We describe the development of an automated agent that can negotiate efficiently with people in crises. The environment is characterized by two negotiators, time constraints, deadlines, full information, and the possibility of opting out. The agent can play either role, with communications via a pre-defined language. The model used in constructing the agent is based on a formal analysis of the crises scenario using game-theoretic methods and heuristics for bargaining. The agent receives messages sent by its opponent, analyzes them and responds. It also initiates discussion on one or more parameters of an agreement. Experimental results of simulations of a fishing dispute between Canada and Spain indicate that the agent played at least as well as, and in the case of Spain, significantly better than a human player.
Kraus, Sarit, Penina Hoz-Weiss, Jonathan Wilkenfeld, David Andersen, and Amy Pate. 2008. "Resolving Crises through Automated Bilateral Negotiations." Artificial Intelligence Journal 172 (January): 1-18. http://www.sciencedirect.com/science/article/pii/S0004370207001051