Optimal Consensus Building Using Negotiation History via Transformer

Abstract

In recent years, ”automatic negotiation agents,” which use AI to help solving negotiation problems have attracted much attention. A negotiation problem leads to the convergence of an agreement satisfactory to decision makers among multiple alternatives. The introduction of automatic negotiation agents is expected to maximize the utility of decision makers for large-scale negotiation problems. The best current strategies for automatic negotiation agents use deep reinforcement learning (DRL). However, DRL method cannot handle the case where there is no environment for prior negotiation. The objective of this study is to acquire negotiation strategies by using negotiation histories with other agents for two-party negotiations. In the proposed method, we introduce a Decision Transformer that can handle causal relationships by time series more explicitly than conventional DRL. Computer experiments show that the proposed method performs better than Behavior Cloning (BC) with the Transformer.