Dry-land algorithm

In Section 4.7, it was shown that the dry-land algorithm is important for belief revision when acts do not have preconditions. In the example, the travel agent chooses between offering a window seat and chatting, and the dry-land algorithm infers from its choice of chat that it either does not have a window seat or that it believes the user does not want one. The system's model of the user is thus revised. Here is another interesting example that would be a worthwhile demonstration of the algorithm. It is the problem of a job interviewee who chooses not to admit that he does not have a driver's license, until he is explicitly questioned. Instead he chooses to talk about something more impressive. In this context, the interviewer can use the dry-land algorithm to revise downwards the belief that he has a driver's license. He must do this since if he thought the interviewee did have one, then his rational choice would on the contrary have been to boast about it. The belief is revised to the decision surface between admitting and keeping quiet.

The planner is capable of handling this problem, but it has not been tested. Perhaps in the future the planner could be used to analyse job interview transcripts. A diagram of the plan library is given in figure 6.5.

Figure 6.5: Game tree for the interview problem