Information Technology Reference
In-Depth Information
There is a general principle behind the problem we have encountered
with this example. Often a mere state constraint does not contain sucient
information as to what eects it is expected to trigger. From the perspective
of logic, all fluent changes that correct a violation of a constraint are equal.
Some of these changes, however, would never occur in reality as indirect
eect. Consequently, additional domain knowledge is required that allows to
distinguish the deductions which correspond to real eects from the mere
logical consequences which have no equivalent in reality. This is a challenge
of very general nature, which we will have to face at various places throughout
the topic. Our objective, therefore, is to develop a solution on the basis of
a principle (yet to be found) which is as universal as the problem to be
addressed.
2.3 Categorizing Fluents
At the end of the previous section we arrived at the insight that in general a
mere state constraint is insucient for a correct handling of indirect eects.
This raises the issue of the nature of additionally required information. With
regard to our example domain, it would of course be sucient to add the very
specic knowledge that \when closing the switch, an activation of the light
bulb is likely to occur, as opposed to the other switch changing its position."
Yet our aim is to nd a general criterion that helps us telling apart the correct
indirect eects. That is to say, we seek the
reason
for us expecting the light
turns on rather than that the switch moves.
In fact, one essential dierence between the light bulb and the switch is
fairly apparent. Namely, the state of a switch can be aected only by direct
operation. Its state is, in a certain sense, independent of the values of other
fluents. In contrast, there are no means to directly operate the light bulb.
Rather its state completely depends on the values of other fluents. From
this perspective, it is no surprise that toggling a switch may indirectly aect
the light but not the other switch. This observation suggests the following
renement of our rst approach to the Ramication Problem.
In order that only expected indirect eects are generated, the fluents of a
domain are divided into two categories. One of which consists of those fluents
that represent state components which are manipulated by direct operation.
The other category contains all fluents that represent state components which
depend on other components. Let us call
primary
fluents of the former kind
and
secondary
those of the latter. What fluent belongs to which category
cannot, of course, be gathered from the state constraints. Therefore, the
categorization needs to be given as part of the domain specication. On the
basis of this additional information, whenever the procedure of minimizing
change oers a choice between a primary and a secondary fluent, then the
former is preferred to remain unchanged|hence the latter is preferred to
adapt as indirect eect.
