COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	Here we suppose that all the preconditions are absorbed into
C00005 ENDMK
C⊗;
Here we suppose that all the preconditions are absorbed into
fluents that may prevent the action.  If the conditions for
the action are a disjunction it may be better not to do it
that way in order to get simpler formulas.  Thus we can
have different axioms for going by bicycle and going on foot
each of which mentions its main conditions.  Hmm!  If different
conditions may prevent going on foot or going by bicycle, then
the action go should contain the mode of travel as a parameter.
The problem of circumscribing what something depends on again
arises.

∀p z s.(¬prevented(p,go(z),s) ⊃ holds(at(p,z),result(p,go(z),s)))

∀p z x y s.(walkable x ∧ holds(at(p,y),s) ∧ att(y,x) ∧ att(z,x)
∧¬prevented(p,go(z,walking),s)
⊃ holds(at(p,z),result(p,go(z,walking),s)))

∀y x.(depends(y,x) ≡ ∃f.∀s.(y(s) = f(x(s))))
Quite a lot of imitation would be required to get the second order
logic out of this one.

The principle of rationality is normally cited as "He will do what
he thinks will achieve his goals".  Formalizing this and using it
to predict action is tedious, because the reasoning the subject
goes through has to be modelled.  (It has to be worked out that
imputed reasoning takes only <constant> more work than direct
reasoning).  Often the default is that the participants understand
the situation, and this permits simplified variants of the principle
of rationality.  Here are some:

∀p g a s.(holds(g,result(p,a,s)) ∧ wants(p,g) ⊃ NORMALLY holds(does(p,a),s)

Well there seems to be only one.

*****

The above is the simplest case of deciding what someone will do.  It
assumes there is only one goal and only one action that will achieve
it.  What is the next simplest case?  Can other cases be assimilated
to the single action case by elaborating the goal to "achieving the
goal in the best way"?