IIT Delhi researchers create AI-Agent 'AILA' to conduct real experiments like human scientists

New Delhi: Imagine walking into a laboratory where an AI agent is carefully adjusting a microscope, running experiments and analysing results, all without human intervention.

Researchers from the IIT Delhi, along with collaborators from Denmark and Germany, have turned it into reality, as detailed in their recent Nature Communications publication titled "Evaluating large language model agents for automation of atomic force microscopy."

Until now, AI models like ChatGPT have served mainly as digital assistants, helping draft documents, answer questions, and analyse data. However, the researchers have now pushed these boundaries much further, developing AILA (Artificially Intelligent Lab Assistant), an AI agent capable of stepping into real laboratories and carrying out scientific experiments from start to finish, much like a human scientist.

"AILA helps me with my daily experimental tasks and speeds up my research progress significantly. Previously, it would take a full day to optimise microscope parameters for high-resolution, noise-free images. Now, the same task is completed in just 7–10 minutes," said Indrajeet Mandal, PhD scholar at the School of Interdisciplinary Research.

The research focused on the Atomic Force Microscope (AFM), a sophisticated instrument that examines materials at incredibly tiny scales. Remarkably, AILA can now control this complex device, make real-time decisions during experiments, and analyse results independently.

Anoop Krishnan, Professor at Civil Engineering and Yardi School of AI, IIT Delhi, explained, "Previously, AI could only help you write about science. Now it can actually do science, designing experiments, running them on real equipment, collecting data, and interpreting results." The researchers discovered that being excellent at answering science questions does not automatically make AI good at doing science.

Models that aced materials science quizzes struggled with real laboratory situations requiring quick adaptation.

"It is like the difference between knowing driving rules from a textbook versus navigating busy city traffic," Mandal said.

"The research also uncovered critical safety concerns. AI agents occasionally deviated from instructions, highlighting the need for robust safeguards to be implemented to prevent accidents or equipment damage as laboratories move toward greater automation," he added.

This breakthrough aligns with India’s AI for Science and Engineering initiative. The government has recently announced funding through the Anusandhan National Research Foundation (ANRF) to accelerate AI-driven research across the country, he said.

Krishnan noted that technologies like AILA could prove transformative for the nation's scientific ecosystem.

"Autonomous lab assistants can democratise access to advanced experimental capabilities," he said.