Robots made from living cells learned how to replicate themselves

Xenobots are "living robots" made up of frog cells, and can be programmed, in a way, by having a supercomputer design how their cells are arranged.

Now, researchers have discovered that they've taught themselves to replicate.

"This is one of those rare moments when you remember why you were a scientist in the first place," said Joshua Bongard, a computer scientist at the University of Vermont who co-led the research.

In a new study, Bongard and his colleagues from Tufts University and Harvard's Wyss Institute for Biologically Inspired Engineering found that the xenobots would autonomously collect loose single cells in their environment, gathering hundreds of cells together until new xenobots had formed.

"This took a little bit for us to wrap our minds around," he said. "There's no programming here. Instead, we're designing or shaping these xenobots, and what they do, the way they behave, is based on shape."

AI designed 'Pac-Man' parents

Xenobots are made up of thousands of skin cells from the Xenopus laevis frog.

"We take a couple of thousand of those frog cells and we squish them together into a ball and put that in the bottom of a petri dish," Bongard told Quirks & Quarks host Bob McDonald. 

"If you were to look into the dish, you would see some very small, what look like specks of pepper, moving about in the bottom of the petri dish."

The xenobots initially received no instruction from humans on how to replicate. But when researchers added extra cells to the dish containing xenobots, they observed that the xenobots would assemble them into piles.

"Cells early in development are sticky," said Bongard. "If the pile is large enough and the cells stick together, the outer ones on the surface will grow very small hairs, which are called cilia. And eventually, after four days, those cilia will start to beat back and forth like flexible oars, and the pile will start moving."

"And that's a child xenobot." 

However, those new xenobots were smaller and weaker, and unable to reproduce themselves. So the researchers turned to artificial intelligence (AI) to test body shapes, which would make the xenobots more effective at reproducing. The computer ultimately ended up on a shape that allowed the bots to gather more cells together, which ultimately produced larger, more robust children.

"It tried several billion designs, several billion shapes until it came up with a particular shape," he said. "It was a flattened disk and the AI had cut a wedge out of the disk. And in essence, what we had was a 'Pac-Man' shape."

"When we built those designs and placed them in the dish, those nine parents produce children, which produced grandchildren, great-grandchildren, great-great-grandchildren, before the process ran out of gas. So in essence, the AI had increased the replicative ability in this robotic technology," said Bongard.

Understanding exponential growth, and how to squash it

While science fiction films like The Terminator and The Matrix warn against encouraging the self-replication of robotics, Bongard believes that this is exactly why xenobots are an important research tool.

"All of us are somewhat apprehensive of very small biological things that replicate out of control. That's something we're grappling with at the moment," he said, referring to the current exponential growth of COVID-19. 

"We want to use the xenobots as a new scientific tool to understand the conditions under which living systems replicate and grow exponentially. And we also want to try and develop technologies — like the AI I mentioned — that can step in and amp up the replication of good things and step in and snuff out the replication of bad things."

Produced and written by Amanda Buckiewicz.