed to Solve the Frame Problem
Brian A. Haugh, Martin Marietta Laboratories
Knowledge and the Frame Problem
Leora Morgenstern, IBM T. J. Watson Labs
Focus of Attention, Context, and the Frame Problem
J. Terry Nutter, Virginia Tech.
Intentionality and Defaults
Donald Perlis, University of Maryland
Towards a Logic of Dynamic Frames
Erik Sandewall, University of Linkoping - Sweden
An Atemporal Frame Problem
Lynn Andrea Stein, Massachusetts Institute of Technology
Abandoning the Completeness Assumptions: A Statistical Approach to the Frame Problem
Josh D. Tenenberg, University of Rochester
The Myth of Domain-Independent Persistence
Jay C. Weber, Lockheed Artificial Intelligence Center
System Dynamics and the Qualification Problem
Daniel S. Weld, University of Washington