MUSE: An Interactive Meta-Agent for Understanding and Steering LLM-powered Data Science Systems

2026-08-17

AI Summary

This study addresses the challenges of diagnosing system behaviors and regulating reasoning processes in LLM-based data science systems by proposing MUSE, an interactive meta-agent. Through multi-level execution trace reconstruction, context-anchored interaction, and mixed-initiative guidance mechanisms, MUSE precisely maps user intent to system remediation commands, enabling transparent comprehension and efficient intervention of agent workflows. Experimental results demonstrate that MUSE significantly improves debugging efficiency for complex tasks while effectively enhancing users’ cognitive sense of control and operational confidence. These findings establish a novel paradigm for explainable human-AI collaboration, bridging the gap between opaque autonomous agents and interpretable human oversight in sophisticated data science applications.

Abstract

Recent advances in large language models have enabled a new class of agentic data science systems that allow users to complete complex data science workflows through natural language. Although these systems can significantly reduce manual effort, it remains difficult to diagnose their behavior and steer the reasoning process when failures or unexpected outputs occur. We present MUSE, an interactive meta-agent that enhances user understanding and control of agentic data science systems by (1) dynamically restructuring low-level execution traces into multiple semantic levels that support navigation from high-level overviews to low-level implementation details; (2) enabling users to reference specific workflow steps in context to ask grounded questions, provide feedback, and revise problematic steps without manually locating relevant execution history; and (3) supporting mixed-initiative steering by surfacing suspicious steps for inspection, scaffolding the repair process, and translating user repair intent into contextualized instructions for the underlying agent. In a between-subjects study (n = 15), MUSE improved task efficiency and increased users' confidence in understanding and steering agentic data science workflows.

Problem

• Agentic Data Science Systems
• Large Language Models
• System Diagnosis
• Reasoning Steering

Innovation

• Meta-Agent
• Interactive Steering
• Execution Trace Restructuring
• Mixed-Initiative
• Agentic Data Science