Show HN: Mediator.ai – Using Nash bargaining and LLMs to systematize fairness
Mediator.ai harnesses LLMs and Nash bargaining to systematically resolve disputes, crafting fair agreements by understanding and synthesizing each party's preferences. This Show HN has captured the community's attention for its innovative application of AI to a traditionally human-centric challenge. The discussion delves into the complexities of algorithmic fairness, power dynamics, and the vast potential for AI to democratize conflict resolution.
The Lowdown
Mediator.ai presents an innovative approach to dispute resolution, combining large language models (LLMs) with the Nash bargaining solution and genetic algorithms to systematize fair agreement generation. The founder, inspired by a personal prenup experience, recognized the challenge in defining utility functions for Nash bargaining and saw LLMs as a way to estimate preferences through comparative analysis. The service allows parties to privately articulate their needs to an LLM, which then uses these preferences to guide a genetic algorithm in finding mutually acceptable deal terms.
Key aspects of Mediator.ai:
- LLM-powered preference capture: Parties interact with an LLM to express their wants and needs, which helps translate qualitative preferences into quantifiable utility estimates.
- Nash bargaining application: Addresses the long-standing difficulty of applying John Nash's game theory solution by using LLMs to infer the necessary utility functions.
- Genetic algorithm for agreement generation: The system iteratively drafts and refines agreements, aiming for an optimal solution that maximizes the product of the parties' utilities, representing a "fair" outcome.
- Real-world applicability: Demonstrated with a detailed example of a bakery partnership dispute (Maya and Daniel) where the tool proposes a nuanced 60/40 equity split with a path for the less active partner to regain equity, alongside other crucial clauses.
Mediator.ai aims to democratize mediation, making sophisticated negotiation strategies accessible for various disputes, from personal relationships to business conflicts. By automating the process of finding compromises, it seeks to move past subjective arguments and towards systematically derived, acceptable solutions.
The Gossip
Fairness in the Face of Power
Commenters debated the concept of "fairness" within algorithmic mediation, particularly when power dynamics or self-interest are at play. Some argued that Nash bargaining inherently accounts for rational self-interest and prevents one party from simply demanding the maximum, as it's sensitive to differences in utility, not absolute values. However, skepticism remained regarding its real-world applicability in contentious, low-trust situations where parties might prioritize being "ahead" over a truly fair outcome, especially if one side holds significant power or indifference.
The Bakery Dilemma: A Case Study Critique
The provided example of a bakery partnership dispute sparked significant discussion. While the proposed 60/40 split with a path back to 50/50 was presented as a fair compromise, many questioned its equity. Some felt the less active partner, Daniel, was "screwed over" given an initial 50/50 agreement and his continued contributions, such as paying rent. Others countered that Daniel hadn't maintained his 50% commitment to the business, making the proposed solution a reasonable way to acknowledge differing levels of engagement and provide a structured path for future equity restoration.
Mediating the Masses: Broad Applications & Limitations
The community enthusiastically explored the vast potential applications of AI-driven mediation, suggesting its utility for small-scale disputes like roommate arrangements, HOA conflicts (highlighting the high cost of traditional mediators), and even contemplating its use in large geopolitical conflicts. Concurrently, practical concerns were raised, including the critical need for clear privacy policies given the sensitive nature of information shared during mediation, and feedback on the website's user experience and clarity.