This demonstration looks at the revision of the system's model of the user's stereotype model of the system, in response to the first act in the dialogue. First, the offer alternative was explored. Recall figure 4.17 which shows the decision surface between offer and chat. To force the system to choose offer, bel(have-seat) was set mutually to 0.2, whereas intend(window-seat) was set mutually to 0.8. The belief revision mechanism revises beliefs at all levels, and therefore every second level remains identical through the course of the dialogues. The user's belief revision mechanism would be expected to update bel(have-seat) since it is a precondition to offer. Also it would be expected that the "dry-land" algorithm would maintain the user's beliefs about the system by pushing the belief state towards the centre of the offer region. It turns out that in the user's model of the system, intend(window-seat) increases to 0.833 in this dialogue, approaching the approximately 0.9 centre. bel(have-seat) approaches the approximately 0.4 centre by increasing to 0.311. Then, the precondition is observed, and the corresponding belief is set to 1, by ordinary belief revision.
The chat alternative was also explored. To force the system to choose chat, bel(have-seat) was set to 0.8, whereas the intend(window-seat) was set to 0.1. It would be expected that these beliefs would be pushed towards the centre of the region in the user's model. That is what happened. intend(window-seat) moved to 0.16, and bel(have-seat) moved to 0.66.