A Lab Without Humans: China Unveils an AI System That Develops Materials Autonomously

Scientists at the Chinese Academy of Sciences (CAS) have built a multi-agent AI system for developing new materials that closes the entire loop — from forming a hypothesis to controlling robots in a physical lab — with no human involvement at the intermediate stages. According to Jiqizhixin (机器之心), it's a working example of an "autonomous closed loop" in materials science: the traditional trial-and-error method (the so-called "Edisonian approach") gives way to a combination of several specialized language models and robotic manipulators.

Three Agents Instead of a Research Department

The system is built on a multi-agent architecture that mimics an entire research institute. One agent acts as the "librarian": it analyzes thousands of scientific papers and extracts data on past successful and failed experiments. The second agent, the "theorist," models atomic interactions at the quantum level and predicts the properties of a future material before it is ever synthesized. The third agent, the "engineer," turns abstract chemical formulas into concrete instructions for lab equipment — a protocol a robot can actually execute. The agents don't work in isolation: they exchange data, argue, and correct each other's actions in real time. The result behaves like a full research team rather than a set of separate programs predicting molecular properties.

Closed

Loop: Hypothesis → Experiment → Analysis → New Iteration

When the "engineer" sends a command, the robot physically mixes reagents or sinters powders — with no human standing at the instrument. If the result doesn't match the theoretical model's prediction (and in science that happens almost every time), the system doesn't hit a dead end. It analyzes the sensor data, pinpoints where theory diverged from practice, and launches the next iteration of the experiment on its own. This is the autonomous closed loop that tech giants have talked about for years. The human's job is reduced to setting the initial search parameters and watching a material with the target properties take shape on screen.

Why It Matters: Materials R&D Has Hit a Ceiling

For years, classical materials science has worked through combinations of elements in search of the right superconductor or catalyst. This is the "Edisonian approach": endless trial and error, where every failed attempt costs weeks of lab work. The industry is now hitting a technological ceiling on several fronts at once, from battery capacity to solar panel efficiency, and it needs new materials faster than conventional science can deliver them. The CAS project shows that combining LLMs with robotics can shorten the R&D cycle by tens or even hundreds of times, because the system forms the hypothesis, plans the experiment, and runs it by itself — while Western startups stay focused on generating text and images.

What Happens to Human Scientists

The direct consequence of automating routine lab work is a shift in the role of graduate students and researchers, whose job used to consist of monotonously mixing reagents and calibrating instruments. That routine moves to robots controlled by the multi-agent system. Humans are left to define tasks for the AI agents and interpret the results, including those that contradict classical scientific intuition. In this model, a scientific discovery effectively becomes the product of managing neural networks well, not the result of a single scientist's lucky insight.

What Did the CAS Researchers Present?

A multi-agent AI system for developing new materials that operates as an autonomous closed loop: it formulates a hypothesis, plans an experiment, and carries it out using robotic manipulators in a physical lab — without step-by-step human involvement at each stage.

How Are Roles Divided Among the System's Agents?

The system is structured as a virtual research institute with three specialized agents: the "librarian" analyzes scientific papers and data from past experiments, the "theorist" models atomic interactions at the quantum level, and the "engineer" translates chemical formulas into instructions for lab robots. The agents exchange data and correct each other's actions in real time, mimicking the work of a research team.

What

Happens If an Experiment's Result Doesn't Confirm the Theory?

The system analyzes sensor data from the lab equipment, determines where the theoretical model diverged from the practical result, and automatically launches the next iteration of the experiment — without pausing for human analysis. That's what makes the loop "closed": hypothesis, experiment, and correction all happen without leaving the autonomous circuit.