Ascribing mental qualities - notes for Texas philosophy colloquy 1987 Nov 13

Abstract: Acribing mental qualities like beliefs, intentions and wants to a
machine is sometimes correct if done conservatively and is sometimes necessary
to express what is known about its state.  We propose some new definitional
tools for this: definitions relative to an approximate theory and second order
structural definitions.

1. Artificial intelligence and philosophy
	AI needs comprehensive if naive world views. (maybe several of them)

2. conservative piecemeal ascription
	complex temperature control system
	thermostat
	baby or child

3. approximate theories

4. second order definitions - definitions based on satisfying axioms
	argument from cryptography that even approximate matches
with axioms are almost always unique

(example:	AN OPTIMIST IS A xUy THAT HASN'T MUCH EXPERIENCE - DON MARQUIS.

		LE PRISONNIER EST xORT.  IL N'A RIEN DIT.)

5. cartesian counterfactuals - generalized cartesian counterfactuals
	counterfactuals relative to a theory
	the example of the skier

Extended remarks on counterfactuals relative to a theory

	It now seems to me that the remarks in the paper are insufficient.

	Consider ``If the skier had bent his knees he wouldn't have fallen''.
In our treatment, the truth value of this assertion depends on the state of
the world and also on the theory of skiing shared by the two ski instructors.
(This seems to be different from the Stalnaker-Lewis notion of counterfactuals
which would try to make its truth dependent only on the state of the world, but I
don't insist on my understanding of that notion, which is certainly incomplete.)
The facts about the slope and the parameters of the skier are to be taken
from observation.  The counterfactual can be refuted by a fact, e.g. it would
be refuted if the skier had hit a tree root.  However, the observed parameters
of the skier are fitted into the theory which regards a skier as a stick figure
with joints and which takes the movements of the joints as exogenous ``driving
functions'' of the dynamics.

	One might try to eliminate the theory as a parameter of the counterfactual
by postulating ``the best theory that fits the facts''.  However, this would
make the truth of the counterfactual depend on a large part of the common
experience and reading of the ski instructors.  Keeping the theory as a parameter
is going to work much better in practice.

	What makes the counterfactual unproblematic is that while we may regard
the world as deterministic (or probabilistic in ways independent of the skier's
choices), the approximate theory has exogenous inputs on whose alternate values
the meaning of the counterfactual depends.