[3 Response by Georgeff]
--------------------------------------------------------------------------------------------------
Let us first consider so-called Beliefs. In AI terms, Beliefs represent knowledge of
the world. However, in computational terms, Beliefs are just some way of representing
the state of the world, be it as the value of a variable, a relational database, or symbolic
expressions in predicate calculus.

Desires (or, more commonly though somewhat loosely, Goals) form another essential
component of system state. Again, in computational terms, a Goal may simply be
the value of a variable, a record structure, or a symbolic expression in some logic. The
important point is that a Goal represents some desired end state. Conventional computer
software is �htask oriented�h rather than �hgoal oriented�h; that is, each task (or subroutine)
is executed without any memory of why it is being executed. This means that the system
cannot automatically recover from failures (unless this is explicitly coded by the
programmer) and cannot discover and make use of opportunities as they unexpectedly
present themselves.

These committed plans or procedures are called, in the
AI literature, Intentions, and represent the third necessary component of system state.

Computationally, Intentions may simply be a set of executing threads in a process that
can be appropriately interrupted upon receiving feedback from the possibly changing
world.





[4 Response by Pollack]
--------------------------------------------------------------------------------------------------
It has been generally accepted for many years that agents cannot
possibly perform optimizations over the space of all possible courses of action [17].
Bratman�fs Claim is aimed precisely at helping reduce that space to make the required
reasoning feasible.




[3 Response by Georgeff]