ABSTRACT
The real world is not a nice place, at least for intelligent artificial agents. Consider an autonomous factory robot whose job is loading trucks. A foreman tells it “Put the box of widgets on the Acme truck, then meet me back here at 3 o’clock.” The robot dutifully begins considering how to do this. Two things are immediately obvious. First, some partial commitment to future actions is needed, or else the sequence of the two events will not be guaranteed to be correct. Consequently, a plan should be formulated. Second, the robot probably has not been given enough information—there may be several boxes of widgets and more than one Acme truck at the loading dock. Consequently, the robot must ask the foreman for additional information.
