Collaborative Problem-Solving in an Optimization Game

TL;DR

This paper presents a novel dialogue game where agents collaboratively solve a complex Traveling Salesman problem, combining language models with symbolic reasoning, achieving notable success in self-play and human collaboration.

Contribution

It introduces a new collaborative dialogue framework for NP-hard problems, integrating LLM prompting with symbolic state tracking for improved problem-solving.

Findings

Best agent solves 45% of games optimally in self-play

Agent successfully collaborates with humans and generalizes to new graphs

Demonstrates effective integration of language models with symbolic reasoning

Abstract

Dialogue agents that support human users in solving complex tasks have received much attention recently. Many such tasks are NP-hard optimization problems that require careful collaborative exploration of the solution space. We introduce a novel dialogue game in which the agents collaboratively solve a two-player Traveling Salesman problem, along with an agent that combines LLM prompting with symbolic mechanisms for state tracking and grounding. Our best agent solves 45% of games optimally in self-play. It also demonstrates an ability to collaborate successfully with human users and generalize to unfamiliar graphs.