BALAR : A Bayesian Agentic Loop for Active Reasoning

2026-05-06

AI Summary

This work addresses the limited active reasoning capability of large language models (LLMs) in interactive tasks, particularly their inability to recognize missing information and ask effective clarifying questions. The authors propose a general, fine-tuning-free outer-loop algorithm that introduces Bayesian active inference into LLM-based interaction for the first time. By maintaining a belief over latent task states, selecting clarification questions that maximize expected mutual information, and dynamically expanding the state representation, the method enables task-agnostic, multi-turn structured dialogue. Evaluated on three benchmarks—AR-Bench-DC, AR-Bench-SP, and iCraft-MD—the approach achieves accuracy improvements of 14.6%, 38.5%, and 30.5%, respectively, substantially outperforming existing baselines.

Abstract

Large language models increasingly operate in interactive settings where solving a task requires multiple rounds of information exchange with a user. However, most current systems treat dialogue reactively and lack a principled mechanism to reason about what information is missing and which question should be asked next. We propose BALAR (Bayesian Agentic Loop for Active Reasoning), a task-agnostic outer-loop algorithm that requires no fine-tuning and enables structured multi-turn interaction between an LLM agent and a user. BALAR maintains a structured belief over latent states, selects clarifying questions by maximizing expected mutual information, and dynamically expands its state representation when the current one proves insufficient. We evaluate BALAR on three diverse benchmarks: AR-Bench-DC (detective cases), AR-Bench-SP (thinking puzzles), and iCraft-MD (clinical diagnosis). BALAR significantly outperforms all baselines across all three benchmarks, with $14.6\%$ higher accuracy on AR-Bench-DC, $38.5\%$ on AR-Bench-SP, and $30.5\%$ on iCraft-MD.

Problem

• active reasoning
• interactive dialogue
• information gap
• question selection
• latent state

Innovation

• Bayesian reasoning
• active questioning
• multi-turn interaction
• latent state belief
• expected mutual information