Correctness Requirements for Multiagent Commitment Protocols Pınar Yolum Department of Artificial Intelligence, Vrije Universiteit Amsterdam De Boelelaan 1081a, 1081 HV Amsterdam, The Netherlands pyolum@few.vu.nl Abstract. Commitments are a powerful abstraction for representing the interac- tions between agents. Commitments capture the content of the interactions declar- atively and allow agents to reason about their actions. Recent work on multiagent protocols define agent interactions as the creation and manipulation of commit- ments to one another. As a result, commitment protocols can be executed flexibly, enabling the agents to cope with exceptions that arise at run time. We study the correctness requirements of commitment protocols that are neces- sary to ensure correct specification and coherent execution. We analyze and for- malize various requirements for commitment protocols and draw relations among them. The main contribution of this analysis is that it allows protocol designers to develop correct protocols by signaling possible errors and inconsistencies that can possibly arise at run time. Since the requirements are formal, they can be incorporated in a software tool to automate the design and specification of com- mitment protocols. 1 Introduction Multiagent systems consist of autonomous, interacting agents. To operate effectively, the interactions of the agents should be appropriately regulated. Multiagent interaction protocols provide a formal ground for enabling this regulation. However, developing effective protocols that will be carried out by autonomous agents is challenging [8]. Similar to the protocols in traditional systems, multiagent protocols need to be speci- fied rigorously so that the agents can interact successfully. Contrary to the protocols in traditional systems, multiagent protocols need to be specified flexibly so that the agents can exercise their autonomy to make choices as best suits them or to handle unexpected situations that arise at run time. This basic requirement rules out many traditional for- malisms, such as finite state machines (FSMs) or Petri nets. These formalisms only specify sequences of actions and leave no room for the agents to act flexibly. Recently, social constructs are being used to specify agent interactions. These ap- proaches advocate declarative representations of protocols and give semantics to pro- tocol messages in terms of social (and thus observable) concepts. Alberti et al. specify interaction protocols using social integrity constraints and reason about the expectations of agents [1]. Fornara and Colombetti base the semantics of agent communication on commitments, such that the meanings of messages are denoted by commitments [6]. Yolum and Singh develop a methodology for specifying protocols wherein protocols