TL;DR
This paper proposes a framework for developing virtual negotiation agents that mimic human social and emotional strategies, moving beyond purely rational offer exchanges. The goal is to create agents that can effectively teach negotiation skills by building rapport, utilizing deception, and understanding long-term relationships, crucial for more realistic human-agent interactions.
What Problem Does This Paper Address?
Current virtual agents used in negotiation often operate under strict, rapid offer-exchange protocols, largely ignoring the nuances of human interaction. Human negotiation, however, is deeply influenced by factors like trust, rapport, emotions (anger, threats), and the long-term implications of relationships. These agents are therefore poor models for teaching negotiation or acting as realistic partners. The paper aims to bridge this gap by developing agents that incorporate these human-like social and emotional dynamics, particularly in the context of repeated interactions and building reputation within a marketplace of agents.
Methods
The authors propose a three-step approach:
- Empirical Work: Conducting human subject studies to identify effective human negotiation strategies, such as "favors and ledgers" (accepting short-term unfairness for future gains) and understanding the role of trust and social dependencies. This step emphasizes the need for data-driven insights into how humans interact with both other humans and agents.
- Platform Development: The development of a platform called IAGO (Interactive Arbitration Guide Online) is highlighted. IAGO supports communication across multiple channels: offers, messages, and emotions, enabling agents to lie, intimidate, and build rapport. It also logs negotiation metrics for validation.
- Agent Development and Validation: The proposed work focuses on using IAGO to develop agents that build rapport and trust, effectively use deception, and maintain models of opponent preferences and strategies. A key future direction is to move beyond dyadic interactions to agents operating in a marketplace, prioritizing long-term value and reputation.
Datasets
Not discussed in the paper. The authors mention using empirical data gathered from "numerous human-human and human-agent studies" but do not specify particular datasets.
Key Insights and Ideas
- Beyond Rationality: Negotiation is not just about optimal offer exchange; social factors like trust, rapport, and emotions are critical for effective human negotiation.
- Long-Term Value: In repeated negotiations, building relationships and reputation can lead to greater long-term value than purely self-interested, short-sighted strategies.
- Emotional Intelligence: Agents need to be able to express and interpret emotions, not just as a communication channel, but as a strategic tool to influence outcomes.
- Reputation Matters: In a multi-agent marketplace, a negotiator's reputation, built over time through their tactics (e.g., misrepresentation, hardline offers), will significantly impact their success.
- Dyadic vs. Multi-Agent: Current agent research often focuses on one-on-one interactions, but real-world negotiation often involves complex networks and evolving relationships.
Key Results
Not discussed in the paper. The paper outlines a proposed work plan and framework rather than presenting specific experimental results or quantitative findings.
Limitations and Future Work
Limitations: * The current models of human-agent interaction are largely based on dyadic (one-to-one) interactions, which oversimplify true human negotiation. * While the IAGO platform supports emotional communication, the paper doesn't detail the specific mechanisms or effectiveness of these emotional strategies in agent development.
Future Work: * Develop agents that operate within a marketplace of agents, focusing on building long-term value and reputation. * Enhance agents' awareness of social reputation and the implications of their tactics on it. * Continue to base agent models on empirical data from human-human and human-agent studies. * Validate agents by prototyping and testing them over the web using the IAGO platform. * Explore automatic, real-time analysis of negotiation performance for on-the-fly feedback in training contexts.
Summary updated