Chemist Builds AI That Runs Lab Experiments in Plain English

A kid who didn't own a computer until age 19 just taught artificial intelligence how to run a chemistry lab.

Gabriel Gomes, a chemical engineer at Carnegie Mellon University, grew up in a small Brazilian town where technology felt distant. Today, he's created Coscientist, an AI agent that lets scientists run complex experiments using plain English instead of code.

The breakthrough solves a problem Gomes spotted when his university invested $50 million in cloud lab equipment. The catch? Scientists would need programming skills to use it. Many chemistry professors weren't interested in learning to code just to access state-of-the-art tools.

Then GPT-4 launched in March 2023. Gomes woke up at 6 a.m. with an idea: what if scientists could simply tell the lab what they wanted in normal conversation?

Coscientist works like a cooking assistant for experiments. Tell it "run this reaction" and it figures out the recipe, checks available equipment, and provides step-by-step instructions. Scientists can even share photos or videos for troubleshooting help.

The first test was simple but revealing. The team asked their robot to "draw something cute" on a plate using food coloring. Coscientist drew a fish without any specific programming instructions.

The real impact goes deeper than cute drawings. Gomes's team built one of the largest datasets of experimental chemical reactions ever assembled. Creating it manually would have required exhausting amounts of repetitive work that most researchers simply avoid.

The Ripple Effect

The technology is already changing lives in Gomes's lab. One student joined in 2024 worried he lacked programming skills. Within months, using Coscientist and learning from AI tools, he was running cutting-edge autonomous research projects.

Gomes spent 2023-2024 at the White House Office of Science and Technology Policy, advising on AI safety. He's clear-eyed about risks but emphasizes that scientists should treat these tools as assistants, not oracles. The AI can become too eager to please, so researchers must verify results instead of accepting them blindly.

The first-generation college student who once couldn't afford a computer now sees a future where AI eliminates the drudgery holding back scientific discovery. His advice to fellow scientists? Learn how these tools work, embrace them carefully, and use them to explore research areas previously blocked by human limitations.

Scientists can now spend less time on repetitive tasks and more time asking the big questions that matter.