From Automation to Autonomy: A New Era in AI-Driven Scientific Discovery
Excited to share our latest work: “From Automation to Autonomy: A Survey on Large Language Models in Scientific Discovery” now available on arXiv!
In this comprehensive survey, we examine how Large Language Models (LLMs) are fundamentally reshaping scientific research—evolving from simple automation tools into increasingly autonomous scientific agents.
What makes our survey unique?
We introduce a novel three-level taxonomy that captures the progression of LLMs in scientific workflows:
- LLM as Tool: Performing specific tasks under direct human supervision
- LLM as Analyst: Processing complex information with reduced intervention
- LLM as Scientist: Autonomously conducting major research stages
Rather than focusing on narrow applications, we analyze LLMs through the lens of the scientific method itself—from observation and hypothesis development to experimentation, analysis, and refinement.
Key insights:
- LLMs are rapidly advancing from discrete task execution to sophisticated multi-stage workflows
- Current “Level 3” systems can now perform end-to-end research tasks with minimal human guidance
- The field faces critical challenges in robotic automation, transparency, self-improvement, and ethical governance
Our survey maps 90+ research works across this evolving landscape, providing both a conceptual framework for understanding the current state of the field and strategic foresight for its future.
As LLMs continue their progression toward greater autonomy in scientific discovery, we hope this work serves as a valuable resource for researchers navigating this transformative frontier.
Check out the full paper: https://arxiv.org/pdf/2505.13259 Code and resources: https://github.com/HKUST-KnowComp/Awesome-LLM-Scientific-Discovery