Dawn of the Rat-Brained Robots

In the heady, optimistic days of 2004, it seemed all but certain that airline pilots were soon to be replaced by none other than lab-grown rat brains.

Okay, not exactly. But it's hard not to laugh over a decade later at a University of Florida press release announcing the dawn of "living computers." A team at the university had cultured some rat neurons, and the culture had managed to fly a virtual plane. A lab-grown rat brain flew a simulated plane!

"They may someday be used to fly small unmanned airplanes or handle tasks that are dangerous for humans, such as search-and-rescue missions or bomb damage assessments," the release continued, and our rat-brain filled future seemed nigh.

The 2000s saw a bevy of this sort of research into so-called animats—electronic systems that were controlled, not by traditional silicon chips, but organic, lab-grown neuronal networks, or rudimentary brains. The University of Reading, University of Florida, and Georgia Tech all had (and still have) animat teams. But then, and even now, real-world applications for this sort of research remain a very basic, and a long ways off.

Thomas DeMarse, a professor at the University of Florida, did indeed succeed at growing a culture of rat-neurons that could fly a simulated plane in 2004 (and by fly, I mean travel straight and level), building on earlier research at Georgia Tech. And in 2008, the University of Reading grew a similar culture that prevented a wheeled robot from running into walls. But to think of these cultures as "brains" is the way that you or I might think of brains is disingenuous at best.

"These are not sentient beings. I wouldn't want to equate it even to an ant," said Slawomir Nasuto, a professor of cybernetics at the University of Reading.

Nasuto was part of a team at the university that worked on an animat project from 2006 to 2009. Like other universities, they employed a process in which cells were seeded onto a dish lined with electrodes. As the cells grew and formed connections with one another, neural activity began to take place. The electrodes can read this activity, as well as deliver stimulus back to the cells—effectively creating an input and output mechanism for the lab-grown brain to process information about its environment when connected to something like, say, a rudimentary robot.

However, "I wouldn't call it learning in a way that probably most people would understand, even for such simple systems," Nasuto said. "The animats that have been constructed so far—and that includes all the groups that I'm aware of—are effectively constructed in a very mechanistic way. So I wouldn't like to say that they are understanding or anything."

According to Nasuto, animats are actually more akin to Braitenberg vehicles. First proposed by the neuroscientist Valentino Braitenberg, these were dumb mechanical vehicles that were designed in such a way that otherwise simple stimuli would cause the vehicles to respond in unexpectedly complex ways—so much so that their behaviour could be mistaken for actual intelligence. (For example, a Braitenberg machine fitted with two light sensors connected directly to each of a vehicle's wheels can be configured to essentially follow, or drive, in the direction of a light source, but entirely mechanically.)

Animats, said Nasuto, are essentially hardwired to work in similar ways. Though some attempts have been made at rudimentary conditioning—essentially, applying repeated stimuli until the neurons respond in a way that meets a researcher's needs—it wouldn't be fair to say these robots possessed the capacity to learn. Rather, they can be programmed to respond to their environment in a very primitive, simply-defined way.

That may not bring us any closer to rat-brain pilots and rescue robots, but it's an important step in understanding the thing researchers are actually trying to model—the mind.

Animats "[are] considered to be often a platform," Nasuto said. "It's an interesting platform in its own right, but it's also a platform that enables [researchers] to investigate other questions related to neuroscience and related to, say, finding out the mechanisms of brain information processing, both in health and in disease."

In other words, no one has set out with the sole intent to build a robot controlled by living neurons on lab-grown brains just yet—but research being done in other areas, as a by-product, could get us there one day.