Information Technology Reference
In-Depth Information
Despite being a necessary precondition, the absence of obstacles like
\there is a potato in the tail pipe" should, however, not be added to the
condition part of the above action law. For otherwise this would require what
has been said inappropriate, namely, the inspection of all abnormal disquali-
cations of this action prior to assuming it executable. The general challenge,
therefore, is to weaken the strict conclusion that actions are guaranteed to
producing the expected eect once all specied preconditions are satised.
This conclusion should become an assumption
by default
.Assuchitisto
be made as long as there is no information to the contrary. Developing a
formal account of this concept within the framework of an action theory is
the
Qualication Problem
. Solving it is necessary in view of applying action
theories to real-world environments, which do not conform with the idealis-
tic view in that most if not all actions are potentially subject to abnormal
disqualication.
Let us state the problem more precisely. Suppose given the two fluents
runs
and
in
(
pt
) stating, respectively, whether the engine of the car is run-
ning and whether its tail pipe houses a potato. As above, the action
ignite
shall be specied by the action law
g
.
Consider the situation where the only available information concerns the state
of the engine, which is known to be o. Nothing is known as to whether or
not the tail pipe is clogged. The formal specication of this scenario thus
consists of the observation
transforms
f:
g
into
f
ignite
runs
runs
:
runs
after []
Then by default it should follow that
runs
after [
ignite
]. The argument
supporting this conclusion would be that there is no hint at a potato being
placed in the tail pipe and the only `regular' precondition of the action, viz.
that the engine is not running already, is true in the initial state.
1
Our action
is therefore to be assumed executable and so to have the expected eect. But
now consider the situation where we do not only know the engine is o but
also that there
is
a potato in the tail pipe, that is,
:
runs
after
[]
(
)
after
[]
in
pt
Then it should no longer follow that igniting the engine is possible nor, hence,
that the observation
runs
after [
ignite
] holds.
This example reveals a crucial principle with which the Qualication
Problem is inherently connected. Namely, additional observations, in our case
in
(
pt
) after [ ], may force the withdrawal of previously valid conclusions. A
so-called
nonmonotonic
entailment relation for observations is required to
this end. This is in contrast to our current action theory being monotonic in
that sense: Whenever two sets of observations
O
1
and
O
2
satisfy
O
1
O
2
,
1
By \regular" preconditions we generally mean both the conditions occurring in
an action law and the implicit qualications deriving from state constraints.
