AI for Negotiation

Advisor: Ray Friedman

This project focuses on the application of AI in negotiation research, with two distinct types of systems under development:

1. AI for Coding Transcripts

A web application designed for negotiation researchers to automate the categorization of negotiation transcripts. Key functionalities include:

• Categorization: Automatically assigning codes to transcript segments based on predefined coding schemes provided by users. These codes represent categories such as “Substantiation” or “Providing Information,” depending on the coding scheme used.
• Multi-Round Validation: Conducting five rounds of categorization to ensure high consistency and accuracy. The system assigns a consistency value to the outputs, highlighting reliability.
• Efficiency and Cost-Effectiveness: Compared to traditional human coding, which often costs over $5,000 per transcript and requires hours of effort, the AI-based approach offers significant advantages in time efficiency and cost savings.

Challenges

• Data Preprocessing: Utilizing a private dataset of high-quality transcripts and transforming unstructured text into structured data for efficient processing.
• Model Evaluation: Testing various large language models (LLMs), including GPT-4, Claude 2, Claude 3 Sonnet, Claude 3 Opus, Claude 3.5 Sonnet, and Claude 3.5 Sonnet-latest.
• Optimization: Exploring different prompt designs, in-context learning training sets, and evaluation metrics to identify the most effective combinations.

This rigorous approach has yielded exceptional results, with significant contributions to the field. Notably, Professor Ray Friedman will present this work at MIT and Harvard in March 2025.

2. AI for Simulating Real-World Negotiation Scenarios

An AI agent system designed to simulate realistic negotiation scenarios, generating transcripts based on assigned conditions. This tool addresses the high costs associated with collecting real-world negotiation data.

Current Progress

The project is currently focused on the development of this system after successfully building two models for coding transcripts. The simulation system aims to provide a cost-effective and scalable solution for generating high-quality negotiation data.

Learn More

Visit our lab page for further details: AI Negotiation Lab - Vanderbilt University.

Acknowledgments

This project is led by Professor Ray Friedman, with substantial progress made thanks to the dedicated efforts of our research team. The innovations achieved have positioned our lab at the forefront of AI in negotiation research.