(biassed) summary of the argument so far.
phayes@cs.uiuc.edu
Message-id: <199306010725.AA19857@dante.cs.uiuc.edu>
Date: Tue, 1 Jun 1993 02:28:49 +0000
To: interlingua@ISI.EDU
From: phayes@cs.uiuc.edu
X-Sender: phayes@dante.cs.uiuc.edu
Subject: (biassed) summary of the argument so far.
Cc: sowa@turing.pacss.binghamton.edu, cmenzel@kbssun1.tamu.edu, cg@cs.umn.edu,
phayes@cs.uiuc.edu
Let me try to summarise the core of the argument. Model theory talks of
possible interpretations of formal languages and describes them using the
language of set theory. It seems to Pat that this allows us to interpret
our languages as referring to physical things like trees and cats, since
set theory is agnostic about the nature of the individuals in the sets
which it describes. John disagrees, claiming that set theory must refer
to abstract entities, which are the subjectmatter of pure, as opposed to
applied, mathematics. John believes that model theory, or indeed any part
of 'pure' mathematics, can relate formal languages only to such an abstract
domain, which he identifies with the domains of computational or mental
constructions (an identification which Pat disagrees with, accusing John
of being a constructivist; but this disagreement constitutes a minor theme
in the argument). John therefore(?) insists that any account of how this
entire mathematical/computational system can relate to the actual, physical,
world, must go beyond model theory (or indeed any part of pure mathematics)
and concern itself with 'grounding', becoming involved with machine perception,
or at any rate epistemology and metaphysics. Pat disagrees,
claiming that it is useful to be able to simply consider interpretations of
formal languages over domains, including physical domains, which do not need
to be any more tightly specified than is required by the branch of mathematics
being used: in this case, set theory. To do so does not, Pat suggests, involve
solving any epistemological questions and need not assume any progress in
understanding how formalisms are grounded: that is concerned with how their
accuracy is ensured, while semantics is concerned only with what they could
possibly mean. He rejects John's claim that 'pure' mathematics can refer only
to abstractions, finding the idea of a set of trees, say, (or even a set of
pairs
of trees and integers) quite congenial and not muddled or confusing. John
interprets the claim that model theory can discuss relationships between
symbols
and the real world, especially the real physical world, as amounting to a claim
that the grounding problem has been magically solved or is trivial, and rejects
it on these grounds: he observes repeatedly that model theory has not
solved this
deep and complex problem. Pat repeatedly concurs, but does not find this to
be a
criticism of model theory so much as simply an observation about what is its
business. John is unable to countenance the combination of such talk with the
claim that model theory can describe relations between symbols and the actual
world.
As an example of the mutual misundertanding, Pat has repeatedly referred
to conventional 'tarskian' models, meaning the same by this as John does,
but since he
(Pat) believes the universes of such things, being sets, may legitimately
contain
physical entities; and since John finds such talk incoherent, he (John)
takes Pat
to be referring to some mythical construction which has addressed all the
philosophical problems of grounding, believing Pat to be guilty of intellectual
dishonesty in making such claims (since he (Pat) couldnt, in Johns eyes, have
been referring to ordinary models, which are made of sets, not real stuff....)
On more pragmatic issues, Pat wishes to make many distinctions that John
wishes to deny. Since John believes that model theory can describe only
abstract entities, which he identifies with computational ones, he sees such
models as being computational structures, indeed as a form of representation.
John therefore identifies 'model' in the sense of model theory (model-1) with
'model' in the sense of vivid knowledge base or database, ie a computationally
compact collection of ground facts (model-2). Since in a robot this would be
the robot's 'model' (in a third, informal sense: model-3) of its world,
with its
accuracy maintained by perceptual mechanisms, it stands between the robot's
'formal' knowledge representations and the actual world, acting as a surrogate
world to the former (and therefore a model-1 with a special status) and a
representation (model-3) of the latter, but one whose semantics are given not
using modeltheory but in some other way. John has hinted that this
might be image-like, so that this 'model' represents by being in some sense
similar to the world it represents, ie is an 'analogical' representation
(model-4). Since this is also what the sensors keep uptodate, it can be thought
of as a communication center for the robot, so it is also part of an
architectural
proposal rather reminiscent of the 'blackboard' idea.
Pats position on all this is that is seems a conflation of many (at least 5)
different ideas, and he regrets the resulting confusion, suggesting that it
is in any case unnecessary. More specifically, Pat urges that representations
should be distinguished semantically, ie by what they say, and that model
theory provides a useful and well-thought-through framework for analysing this.
As an example, he points out that a DB is better thought of as a collection
of ground assertions than as a model-1, and these have sharply different
functional roles in inference. John claims that a DB can be regarded as a
model-1 or as a model-2 because these are 'isomorphic'. It seems that John
is more impressed than Pat with the syntactic form of a representation.
Pat also is worried about the methodological implications of insisting that
model
theory can only refer to 'abstract' surrogates of reality, since it seems
to imply
that any representational project must be accompanied by a 'Principia'-like
enterprise for the relevant domain, and this would be highly impractical as
well as completely unneccessary. John has not addressed this concern.
John believes that Pat cannot be misunderstanding him, and is therefore overly
concerned with terminological issues. He suggests that we use 'depiction' as a
neutral term to refer to these things. Pat finds this useless and
water-muddying,
since his whole problem is one of getting distinctions made carefully and not
conflated with one another. Each is trying to be helpful, but finding the
actions
of the other to be so unhelpful as to seem deliberately obstructive.
----
There are many other differences between Pat and John in ways of thinking,
rhetorical style and opinions on various matters, but they are very similar
in that both are determined to argue and each regularly infuriates the other
to the point of apoplexy.
Pat
----------------------------------------------------------------------------
Beckman Institute (217)244 1616 office
405 North Mathews Avenue (217)328 3947 or (415)855 9043 home
Urbana, IL. 61801 (217)244 8371 fax
hayes@cs.stanford.edu or Phayes@cs.uiuc.edu