Skip to Content
Artificial intelligence

A robot scientist will dream up new materials to advance computing and fight pollution

Kebotix is using AI and robotics to brainstorm—and then test—novel compounds.
November 7, 2018
Ms. Tech

In a laboratory that overlooks a busy shopping street in Cambridge, Massachusetts, a robot is attempting to create new materials.

A robot arm dips a pipette into a dish and transfers a tiny amount of bright liquid into one of many receptacles sitting in front of another machine. When all the samples are ready, the second machine tests their optical properties, and the results are fed to a computer that controls the arm. Software analyzes the results of these experiments, formulates a few hypotheses, and then starts the process over again. Humans are barely required.

The setup, developed by a startup called Kebotix, hints at how machine learning and robotic automation may be poised to revolutionize materials science in coming years. The company believes it may find new compounds that could, among other things, absorb pollution, combat drug-resistant fungal infections, and serve as more efficient optoelectronic components. The company’s software learns from 3-D models of molecules with known properties.

Software algorithms are already used to design chemical compounds and materials, but the process is slow and crude. Usually, a machine simply tests slight variations of a material, blindly searching for a viable new creation. Machine learning and robotics could make the process much faster and more effective. Kebotix is one of several startups working on this idea.

The goal is to use machine learning to generate candidate materials. “Discovery is too slow,” says Jill Becker, CEO of Kebotix. “You have an idea for a material, you try to make it, and you test it. Few ideas are tested, with even fewer results.”

The founders of Kebotix: Alán Aspuru-Guzik, Dennis Sheberla, Jill Becker, Semion Saikin, and Christoph Kreisbeck.
Courtesy of Kebotix

Kebotix uses several machine-learning methods to design novel chemical compounds. The company feeds molecular models of compounds with desirable properties into a type of neural network that learns a statistical representation of those properties. This algorithm can then come up with new examples that fit the same model.

Kebotix also uses another network to weed out designs that stray too far from the original and are therefore likely to be useless. Then the company’s robotic system tests the remaining chemical structures. The results of those experiments can be fed back into the machine-learning pipeline, helping it get closer to the desired chemical properties. The company dubs the overall system a “self-driving lab.”

Christoph Kreisbeck, the company’s chief product officer, says Kebotix will start out working with molecules for electronic applications and then try to tackle new polymers and alloys.

“The AI predicts and plans what to do next; the robot automation system very rapidly tests our new molecule,” Kreisbeck says. “The machine can learn from the database and make a better decision for the next round.”

Kebotix was founded by researchers working in the Harvard lab of Alán Aspuru-Guzik, who left Harvard earlier this year to build at lab at the University of Toronto in Canada. Kebotix, which is based at MIT’s VC firm The Engine, recently received $5 million in seed funding. The investment round was led by One Way Ventures, an investment firm that specializes in funding immigrant entrepreneurs. All of Kebotix’s founding team members are immigrants to the US.

Klavs Jensen, a professor in MIT's chemical engineering department, leads a lab that’s developing automated approaches to devising useful new chemicals, including methods that combine machine learning and robotics. He says the catch is that such methods tend to require huge quantities of data, which is generally time consuming and difficult to collect. This also becomes more challenging as the materials get more complicated. “You can definitely do a lot,” Jensen says. “But like anything else, it’s about the quality of the data.”

Jensen says that automation, already commonplace in the pharmaceutical industry, will become increasingly important in materials research. “It won’t replace the expert,” he says, “but you’ll be able to do things a lot faster.”

Deep Dive

Artificial intelligence

Sam Altman says helpful agents are poised to become AI’s killer function

Open AI’s CEO says we won’t need new hardware or lots more training data to get there.

Is robotics about to have its own ChatGPT moment?

Researchers are using generative AI and other techniques to teach robots new skills—including tasks they could perform in homes.

What’s next for generative video

OpenAI's Sora has raised the bar for AI moviemaking. Here are four things to bear in mind as we wrap our heads around what's coming.

An AI startup made a hyperrealistic deepfake of me that’s so good it’s scary

Synthesia's new technology is impressive but raises big questions about a world where we increasingly can’t tell what’s real.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at with a list of newsletters you’d like to receive.