MOOSE-Copilot: A Web-Based Interactive Assistant for Unified Exploratory and Fine-Grained Scientific Hypothesis Discovery

2026-05-28

AI Summary

This work addresses the fragmentation in current large language models (LLMs) between exploratory ideation and refined hypothesis formulation in scientific discovery, as well as their lack of effective human-in-the-loop guidance mechanisms. To bridge this gap, we propose MOOSE-Copilot, a unified framework that seamlessly integrates exploration and refinement phases for the first time, augmented with a structured human–AI interaction protocol. This protocol enables scientists to steer hypothesis generation throughout the process via initial blueprints, phase routing, and regenerative feedback. An accompanying interactive web interface and tree-based visualization tool substantially lower usability barriers. Experimental results demonstrate that, under expert guidance, our approach significantly outperforms fully autonomous baselines and approaches the performance of an idealized oracle, thereby effectively empowering end-to-end, cross-disciplinary scientific discovery.

Abstract

Large language models (LLMs) show remarkable potential in scientific hypothesis discovery. However, existing approaches face two critical limitations: they treat divergent exploratory ideation and convergent fine-grained refinement as isolated tasks, and they operate autonomously with little to no human guidance. We present MOOSE-Copilot, the first unified framework to bridge this abstraction gap through a formalized human-AI interaction (HAII) protocol. Our system empowers scientists to steer the generative process via three explicit signals: initial blueprints, inter-stage routing, and regenerative feedback. Quantitative evaluations demonstrate that injecting these structured expert signals significantly outperforms purely autonomous baselines, establishing a performance ceiling under oracle guidance. Furthermore, to democratize this paradigm, we develop an intuitive web-based interface featuring interactive tree visualization. This explicitly eliminates the steep learning curve of complex command-line agentic tools, empowering interdisciplinary researchers to directly leverage, visually orchestrate, and accelerate end-to-end scientific breakthroughs.

Problem

• scientific hypothesis discovery
• human-AI interaction
• exploratory ideation
• fine-grained refinement
• large language models

Innovation

• human-AI interaction
• scientific hypothesis discovery
• large language models
• interactive visualization
• fine-grained refinement