Argumentation-based negotiation in a social context
Argumentation-based negotiation (ABN) is gaining increasing importance as a fundamental method of interaction in multi-agent systems. In essence, ABN enhances the ways agents can interact within a negotiation encounter. In particular, it allows agents to justify their demands, criticise each others’ proposals, and add comments to their statements during a negotiation encounter. Furthermore, ABN gives them the capability to exchange explicit arguments, such as promises, threats, appeals, and other forms of persuasive locutions, to influence one another during a negotiation dialogue. Such enhancements lead to richer forms of negotiation than have hitherto been possible in game-theoretic or heuristic-based models. Therefore, many argue that endowing the agents with the ability to argue during their negotiation interactions, not only facilitates more realistic rational dialogues, but also allows an effective means of resolving different forms of conflicts endemic to multi-agent societies. Even though ABN is argued to be an effective means of resolving conflicts, its operation within multi-agent systems incurs certain computational overheads. In particular, it takes time for an agent to argue and convince an opponent to change its demands and yield to a less favourable agreement within an ABN encounter. It also takes computational effort for both parties of the conflict to carry out the necessary reasoning required to generate, select, and evaluate an appropriate and a convincing set of arguments required for such an encounter. On the other hand, within a multi-agent society, not all conflicts need to be resolved. In some instances conflicts can be avoided by other nonarguing means. For instance, in certain situations agents may be able to avoid conflicts by finding an alternative resource to achieve their actions instead of arguing over a conflicting one. They also may be able to re-plan to achieve the same objective through a different means and, thereby, remove the conflict without argument. In the presence of such overheads and given the alternatives available, this thesis argues that computationally bounded entities such as agents need to consider two critical questions before they use ABN to manage their conflicts. First is when to argue; that is, under what conditions would ABN, as opposed to other non-arguing methods, present a better option for agents to overcome conflicts. Second is how to argue; that is, a computationally tractable method and a set of strategies to successfully formulate such sophisticated ABN dialogues. To this end, this thesis forwards a detailed theoretical and empirical study to address both these research questions. In more detail, first we formulate a novel ABN framework that allows agents to argue, negotiate and, thereby, resolve conflicts in structured multi-agent systems. The framework is unique in the way that it explicitly captures social influences endemic to such agent societies and, in turn, allows agents to use them constructively in their ABN dialogues. Having formulated the framework, we then map it into the computational context of a multi-agent task allocation scenario. In so doing, we bridge the gap between theory and practice and provide a test-bed to evaluate how our ABN model can be used to manage and resolve conflicts in multi-agent societies. Our experimental analysis on when to argue shows a clear inverse correlation between the benefit of arguing and the resources available within the context. It also shows that arguing selectively is both a more efficient and a more effective strategy than doing so in an exhaustive manner. Furthermore, we show that when agents operate under imperfect knowledge conditions, an arguing approach allows them to perform more effectively than a non-arguing one. On the issue of how to argue, we show that arguing earlier in an ABN interaction presents a more efficient method than arguing later in the interaction. Moreover, during an ABN interaction, allowing agents to negotiate their social influences presents both an effective and an efficient method which will enhance their performance within a society.