Large Language Models for Human–AI Collaboration in Scientific Experimentations

Scientific discovery increasingly depends on the ability to make rapid, informed decisions during complex experiments. Yet in many domains, from microscopy to spectroscopy, progress remains limited by human intuition and fragmented expertise. We present a framework in which large language models (LLMs) serve as reasoning engines for experimental planning and adaptive decision-making. By combining natural-language reasoning with real-time data interpretation and physics-informed constraints, the system assists scientists in defining goals, selecting optimal measurement strategies, and interpreting outcomes dynamically. This human–AI collaboration transforms experimentation from a reactive process into an adaptive, hypothesis-driven workflow, where decisions evolve with each new observation. Demonstrations in electron microscopy highlight how such models can bridge qualitative reasoning and quantitative control, guiding exploration toward regions of highest informational value. The approach establishes a foundation for context-aware, LLM-assisted autonomy, enabling scalable, explainable, and goal-oriented scientific experimentation across instruments and disciplines.