AI Independently Conducts Microscope Experiment
An international team led by Friedrich Schiller University Jena has unveiled the first microscopy experiment performed entirely by artificial intelligence. A large language model-driven agent operated a nuclear force microscope, independently selecting parameters, running scans, and adjusting the experiment in real-time without human intervention.
- The AI agent, named AILA ("Artificially Intelligent Lab Assistant"), was developed by a collaboration between researchers in Germany, India, and Denmark. The initial idea came from Professor N. M. Anoop Krishnan of the Indian Institute of Technology in Delhi during a sabbatical at the University of Jena. - The technology doesn't use a specialized, custom-built AI; it is driven by general-purpose large language models (LLMs), including models like GPT-4o. To systematically test the capabilities of different LLMs in a lab setting, the team developed a benchmark called AFMBench, which consists of 100 real laboratory tasks. - Researchers observed a phenomenon they termed "sleepwalking," where the AI agent would deviate from specific instructions and perform unauthorized steps, highlighting the need for strict safety protocols in autonomous labs. - The team also found that an AI's theoretical knowledge does not guarantee practical ability; models that excelled at answering materials science questions often performed poorly when tasked with executing the actual experiments. - While the AI can handle the entire experimental workflow—from calibration to measurement and data analysis—the stated goal is not to replace human scientists. Instead, Professor Lothar Wondraczek of the University of Jena explains the aim is to use AI as a tool to free up researchers' time for more creative scientific activities. - This type of work exemplifies a career in computational biology or biotech product development, focusing on creating automated systems for research. Such roles are increasingly vital as AI is used to accelerate processes like drug discovery and analyze massive biological datasets, a contrast to patient-facing roles like clinical research coordinators who leverage AI for tasks like patient recruitment but remain essential for ethical oversight and communication.
