next up previous contents
Next: Demonstration 3 Up: Demonstration 2 Previous: Upper branch   Contents

Lower branch

In the lower branch of the game tree, "chat" is chosen by the system. If the user responds with "chat", the state is expected to drift towards bel(have-seat) being false and intend(window-seat) being false, since these are both explanations for the user deciding to chat. This change would be made by the dry-land algorithm. If the user responds with ask, the system responds with give (since its base belief is that it does have a window seat), then both beliefs are set to 1 by precondition inference, with the decaying average possibly moving across the decision surface. Starting at the point [0.5,0.5], the user chose chat, resulting in belief revision by the dry-land algorithm to: p(intend(window-seat)) = 0.486 for the system's belief about the user and p(bel(have-seat)) = 0.321 for the system's belief about the user's belief about the system.

bmceleney 2006-12-19