In a world that sometimes seems full of conflict, it is important to train corporate, military, and governmental officials to negotiate effectively. It is often difficult to practice these skills in a training session because it demands one-on-one interaction between an instructor and a student. This paper describes a simulated negotiating agent that can be used to hone the skills of human negotiators. The work is part of a broader research agenda on conversational agents.
This particular conversational agent operates from a sophisticated script that models such things as negotiation strategy (issue avoidance, integrative/win-win, distributive/win-lose), as well as conversational tactics to achieve each strategy. The system achieves believable interaction by exploiting the fact that the dialog is highly restricted in both topic and style (in this sense, it is an advanced derivative of Weizenbaum’s work on ELIZA in the mid-1960s). This approach seems promising for achieving the stated goal of the work--to provide a useful training environment for humans learning to negotiate. However, such restricted conversational systems have limited value in the face of the more general goal of creating artificially intelligent conversational agents that can handle free-form human-machine communication.