Information Technology Reference
In-Depth Information
is given by global minimization. Abnormalities caused by the performance of
actions need to be preferably accepted or, for that matter, not considered
abnormal at all. The action of starting the car engine being disqualied, for
instance, is a perfectly reasonable, since natural, consequence of introducing
a potato into the tail pipe.
With the last remark we seem to have pinned down the problem to the
question of how to account for the fact that abnormalities may be obtained
as a side eect of other actions. That is to say, the occurrence of an action
disqualication is not abnormal if being evoked as indirect eect of another
action. Put that way, the whole apparatus of Chapter 2 might furnish a
ready fundamental for a solution to the problem of caused abnormalities:
Suppose the expression
disq
(
a
)isa
fluent
stating whether or not action
a
is abnormally disqualied in a state. In the light of this new interpretation, a
disqualifying condition
F disq
(
a
) reveals itself as state constraint so that
disq
(
a
) holds whenever
F
is true. In particular, any such constraint should
give rise to the indirect eect
disq
(
a
) whenever
F
is brought about. Fluent
in
(
pt
) being made true by some action, for instance, is thus expected to
trigger the indirect eect
disq
(
ignite
) according to
in
(
x
)
disq
(
ignite
),
now to be taken as state constraint.
Fluents representing disqualications in conjunction with suitable state
constraints enable us to model, by means of ramication, situations where an
abnormality is caused by other actions. Whenever some
disq
(
a
) is initiated
in this way, then that should no longer receive special attention insofar as
comparing the abnormalities occurring in models is concerned. This reflects
the intention to consider normal any case of abnormal action disqualications
which are deliberately brought about. The question remains, however, how
the `real' abnormalities are assumed away by default. If, as we have argued,
this cannot be done globally over all states, then what is the alternative?
The answer is that abnormalities are to be minimized solely in the initial
state and so to leave it to both persistence and ramication to take care of the
correct evolution of these fluents as time advances. We recall Example 3.2.2
to supply the reader with a feeling of how that strategy solves the problem
of causally motivated preferences among abnormalities. Let the two fluent
formulas
9x:
)
8x
[
heavy
(
x
)
disq
(
insert
(
x
)) ]
in
(
x
)
disq
(
ignite
(3.1)
be state constraints. Given the observation
:
runs
^:
in
(
pt
) after [],itis
consistent to assume the initial state be
S
=
f:
runs
; :
in
(
pt
)
; :
heavy
(
pt
)
; :disq
(
ignite
)
; :disq
(
insert
(
pt
))
g
Noticeably, all abnormalities are denied in this state. It is clear, therefore,
that
S
is preferred according to the proposed criterion. As a matter of fact,
it is the only state that possesses this property. For if
) were true,
then this would entail an abnormality according to the second state constraint
(
heavy
pt
