Self-driving lab

A self-driving lab is a laboratory in which an AI model designs each experiment, robots and instruments carry it out, software reads the result, and the model chooses the next experiment, with no person needed between the steps. Scientists set the goal and the limits, such as a molecule that binds a protein strongly, or a material that conducts heat well. The lab then runs the cycle of design, make, test and learn by itself, day and night.

Here is how it works in general. The model starts with what is already known and proposes an experiment: which ingredients, in what amounts and under what conditions. Robots that move tiny amounts of liquid prepare the samples on plates. Instruments, such as a plate reader or a mass spectrometer, measure what happened. Software collects the measurements and gives them to the model in a consistent form. The model learns from the result and picks the next experiment, usually the one that it expects to teach it the most. The cycle then repeats, often hundreds of times.

Why is it hard? Each instrument comes from a different maker and saves its results in its own format. Before a model can direct them, every instrument and system must be connected, so that data moves between them without a person copying it by hand. That connection, called lab instrument integration, is the first step. Self-driving labs already exist in research, and instrument makers are now building the connections that would bring them into industry.

Take, for example, a drug discovery team. Its AI drug discovery model proposes new molecules, a self-driving lab makes and tests a few hundred of them, and the results go back to the model to improve the next design.

Consider a company that develops a new paint or plastic. With AI for chemistry and materials R&D, the model suggests the next recipe to try, and the lab tests it overnight.

Imagine a biology lab that tests thousands of cell samples. Liquid handling robots prepare the plates for each experiment that the model chooses.

How is a self-driving lab different from ordinary lab automation? An automated lab repeats a fixed method that a scientist wrote, such as the same test on every sample. A self-driving lab decides what to do next from the last result. It is one form of physical AI, in which an AI model directs machines in the real world, rather than only writing text.