Multi-issue negotiation processes by evolutionary simulation : validation and social extensions

We describe a system for automated bilateral negotiations in which artificial agents are evolved by an evolutionary algorithm. The negotiations are governed by a finite-horizon version of the alternating-offers protocol. Several issues are negotiated simultaneously and negotiations can be broken off with a pre-defined probability. In our experiments the bargaining agents have different preferences regarding the importance of the issues, which enables mutually beneficial outcomes. These optimal solutions are indeed discovered by the evolving agents. We also present an extended model of the evolving agents in which the agents use a ``fairness'' norm in the negotiations. This concept plays an important role in real-life negotiations and experimental economics. In the implementation with fairness, agents also evaluate a potential agreement on its fairness and reject unfair proposals with a certain probability. In our model, re-evaluation can take place in each round or only if the deadline of the negotiations is reached. In both cases, fair outcomes can be obtained. When fairness is applied in each round, the results become much more robust and rather insensitive to the actual fairness function. To validate our system, the computational results are compared to game-theoretic (subgame perfect equilibrium) results. The influence of important model settings, like the probability of breakdown in negotiations, the length of the game, or the influence of fairness, as well as the proper settings of the EA parameters and their sensitivity are substantially investigated in this validation part.