Can agents acquire human-like behaviors in a sequential bargaining game? - Comparison of roth's and Q-learning agents

Abstract

This paper addresses agent modeling in multiagent-based simulation (MABS) to explore agents who can reproduce human-like behaviors in the sequential bargaining game, which is more difficult to be reproduced than in the ultimate game (i.e., one time bargaining game). For this purpose, we focus on the Roth's learning agents who can reproduce human-like behaviors in several simple examples including the ultimate game, and compare simulation results of Roth's learning agents and Q-learning agents in the sequential bargaining game. Intensive simulations have revealed the following implications: (1) Roth's basic and three parameter reinforcement learning agents with any type of three action selections (i.e., e-greed, roulette, and Boltzmann distribution selections) can neither learn consistent behaviors nor acquire sequential negotiation in sequential bargaining game; and (2) Q-learning agents with any type of three action selections, on the other hand, can learn consistent behaviors and acquire sequential negotiation in the same game. However, Q-learning agents cannot reproduce the decreasing trend found in subject experiments. © Springer-Verlag Berlin Heidelberg 2007.