To do this, an agent must have a good model of the other agent's activities, position, sensor capabilities, and of their internal models of the world. The former three are needed so that the agent can determine whether the others are at vantage points to witness the event and whether they have the necessary sensory and attentional resources to detect the event when it occurred. The latter is needed so that the agent can predict whether the event looked like an event to the other agents. For example, suppose during a cooperative underwater photography task (e.g., the MAVIS task [Turner et al., 1991]), AUV A notices that AUV B's light turned off when, from A's standpoint, it should not have. AUV A should predict that B likely (but not necessarily) also detected the event; it should also predict that another of the group's AUVs, C, that is not involved in the current task, would likely not have noticed-to it, seeing B's light go off would not mean that something unexpected had happened. Consequently, A needs only be concerned with B having simultaneously noticed the event.
If an agent believes that others have also noticed an event, then it needs to
enter into negotiation (see, e.g., Sycara, 1989) to
decide which of them will handle the event. The possibilities are: the agent,
the agent and others (possibly including the one being negotiated with), or
others and not the agent. If the agent has no responsibility for the event,
then its processing of it can terminate.