Copy Link
Add to Bookmark
Report
AIList Digest Volume 2 Issue 062
AIList Digest Wednesday, 23 May 1984 Volume 2 : Issue 62
Today's Topics:
Philosophy - Identity & Essence & Reference,
Seminars - Information Management Systems & Open Systems
----------------------------------------------------------------------
Date: Mon, 21 May 84 00:27:38 pdt
From: Wayne A. Christopher on ttyd8 <faustus%ucbernie@Berkeley>
Subject: The Essence of Things
I don't think there is much of a problem with saying that two objects are
the same object if they share the same properties -- you can always
add enough properties (spatio-temporal location, for instance) to effectively
characterize everything uniquely. Doing this, of course, means that sometimes
we can accurately say when two things are in fact the same, but this
obviously isn't the way we think, and not the way we want computers to be
able to think. One problem lies in thinking that there is some sharp
cut-off line between identity and non-identity, when in fact there isn't
one. In the case of the Greek Ship example, we tend to say, "Well, sort of",
or "It depends upon the context", and we shouldn't begrudge this option
to computers when we consider their capabilities. It obviously isn't as
simple as adding up fractional measures of identity, which is obvious
from the troubles that things like image recognition have run into, but
it is something to keep in mind.
Wayne Christopher
------------------------------
Date: 21 May 1984 9:30-PDT
From: fc%USC-CSE@USC-ECL.ARPA
Subject: Re: AIList Digest V2 #59
Flame on
It seems to me that it doesn't matter whether the ship is
the same unless there is some property of sameness that is of interest
to the solution to a particular problem. Philosophy is often pursued
without end, whereas 'intelligent' problem solving usually seems to have
an end in sight. (Is mental masterbation intelligence? That is what
philosophy without a goal seems to be to me.)
Marin puts this concisely by noting that intelligence exists
within a given context. Without a context, we have only senseless data.
Within a context, data may have content, and perhaps even meaning. The
idea of context boundedness has existed for a long time. Maybe sombody
should read over the 'old' literature to find the solutions to their
'new' problems.
Fred
Flame off
------------------------------
Date: 9 May 84 10:12:00-PDT (Wed)
From: hplabs!hp-pcd!hpfcla!hpfclq!robert @ Ucb-Vax
Subject: Re: A topic for discussion, phil/ai pers
Article-I.D.: hpfclq.68500002
I don't see much difference between perception over time and perception
at all. Example: given a program understands what a chair is, you give
the program a chair it has never seen before. It can answer yes or no
whether the object is a chair. It might be wrong. Now we give the
program designed to recognize people examples of an Abraham Lincoln
at different ages (with time). We present a picture of Abraham
Lincoln that the program has never seen before and ask is this
Abe. The program might again answer incorrectly but from a global
aspect the problem is the same. Objects with time are just classes
of objects. Not that the problem is not difficult as you have said,
I just think it is all the same difficult problem.
I hope I understood your problem. Trying hard,
Robert (animal) Heckendorn
..!hplabs!hpfcla!robert
------------------------------
Date: 18 May 84 5:56:55-PDT (Fri)
From: ihnp4!mhuxl!ulysses!unc!mcnc!ecsvax!unbent @ Ucb-Vax
Subject: Greek Ships, Lincoln's axe, and identity across time
Article-I.D.: ecsvax.2516
Finally got a chance to grub through the backlog and what do I find?
Another golden oldie from intro philosophy!
Whether it's a Greek ship or Lincoln's axe that you take as an
example, the problem concerns relationships among several concepts,
specifically "part", "whole", and "identity". 'Identical', by the way, is
potentially a dangerous term, so philosophers straightaway disambiguate it.
In everyday chatter, we have one use which means, roughly, "exactly similar"
(as in: "identical twins" or "I had the identical experience last week").
We call that "qualitative identity", or simply speak of exact similarity
when we don't want to confuse our students. What it contrasts with is
"numerical identity", that is, being one and the same thing encountered at
different times or in different contexts.
Next we need to notice that whether we've got one and the same thing
at different times depends on how we specify the *kind* of "thing" we're
talking about. If I have an ornamental brass statuette, melt it down, and
cast an ashtray from the metal, then the ashtray is one and the same
*quantity of brass* as the statuette, but not one and the same *artifact*.
(Analogously, you're one and the same *person* as you were ten years ago,
but not exactly similar and not one and the same *collection of
molecules*.)
It's these two distinctions which ariel!norm was gesturing at--and
failing to sort out--in his talk about "metaphysical identity" and
"essential sameness". Call the Greek ship as we encounter it before
renovation X, the renovated ship consisting entirely of new boards Y, and
let the ship made by reassembling the boards successively removed from X be
Z. Then we can say, for example, that Z is "qualitatively identical" to X
(i.e., exactly similar) and that Z is one and the same *arrangement of
boards* as X (i.e., every board of Z, after the renovation, is "numerically
identical" to some board of X before the renovation, and the boards are
fastened together in the same way at those two times, before and after).
The interesting question is: Which *ship*, Y or Z, which we
encounter at the later time is "numerically identical to" (i.e., is one and
the same *ship* as) the ship X which we encountered at the earlier time?
The case for Y runs: changing one board of a ship does not result in a
*numerically* different ship, but only a *qualitatively* different one. So
X after one replacement is one and the same ship as X before the
replacement. By the same principle, X after two replacements is one and the
same ship as X after one replacement. But identity is transitive. So X
after n replacements is one and the same ship as X before any replacements,
for arbitrary n (bounded mathematical induction). The case for Z runs: "A
whole is nothing but the sum of its parts." Specifically, a Greek ship is
nothing but a collection of boards in a certain arrangement. Now every part
of Z is (numerically) identical to a part of X, and the arrangement of the
parts of Z (at the later time) is identical to the arrangement of those
parts of X (at the earlier time). Ergo, the ship Z is (numerically)
identical to the ship X.
The argument for Z is fallacious. The reason is that "being a part
of" is a temporally conditioned relation. A board is a part of a ship *at a
time*. Once it's been removed and replaced, it no longer *is* a part of the
ship. It only once *was* a part of the ship. So it's not true that every
part of Z *is* (numerically) identical to some part of X. What's true is
that every part of Z is a board which once *was* a part of X, i.e., is a
*former* part of X. But we have no principle which tells us that "A whole
is nothing but the sum of its *former* parts"! (For a complete treatement,
see Chapter 4 of my introductory text: THE PRACTICE OF PHILOSOPHY, 2nd
edition, Prentice-Hall, 1984.)
What does all this have to do with computers' abilities to think,
perceive, determine identity, or what have you? The following: Questions
of *numerical* identity (across time) can't be settled by appeals to
"feature sets" or any such perceptually-oriented considerations. They often
depend crucially on the *history* of the item or items involved. If, for
example, ship X had been *disassembled* in drydock A and then *reassembled*
in drydock B (to produce Z in B), and meanwhile a ship Y had been
constructed in drydock A of new boards, using ship X as a *pattern*, it
would be Z, not Y, which was (numerically) identical to X.
Whew! Sorry to be so long about this, but it's blather about
"metaphysical identity" and "essences" which gave us philosophers a bad name
in the first place, and I just couldn't let the net go on thinking that Ayn
Rand represented the best contemporary thinking on this problem (or on any
other problem, for that matter).
Yours for clearer concepts, --Jay Rosenberg
Dept. of Philosophy
...mcnc!ecsvax!unbent Univ. of North Carolina
Chapel Hill, NC 27514
------------------------------
Date: 20 May 84 18:55:44-PDT (Sun)
From: hplabs!hao!seismo!ut-sally!brad @ Ucb-Vax
Subject: identity over time
Article-I.D.: ut-sally.232
Just thought I'd throw more murk in the waters.
Considering the ship that is replaced one board at a time:
using terminology previously devised for this argument, call
the original ship X, the ship with all new boards Y and
the ship remade from the old boards Z, Robert Nozick
would claim that Y is clearly the better candidate for "X-hood"
as it is the "closest continuer." The idea here is that
we consider a thing to be the same as another thing when
1) It bears an arbitrary "close enough" relation
(a desk that has been vaporized just can't be pointed to as
the 'same desk'). and
2) It is, compared to all other candidates for the
title of 'the same as X', the one which represents the most
continuous existence of X.
To be a little less hand wavy: If one considers Z rather
than Y to be the same as X then there is a gap of time in which
X ceased to exist as a ship, and only existed as a heap of lumber
or as a partially built ship. Whereas if Y is considered to be the
same as X there is no such gap.
Disclaimers: 1) The idea of "closest continuer" is Nozick's, the
(probably erroneous) presentation is my own.
2) I consider the whole notion to be somewhere be-
tween Rand and Rosenberg; i.e. it's not the best comment I've seen
on the subject, but it is another point-of-view.
Brad Blumenthal {No reasonable request refused}
{ihnp4,ctvax,seismo}!brad@ut-sally
------------------------------
Date: 17 May 84 12:50:35-PDT (Thu)
From: decvax!cca!rmc @ Ucb-Vax
Subject: Re: Essence
Article-I.D.: cca.528
What we are discussing is one of the central problems of the
philosophy of language, namely, the problem of reference. How do humans
know what a given name or description refers to?
Pre WWI logic was particularly interested in this question, as they
were building formal systems and tried to determine what constants and
variables really meant. The two major conflicting theories came from
Bertrand Russel and Gottlieb Frege.
Russell believed in a dichotomy between the logical and gramatical
forms of a sentence. Thus a proper name was not really a name, but just
a description that enabled a person to pick out the particular object to
which it refered. You could reduce any proper name to a list of
properties.
Frege, on the other hand, considered that there were such things as
proper names as grammatical and logical entities. These names had a
"sense" (similar to the "essense" in some of the earlier msgs on this
topic) and a "reference" (the actual physical thing picked out by the
name). Although the sense is sometimes conveyed by giving a
description, it is not identical to the description you would give in
trying to explain the name to someone.
Now there have been many developments of both theories. Behaviorists
tend to build "complexes of qualities" theories of meaning which read a
lot like Russell's work, but there are lots of differences in
implementation and mechanism. Linguists and modal logicians tend to
build theories closer to Frege's.
I think the most important recent book on the subject is "Naming and
Necessity", by Saul Kripke (along with Willard VO Quine and Hillary
Putnam, probably the top philosophers in North America today). The
book is a transcript, not much edited except for explanatory footnotes,
of a series of lectures trying to explain how proper names might work.
The arguments against the "quality cluster" theories seem pretty
conclusive. They include the way we use counterfactuals, that is
talking about an object or a person if they were different than they
actually were (like, what would Babbage have been like if he had lived
in an age of VLSI chips? or what would Mayor Curly of Boston been
like if he hadn't been a crook?) These discussions can get pretty far
away from reality, and this indicates that the names we use allow us to
keep track of who or what we mean without getting confused by the
changes in qualities and properties. The properties and qualities are
not what provide the "sense" or "essense" of the name.
Kripke goes on to suggest that we understand names through a
"naming" and a "chain of acquaintances". For example, Napoleon was
named at his christening, and various people met him, and they talked to
people about him, and this chain of acquaintances kept going even after
he was dead. Thus there is a (probably multi-path) chain of
conversations and pointings and descriptions that leads back from your
understanding of the name "Napoleon" to the christening where he
received his name. I am not sure that this is a correct appraisal of
the mechanism for understanding names, but it certainly is the best I
have heard.
Leonard (?) Linsky has recently written a book attacking this and
similar views, and indicating that a synthesis of the Russell and Frege
theories still has problems but avoids most of the pitfalls of
acquaintances. Unfortunately I have not yet read that book.
For other works in the area, certainly read Quine's Word and Object
and the volume of collected Putnam papers on language. Also works by
Searle and Austin on speech acts are useful for thinking about the
clues, both verbal and non-verbal, that allow us to make sense of
conversations where not everything is stated explicitly.
Enjoy!
R Mark Chilenskas
chilenskas@cca-vms
decvax!cca!rmc
------------------------------
Date: Mon 21 May 84 12:12:05-EDT
From: Jan <komorowski%MIT-OZ@MIT-MC.ARPA>
Subject: Seminar - Information Management Systems [Harvard]
[Forwarded from the MIT bboard by SAWS@MIT-MC.]
Wednesday, May 23 Professor Erik Sandewall from Linkoping University, Sweden
will talk at Harvard in the colloquium series.
Theory of Information Management Systems
4:00PM
Aiken Lecture Hall, Tea in Pierce 213 at 3:30
It is often convenient and natural to view a data base as a network consisting
of nodes, arcs from nodes to nodes, and attribute values attached to nodes.
This view occurs in artificial intelligence (eg semantic networks), data base
theory (eg. entity-relationship models), and office systems (eg. for
representation of the virtual office).
Unfortunately, the network view of data bases is usually treated informally, in
contrast to the formal treatment that is available for relational data bases.
The theory of information management systems attempt to remedy this situation.
Formally, a network is viewed as a set of triples <f,x,y> where f is a function
symbol, x is a node, and y is a node or an attribute value. Two perspective
on such networks are of interests:
1) algebraic operations on networks allow the definition of cursor-related
editing operations, and of line-drawing graphics.
2) by viewing a network as an interpretation on a variety of first-order logic,
one can express constraints on the data structures that are allowed there. In
particular, both "pure Lisp" data structures and "impure" structures (involving
shared sublists and circular structures) can be characterized. Proposition can
be also used for specifying derived information as an extension of the
interpretation. This leads to a novel way of treating non-monotonic reasoning.
The seminar emphsizes mostly the second of these two approaches.
Host: Jan Komorowski
------------------------------
Date: 21 May 1984 11:10-EDT
From: DISRAEL at BBNG.ARPA
Subject: Seminar - Open Systems
[Forwarded from the MIT bboard by SASW@MIT-MC.]
This Wednesday, at 3:00 Carl Hewitt of the MIT AI LAB will be speaking
on "Open Systems". The seminar will be held in the 3rd floor large
conference room.
Open Systems: the Challenge for Intelligent Systems
Continous growth and evolution, absence of bottlenecks, arm's-length
relationships, inconsistency among knowledge bases, decentralized
decision making, and the need for negotiation among system parts are
interdependent and necessary properties of open systems. As our
computer systems evolve and grow they are more and more taking on the
characteristics of open systems. Traditional foundational assumptions
in Artificial Intelligence such as the "closed world hypothesis", the
"search space hypothesis", and the possibility of consistently
axiomatizing the knowledge involved become less and less applicable as
the evolution toward open systems continues. Thus open systems pose a
considerable challenge in the development of suitable conceptual
foundations for intelligent systems.
------------------------------
End of AIList Digest
********************
