Information Technology Reference
In-Depth Information
incorporated by any action theory meant to address other than articially
simplied domains.
Finally, the innocent word \adequate," which the reader may even have
overlooked in the above denition, is probably the most crucial aspect. It
means that the specication of actions and their eects shall be as natural
as possible. For example, it would not only be inconvenient but clearly most
unnatural to explicitly state the result of executing an action in every possi-
ble situation. Rather one wants to specify that, say, carrying the newspaper
always causes it to change its location|no matter what color it is or whose
party the current President belongs to etc. Likewise, one would want to avoid
re-specifying the eects of transporting an object in case it contains (or is
underneath or attached to etc.) another object. Rather the fact that this
additional object also changes its location should be inferable from general
knowledge of the world. In providing all this, action theories always include
a more or less implicit general notion of time, change, and causality. This
indicates that the adequacy requirement is what makes action theories so
special|and it is what this topic is all about.
Action theories have much in common with logic. They are based on a for-
mal language and they include an entailment relation among the expressions
in this language. This relation determines the way conclusions are drawn from
specications. Yet entailment in action theories is somewhat dierent from en-
tailment in so-called general purpose logics, such as classical rst-order logic,
say. The reason is that the entailment relation reflects the special notions
of change and causality inherent in action theories. This usually makes the
formal denition of how to draw conclusions much more complex compared
to the majority of general purpose logics. It also means that any enrichment
of the ontology of an action theory necessitates changes in the denition of
entailment.
Both the fair complexity and frequent changes of the notion of entailment
in action theories constitute an important drawback in view of automating
reasoning. This favors general purpose logics as means to this end. Research
in automated deduction in rst-order logic, for instance, has made noticeable
progress in the past decades. It might be unwise not to exploit this devel-
opment for automating reasoning in action domains. Fortunately this can be
achieved without losing the major advantage of action theories, namely, their
naturalness when it comes to formalizing domains and scenarios. What needs
to be done towards this end is to axiomatize in, for instance, rst-order logic
the characteristics of an action theory. In other words, the implicit theory
of change and causality is to be made explicit. It became common to call
\foundational axioms" the resulting encoding that characterizes a particular
action theory. Additional axioms then represent knowledge about a specic
action domain and scenario. Together they provide a (hopefully) suitable
encoding which allows to draw all conclusions suggested by the underlying
action theory but by means of a general purpose logic. Notice that the ques-
tion whether such an encoding is suitable is a precise mathematical problem
