Information Technology Reference
In-Depth Information
this paper marked the beginning of a shift away from the original concept
of the Qualication Problem. Instead of being concerned with unlikely ac-
tion preconditions to be assumed away by default, the Qualication Problem
there has been taken as the task of extracting implicit action preconditions
from general knowledge, e.g., state constraints|a task which we have already
considered in the context of the Ramication Problem (see Section 2.8). The
fundamental dierence to the plain Qualication Problem is that any such
implicit condition needs to be explicitly veried, hence cannot be assumed
away prior to concluding that the action in question is executable. This rein-
terpretation of the Qualication Problem has been taken up in subsequent
work, e.g., [38, 68, 99]. There is no urge for arguing against the view that
accounting for implicit action preconditions is part of some broader Quali-
cation Problem. But denying that the core is to providing means to assume
away, by default, the occurrence of
a priori
unlikely obstacles amounts to an
oversimplication of the problem.
One of the marginal consequences of dealing with the Qualication Prob-
lem is that a property called \restricted monotonicity," which has been
claimed generally desirable for theories of actions in [67], is no longer so
when facing qualication scenarios. A formalism possesses this property if
additional observations can only increase the set of observations that are en-
tailed by a domain description. This kind of monotonicity is obviously not
appropriate in case of observations following by default.
Exploiting solutions to the Ramication Problem when addressing the
problem of caused vs. unmotivated action disqualications has rst been pro-
posed in [110]. An earlier general alternative to global minimization, namely,
chronological ignorance
[102, 103], is in principle capable of treating correctly
our key example with the potato being deliberately placed in the tail pipe
(viz. Example 3.2.2), but the approach suers from another, inherent limita-
tion. Roughly speaking, the crucial idea underlying the principle of chrono-
logical ignorance is to assume away, by default, abnormal circumstances but
to prefer minimization of abnormalities at earlier timepoints.
11
Formally, a
certain kind of modal logic is employed as a means to express the distinction
between provable facts and propositions which might or might not be true.
Our example domain, for instance, one would formulate in this framework by
these two action descriptions:
2
T
rue
(
(
x
)
;t
)
^ 2
T
rue
(
:
(
x
)
;t
)
^ 3
T
rue
(
:
(
x
)
;t
)
insert
in
heavy
2
T
rue
(
in
(
x
)
;t
+ 1)
(3.15)
the perspective of causality. The strong resemblance is apparent to the obser-
vation that a potato being too heavy to lift is truly abnormal as opposed to
the expectation that one will fail to start the car after attempting to place a
potato in the tail pipe.
11
This explains the naming: Potential abnormal disqualications are
ignored
whenever possible, and this is done in
chronological
order. Assuming away
obstacles whenever their occurrence cannot be proved, Shoham also calls the
\ostrich" principle, or: \what-you-don't-know-won't-hurt-you."
