Information Technology Reference
In-Depth Information
of (3.1). Having both
:disq
(
)
2 S
(the only
regular precondition of inserting a potato), it follows that action
(
))
2 S
and
:
(
insert
pt
in
pt
)
is executable in
S
. The direct eect being
in
(
pt
), we expect the additional,
indirect eect
disq
(
ignite
) deriving from the rst one of our state con-
straints (3.1). The successor state resulting from performing
insert
(
pt
)thus
correctly entails an abnormal disqualication of action
(
insert
pt
. To stress the
point, we again draw the reader's attention to the fact that the initial state
S
,
although entailing an abnormality at a later timepoint, is in itself free of ab-
normalities.
Restricting minimization of abnormalities to the initial state thus provides
a solution to our key example that proved global minimization inadequate
as a formal account of the Qualication Problem. The question immediately
rises, then, whether the improved approach is generally well-founded and not
refutable itself. To begin with, the new minimization strategy is bound to
fail in case it neglects preferences apparently to be made among alternative
ways of minimizing in specic situations. Now, any such preference either
grounds on static reasons, that is, it holds in any state regardless of both
past and future states, or the reasons for a preference lie in the dynamics
of state evolution. As regards static preferences, dierent degrees of
a priori
likelihood of abnormalities shall be neglected for the moment and dealt with
in Section 3.5. For the moment the only possible static reason for preferring
to accept an abnormality is additional information implying its presence. To
account for this is the role of state constraints dening conditions for a fluent
disq
(
a
) being true. Since any acceptable state must satisfy these constraints
by denition, our approach does respect static reasons for preferring to accept
an abnormality.
Reasons to accept an abnormality grounded on the dynamics of state tran-
sition are linked to causality. That is to say, the acceptance of an abnormal
disqualication may be founded on the performance of certain actions. Let
us take for granted the unidirectionality of causality forward in time.
4
Two
points need to be made then. First, suppose that at some stage an action is
performed which brings about a condition for an action
a
being abnormally
disqualied. Whenever this happens, the corresponding fluent
disq
(
a
) should
become true as side eect of the other action. This is achieved by any state
constraint of the form
F disq
(
a
) giving rise to the indirect eect
disq
(
a
)
as soon as
F
gets invoked. Our solution to the Ramication Problem guar-
antees this|provided that the underlying influence information entails that
any fluent occurring in
F
potentially aects
disq
(
a
).
5
4
ignite
This is not meant as oense to the minority of philosophers who take serious
the idea of eects preceding their causes, nor to the minority of physicists
who interpret quantum mechanics in such a way that it gives rise to backward
causality. We appeal to common sense here.
5
This holds even if one abandons the rather idealistic assumption that a domain
specication contain complete knowledge as to the possible reasons for an action
disqualication. The trick is to introduce an articial fluent representing the
`unknown' cause; see Section 3.5 below.
