Scientists are teaching AI-powered robots to run lab experiments

Reshma Shetty, co-founder and COO of Ginkgo Bioworks, walks through an autonomous lab where AI robots replace lab benches. Shetty says using AI has already fundamentally changed the way she practices science. "The really wild moment was the first time I saw a lab notebook entry written by the model," she says. Jodi Hilton for NPR hide caption

Nearly two decades ago, four graduate students from MIT united around a shared idea. "We believed that programming cells would ultimately be more important than programming computers," says Jason Kelly.

Want the latest stories on the science of healthy living? Subscribe to NPR's Health newsletter .

It felt like an outlandish bet at the time. Things like gene editing or testing new molecules typically demanded many hours in the laboratory — carefully mixing hundreds of chemical cocktails by hand and pipetting them into petri dishes, tasks that required an enormous amount of human labor.

Living Better

ChatGPT might give you bad medical advice, studies warn

The first step, they figured, was to speed that process up. So they started a company to replace those human lab workers with robots.

Early potential investors, Kelly recalls, were not excited.

" We were living on ramen, buying equipment on eBay, and we could not raise venture capital," he says of their early days running their startup.

Technology

"The godfather of AI" warns of AI possibly outperforming humans

Then came the artificial intelligence boom. In 2014, Kelly remembers reading a blog post from Sam Altman, roughly a year before he went on to found OpenAI. Kelly recalls that Atlman wrote about the potential to automate biotechnology the same way he imagined automating other kinds of technology. The two started talking.

" I was like, man, thanks for this blog post," Kelly recalls. "We've been around for five years. It is impossible to raise money."

Living Better

'ChatGPT saved my life.' How patients, and doctors, are using AI to make a diagnosis

Eventually, the Silicon Valley money started flowing.

Shetty, left, and Jason Kelly are two of the four co-founders of Ginkgo Bioworks. The group met at MIT, where they hatched an idea to build an automated biotechnology lab. Kelly says it wasn't a popular idea before the AI revolution. "We were living on ramen." Jodi Hilton for NPR hide caption

Today Kelly runs a company, Ginkgo Bioworks, with his former classmates. It has an autonomous laboratory housed in a building overlooking the Boston harbor. Using robotics and AI, Kelly and his co-founders say that they are building the science labs of the future — where human scientists oversee robotic versions of themselves.

Pipetting robots

" Pipetting robots," Kelly says while giving a tour. "I'll show you where we do that."

Science

Philanthropy is a 'significant form of power.' Here's how Jeffrey Epstein exploited that

Robots are arrayed around the lab, each working on separate science projects. They look nothing like humans — more like one-armed machines, each encased in glass like museum displays. A big screen at the front of the room shows a color-coded schedule of the experiments and each robot's tasks for the day. Below it a track resembling an oversized toy train set runs through the room, delivering equipment from one robot to another.

Gingko Bioworks does all kinds of work here including pharmaceutical, agricultural and government contracts. Current projects include engineering microbes for better fertilizer and creating proteins that will make snow or ice. They do a significant amount of research on pharmaceuticals.

The autonomous lab works on a range of pharmaceutical , agricultural, government and other projects. One current assignment includes engineering microbes for better fertilizer and creating proteins that will make snow or ice. Jodi Hilton for NPR hide caption

"That one there," says Kelly, gesturing to a petri dish being ferried from one robot to another, "that has actual live cells in it."

To do this work, scientists use AI to translate experimental designs into instructions for robots about the work they need done in the lab.

Empowering the robots to be the scientists

Recently Gingko's scientists have been experimenting with taking things a step further – empowering the robot to be the scientist.

"The really, really wild moment was the first time I saw a lab notebook entry written by the model," says Reshma Shetty, another of the founders.

Shetty recently worked on a collaboration with OpenAI. Working through ChatGPT, they challenged the bot to create a certain protein. Typically, this level of thinking is left to the scientists, not unlike writing a recipe and handing it to a robot to execute. Now they were asking the bot to write the recipe for them.

"We had no idea if it would even be able to make protein," says Shetty.

The bot performed better than they expected. In comparison to human work, they concluded the protein synthesis was a 40 percent reduction in costs. It ran more than 30,000 experiments in 6 months. They've published these results , though the paper has not been peer reviewed.

Both Shetty and Kelly stress that humans are still needed to provide the right questions and constraints for experiments. Still, Shetty says it has already fundamentally changed the way she practices science.

Most of the robots in this lab do not resemble humans, and they perform their work while encased in glass. Jodi Hilton for NPR hide caption

"Normally, I rush through designing my experiment because I need to get it done so that I can actually do all the pipetting in the lab and set it all up," says Shetty.

Now, she says, she spends more time designing her experiments so that the robot can do them for her overnight.

New access to science comes with risks

Some people warn these new freedoms bring new dangers. Drew Endy, who studies bioengineering at Stanford, says that artificial intelligence opens the door to the possibility of people with little to no training in science running experiments with questionable goals.

He and some colleagues recently wrote a report illustrating the ways artificial intelligence could be used to do things like mass-produce viruses or create other biosecurity threats.

In general, says Endy, "I'm thrilled about AI and science right now as a researcher," but he is also worried about risks including potential bioweapons programs in other countries. "I'm not excited about that."

He notes that regulations and policy to mitigate these risks are within human reach, but need to be prioritized well in advance of a biotechnological disaster or warfare.

Until now, says Endy, biotechnology has been naturally insulated from these risks through intellectual gatekeeping. " Biology has traditionally been hard for people to really gain control over," he says. "AI could nudge it a little bit more towards concentration of power."

For better or worse, Jason Kelly says he foresees a day when the practice of science is democratized.

"I do think you'll have a culture clash," says Kelly, "coming of what happens when everyday people can ask scientific questions."