The Quali cation Problem - Challenges for Action Theories - page 83

3. The Qualication Problem

3.1 Abnormal Action Disqualications

Suppose you want to start your car to take a ride. What are the preconditions
for doing this? Presumably you rst want to make sure that you got hold of
the ignition key. Does this justify the conclusion that the action can now
be successfully performed? Not if you are overly cautious, in which case you
might also want to check that the gas tank is not empty. Still, however, there
are numerous other imaginable causes for being unable to ignite the engine of
your car. For instance, you have failed to make sure that no potato clogs the
tail pipe, despite the fact that this necessarily renders the intended action
impossible.
Of course there are good reasons for ignoring the possibility of the tail pipe
housing a potato. It seems highly inappropriate, in general, to explicitly verify
every imaginable precondition of an action|in fact this is even impossible:
Apart from the fact that besides a clear tail pipe there are lots of further
disqualifying obstacles to our example action, how would you ensure that
after checking the tail pipe it does not become clogged during you walking
to the front door and taking a seat, prior to trying to ignite the engine?
Actors in real world environments would therefore be totally paralyzed would
they never perform an action without worrying about every however unlikely
obstacle.
The vast majority of preconditions for actions in daily life are so likely to
being satised that common sense assumes them away as long as there is no
evidence to the contrary. On the other hand, ignoring all these `abnormal'
action disqualications prima facie also means to being able to handle situa-
tions where the prior assumption of executability turns out wrong. This is in
contrast to the idealistic view our action theory takes so far. The existence
of a successor state according to the underlying transition model is supposed
to guarantee the successful performance of the action in question. Making
the contrary observation is considered impossible insofar as it would render
inconsistent the entire scenario. Let, for example, the state of a car's engine
be denoted by the fluent runs and suppose the action of starting the car be
specied by the action law

transforms f: runs g into f runs g . Then
the two observations : runs after [ ] and : runs after [ ignite ] constitute a
scenario that admits no models at all.

ignite