When Will Negotiation Agents Be Able to Represent Us? The Challenges and Opportunities for Autonomous Negotiators

Tim Baarslag, Michael Kaisers, Enrico H. Gerding, Catholijn M. Jonker, Jonathan GratchIJCAI 2017 · DOI

TL;DR

This paper argues that truly autonomous negotiation agents, capable of representing us in complex real-world scenarios, are still a future prospect. It identifies three key challenges – domain knowledge/preference elicitation, long-term perspective, and user trust – that must be addressed in unison to advance the field beyond its current parallel development along separate autonomy dimensions.

What Problem Does This Paper Address?

Negotiation is a fundamental human activity, yet many people avoid it due to fear or lack of skill, contributing to societal issues. The promise of automated negotiation lies in agents that can secure better deals, reduce costs, and alleviate stress. However, despite significant research, truly autonomous negotiation agents are rarely deployed in real-world applications. This paper addresses the gap between the potential of automated negotiation and its current limited adoption by exploring the technological, societal, and ethical challenges that hinder the development of agents capable of independently representing humans in diverse and complex situations. The authors question what is currently missing and what is needed to enable these agents to fulfill their promise.

Methods

The paper doesn't propose a novel algorithm or technique but rather offers a conceptual framework and a roadmap for future research. It proposes an "autonomy diagonal" with three orthogonal dimensions: self-sufficiency, self-directedness, and interdependence. The authors then cluster existing research in automated negotiation around these dimensions. They identify three major challenges that cut across these dimensions:

  1. Domain Knowledge and Preference Elicitation: This involves agents accurately understanding user preferences and the intricacies of the negotiation domain. The paper highlights the need for "preference elicitation on-the-fly" to minimize user burden and adapt to dynamic preferences.
  2. Long-Term Perspective: Current research often focuses on single encounters. This challenge addresses the need for agents to handle repeated interactions, evolving user preferences, and the complexities of long-term relationships, including the use of virtual currencies and reputation systems.
  3. User Trust and Adoption: This crucial aspect focuses on building user confidence through co-participation, transparency, and clear communication of consequences. The paper emphasizes the tension between an agent's need for independent action and the user's need to understand and trust its reasoning.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Autonomy is Multi-Dimensional: True autonomy in negotiation isn't just about being independent; it's a balance of being self-sufficient, self-directed, and capable of interdependent collaboration. Current research has largely developed along these dimensions in parallel, hindering holistic progress.
  • "Preference Elicitation on-the-Fly" is Crucial: Users are often unwilling or unable to provide extensive preference information upfront. Agents need to intelligently infer and actively elicit preferences during the negotiation process, minimizing user fatigue.
  • Long-Term Negotiation is More Than Repeated Deals: It involves managing evolving user needs, building trust through reputation, and potentially using novel mechanisms like virtual currencies, all while navigating psychological biases like fairness.
  • Transparency is Key to Trust, But Not Always to Opponents: While users need transparent reasoning to trust an agent, unpredictable behavior can be a valuable negotiation tactic against opponents, as long as the consequences are clear to the user.
  • Ethical Considerations Mirror Human Dilemmas: As agents become more sophisticated, they will face ethical quandaries similar to those faced by human intermediaries, raising questions about responsibility and the potential for agents to act unethically to maximize profit.

Key Results

The paper does not present quantitative results from experiments. Instead, it offers a qualitative analysis of the state of automated negotiation research and identifies key challenges and opportunities. The "key results" are conceptual: a framework for understanding autonomy in negotiation and a roadmap for future research focused on integrating the three core challenges.

Limitations and Future Work

The authors acknowledge that their paper does not comprehensively survey all research but rather highlights important focal points. They suggest that future work should focus on:

  • Integrating the Autonomy Dimensions: Advancing research that simultaneously addresses self-sufficiency, self-directedness, and interdependence.
  • Developing Metrics for Preference Elicitation: Creating performance-based metrics to assess the impact of supplementary preference information.
  • Modeling Non-Stationary Preferences: Developing cost-efficient methods for tracking evolving user preferences in long-term negotiations.
  • Enhancing User Trust: Researching mechanisms for co-participation, transparency, and adjustable autonomy to build user confidence.
  • Addressing Ethical Dilemmas: Investigating how agents should handle ethical considerations, potentially mirroring human behavior in representation.
  • Exploring Collaborative Control: Developing mixed-initiative systems where humans and agents collaborate effectively.
  • Understanding Representation Effects: Further research is needed to understand how artificial representation affects negotiation behavior, particularly concerning fairness and ethical conduct.

Summary updated

BibTeX
@inproceedings{baarslag2017when,
  author = {Tim Baarslag and Michael Kaisers and Enrico H. Gerding and Catholijn M. Jonker and Jonathan Gratch},
  title = {{When Will Negotiation Agents Be Able to Represent Us? The Challenges and Opportunities for Autonomous Negotiators}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2017},
  doi = {10.24963/ijcai.2017/653},
}
Figure