ComplLLM: Fine-tuning LLMs to Discover Complementary Signals for Decision-making

2026-02-22

AI Summary

This work addresses the challenge of effectively leveraging complementary information to enhance collective performance in multi-agent decision-making. The authors propose ComplLLM, a novel framework that, for the first time, integrates the principle of complementarity into the post-training phase of large language models (LLMs). By employing decision-theoretic reinforcement learning, ComplLLM uses complementary information as a reward signal to guide LLMs toward proactively generating new signals that provide incremental value over existing decisions. This approach not only strengthens the judgment capabilities of downstream decision-makers but also yields interpretable justifications for the generated complementary signals. Experimental results demonstrate that ComplLLM successfully recovers known complementary information and produces reasonable, explainable supplementary signals in both synthetic and real-world tasks, significantly improving the decision-making efficacy of multi-agent systems.

Technology Category

• Multiagent Systems: Multiagent Learning
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Planning and Decision Support for Human-Machine Teams

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Multi-agent decision pipelines can outperform single agent workflows when complementarity holds, i.e., different agents bring unique information to the table to inform a final decision. We propose ComplLLM, a post-training framework based on decision theory that fine-tunes a decision-assistant LLM using complementary information as reward to output signals that complement existing agent decisions. We validate ComplLLM on synthetic and real-world tasks involving domain experts, demonstrating how the approach recovers known complementary information and produces plausible explanations of complementary signals to support downstream decision-makers.

Problem

• complementarity
• multi-agent decision-making
• large language models
• decision support
• complementary signals

Innovation

• ComplLLM
• complementary signals
• fine-tuning
• multi-agent decision-making
• decision theory