What the Records Don't Carry: A Position on Researcher-AI Co-Adaptation in Exploratory Laboratory Research

Zeyu Li, Ziang Liu, Xiaoman Yang, Dan Luo

Additional Submission Instructions: For the camera-ready version, please include the author names and affiliations, funding disclosures, and acknowledgements.

Track: Track 1: Original Research/Position/Education/Attention Track

Keywords: AI for Science, large language models, exploratory experimentation, tacit knowledge, human-AI co-adaptation, physical laboratory research, position paper

TL;DR: Exploratory laboratory research depends on judgment that formal records do not carry, and reaching it with AI requires researchers and AI systems to co-adapt through sustained interaction.

Abstract: In exploratory research in physical laboratories, much of what guides discovery forms through direct work with materials and is only partly available to formal records. AI systems built on published literature and structured data have no way of knowing what these records do not carry. AI systems fill these gaps without recognizing them and produce output that only the researcher can evaluate. As researchers in a biomaterials and bioengineering laboratory, we demonstrate through a sustained collaboration with LLM-based AI systems how these difficulties compound in practice: judgment from handling materials did not transfer through the files AI received, and the reasoning that trained researchers apply by default was absent from AI’s output. Correcting these gaps surfaced this judgment and reasoning, which had been invisible to both sides. We argue that bringing AI systems into exploratory research in physical laboratories beyond these fragments requires co-adaptation: as the researcher becomes better at recording what would otherwise stay invisible to AI, and AI becomes better at recognizing what records do not carry, both adapt through sustained interaction.

Submission Number: 254