It may be that an agent in a multiagent system needs to contact other agents to perform event diagnosis. One reason for this may be to obtain information that it does not have. For example, an AUV may notice that it is drifting off course; one reason for this may be that it is in a current. If another AUV has recently traveled through the same region, it can ask it if it, too, experienced difficulty maintaining its heading-if so, then it's likely that there is a current present.
Event diagnosis may also involve taking actions; this is true, for example, in medical diagnosis (e.g., perform a lab test) as well in the AUV domain (e.g., attempt to back up to see if the vehicle is trapped in a net). An agent may need to ask others to take actions on its behalf to help it diagnose an event it has noticed. For example, an AUV may ask another to visually inspect its thrusters to look for damage after it notices that its motion has stopped. When such coordination is necessary, the agent faces the usual quandary of how much to tell its partner(s) about why it has requested the actions: Should it simply ask for an action to be performed, or should it give a rationale for the action, so the other agent can use its own knowledge and initiative to better help it?
Even if the agent needs no help from others, if it must take actions to diagnose an event, it must first assess the actions' impact on others. It may, depending on the actions to be taken, need to inform or even negotiate with others.
The knowledge needed for diagnosis in a sense goes somewhat beyond the knowledge of others and their actions, etc., necessary for the other phases of event handling. Event diagnosis also requires the kinds of associational/causal or model-based knowledge needed for other diagnostic tasks.