Generative Negotiation for Creative Agreements in Multi-Agent Systems

Shun Okuhara, Takayuki Ito

Abstract: How can AI systems augment human negotiators' capacity for creative invention, rather than merely optimizing within boundaries humans have already drawn? Automated negotiation has traditionally assumed a fixed space of issues, values, and utility functions, a constraint that renders negotiation epistemically closed. Yet breakthrough agreements arise when negotiators reframe the problem itself. We propose Generative Negotiation, a paradigm that treats negotiation as the co-creation of the space in which agreement becomes possible. We model negotiation spaces as dynamic semantic structures, introduce generative operators to expand them, and identify conditions under which creative surplus arises. We propose a computational architecture that integrates Large Language Models as dimension generators and articulate challenges, including novelty verification, convergence guarantees, and the compounding biases introduced by human and AI agents when jointly expanding negotiation spaces.