COMMENT ⊗   VALID 00003 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Metaphilosophy
C00007 00003		It looks like we might get a meta-philosophical
C00009 ENDMK
C⊗;
Metaphilosophy

	We propose to study ⊗metaphilosophy whose
relation to philosophy is similar to that connecting metamathematics
and mathematics.  We study the behavior of "persons" in "worlds"
where both "persons" and "worlds" are taken as abstract mathematical
objects.  For example, in ⊗metaepistemology the person is a "knowledge
seeking system" seeking knowledge in accordance with an epistemological
strategy.  Such a system represents information in
and experiments in its environment in accordance with the strategy.
If the world and the knowledge seeker are represented as abstract
mathematical objects, then the result of the interaction is a
technical mathematical question.  For example, investigating strategies
in which the system treats the world as a construct from
its inputs may show that they cannot ever "discover" certain facts
about certain kinds of worlds; at least this is a result that I would expect
on the basis of my philosophical opinions and my work on artificial
intelligence.

	While such a result might be a technical mathematical theorem
philosophical controversy will recur as soon as anyone attempts to
apply the results because he will have to argue that the real world
is one to which the theorems apply.  Nevertheless, we can hope that
the theorems we can prove will have a substantial effect on
philosophy just as the theorems of metamathematics have affected
the philosophy of mathematics.

	I don't want to be misunderstood as advocating a crude AI
approach.  However, shall begin by defining several such crude approaches
and explain why we need something more subtle and abstract.  Discussing
them will also have some value in itself.

	1. Imagine the external world as a finite automaton and
the knower as a computer program that experiments with the automaton
giving it inputs and outputs.  The simplest epistemology supposes
that the automaton is characterized by a state diagram and attempts
to discover the diagram experimentally.  E. F. Moore's (1956)
"Gedanken Experiments with Sequential Machines" discusses what
such a program might discover as it depends on the automaton being
investigated.  Here is a summary of his results.

	The information obtained may depend on actions made
before information is available to determine what actions will
get the most information.  Some information may not be accessible
at all.  The most tractable automata are those such that every
state remains accessible no matter what preliminary actions are
taken.  One can never be sure one has determined the structure
of an automaton, because there may always be a hidden counter
that will trigger a change of behavior.  However, if one makes
assumptions about the total number of states, the structure can
often be determined.

	Finite automata presented by their state diagrams are
not a plausible or tractable metaphysical model.  Presentations
as systems of interacting subautomata are somewhat better and
provide models for interesting phenomena like causality and
ability.

2.
	It looks like we might get a meta-philosophical
result about the inadequacy of procedural semantics
or some other form of operationalism.
The intuitive idea is that when we have an idea like
%2virtuus dormitiva%1 that constitutes a borrowing of theory
and haven't yet developed any idea of how to pay it
back, we have a concept without a trace of operational
definition.  Now suppose we have several possible worlds
with different ways of making %2virtuus dormitiva%1 into
a concept connected in some way to observation.  We have
no problem with a "platonic" science, since it allows
hypotheses like %2virtuus dormitiva%1 directly, but any
kind of operationalism that requires concepts to
have procedural definitions, even of Bill Woods's
general kind loses.