Coping with Bad Agent Interaction Protocols When Monitoring Partially Observable Multiagent Systems

Interaction Protocols are fundamental elements to provide the entities in a system, be them actors, agents, services, or other communicating pieces of software, a means to agree on a global interaction pattern and to be sure that all the other entities in the system adhere to it as well. These “global interaction patterns” may serve different purposes: if the system does not yet exist, they may specify the allowed interactions in order to drive the system’s implementation and execution. If the system exists before and independently from the protocol, the protocol may still specify the allowed interactions, but it cannot be used to implement them. Its purpose in this case is to monitor that the actual system does respect the rules (runtime verification). Tagging some protocols as good ones and others as bad is common to all the research communities where interaction is crucial, and it is not surprising that some protocol features are recognized as bad ones everywhere. In this paper we analyze the notion of good, bad and ugly protocols in the MAS community and outside, and we discuss the role that bad protocols, despite being bad, may play in a runtime verification scenario where not all the events and interaction channels can be observed.