AI and the World's Hard Problems · Theoretical computer science · arXiv · 2026-10-19
ParetoChat: An LLM-Based Interface for Interactive Multi-Objective Optimization
Decision makers often struggle to turn qualitative preferences into the numerical weights or reference points that multi-objective optimization methods require. ParetoChat is an LLM-based expert-agent interface that converts natural-language input into scalarization parameters, invokes an optimization system, presents candidate solutions, and collects feedback iteratively. In simulations on multi-objective binary knapsack and set-partitioning instances, with an LLM decision-maker agent, it guided the search efficiently toward preferred solutions compared with DESDEO methods, and an application to courier-order assignment showed users exploring trade-offs through natural-language requests.