Multi agent collaboration needs reliable shared information states
Right Answers, Wrong States: Hidden Information Failures in Multi-Agent Collaboration
Computation and Language
Summary
Sometimes groups of AI agents get the right answer but keep hidden mistakes in their shared information, which can cause problems later. The authors call this issue 'off-query failure' and studied it in healthcare and disaster response tasks. They found that while the agents often solve tasks correctly, their shared understanding of facts is often wrong. They developed a method called ReGround that improves verifying facts, rebuilding a reliable shared state, and making good decisions in these multi-agent settings.
What this means in practice
- •For healthcare data teams: Improve collaboration systems to ensure shared patient information is accurate and consistent during high-stakes medical decision making.
- •For disaster response coordinators: Build coordination tools that verify and correct shared information among agents to ensure reliable decisions during emergencies.
Authors
Herun Wan, Jiaying Wu, Minnan Luo, Zihan Ma, Fanxiao Li, Nancy F. Chen, Min-Yen Kan
Abstract
Multi-agent systems are often judged by whether they reach the correct answer. This can miss a distinct failure: collaboration may leave behind a corrupted information state even when the immediate decision is correct. We call this an off-query failure. To study this failure in collaborative decision support, we introduce OffQuery, which separately evaluates evidence verification (T1), shared-state reconstruction (T2), and task resolution (T3) in two representative high-stakes settings: healthcare and disaster response. Across GPT, Gemini, and Qwen models, standard collaboration shows much stronger task performance than state reliability. Averaged over 21 model--setting combinations, task resolution reaches 64.7%, while evidence verification and state reconstruction reach only 14.3% and 43.1%. We trace this gap to selective information use: current queries often bypass corrupted facts, which become consequential when later tasks require them. We further introduce ReGround, which resolves conflicting evidence, verifies shared facts, reconstructs a trusted state, and reasons over that state. Across seven models from three families, ReGround improves all three capabilities in every evaluated setting, with average relative gains of 309.0%, 82.9%, and 17.6% on T1, T2, and T3. Reliable collaboration therefore requires both a correct decision and a reliable shared state for future reasoning.