Scientist Build a 'Brain' From Rat Cells

The dish is on top of an array of 60 electrodes that allows the scientists to record the neural activity. And from that they can train the neurons to control the aircraft through a flight simulator.

"Here's the trick to it," DeMarse says. "We have these recordings coming out of the dish, and that allows us to get the behavior [of the neurons] into the simulated plane."

But how do the neurons learn how to fly the thing? That's done by electrical pulses into the dish through one of the electrodes. That in effect tells the neurons when they are doing the right thing to keep the plane on course. High frequency, or rapid pulses, stimulate the neurons and enhance the connections between them.

Simply put, by stimulating the neurons the researchers tell them they're on the right track, so they continue to adjust the plane's elevator to keep it from plunging toward the ground during a downdraft, for example. When the plane levels off, the simulator reduces the frequency of the pulses, and the neurons back off from that control surface, allowing the plane to remain on course.

After just a few minutes of that kind of training, the "brain" takes over completely, sending signals to the plane's control surfaces, and using feedback from the simulator to know just which signals to send.

And then, after about 15 minutes, it's all over. The neurons can't remember how to fly the plane anymore, so the next time the experiment is run, the neurons have to be taught all over again.

Melding Biology and Computers

It all sounds very complex, but DeMarse says it's actually a very simple experiment, and he hopes to change that. The experiment currently uses only two electrodes to send and receive signals, and the brain of a mammal uses many avenues, or sensors, to do what it has to do to survive.

So the "brain in a dish" is actually a very simplified brain, even for a rat. But if that's the case, how come we can't teach a rat how to fly an aircraft?

Maybe we could, DeMarse says, if we knew how. But there would still be a bit of a problem.

"Their visual system isn't that good," he says. "The rat wouldn't be able to see past the cockpit."

DeMarse doesn't think the purely biological model he's working with is likely to ever pose a threat to human dominance. Even the relatively simple "brain in a dish" that can fly a simulator is difficult to maintain.

But still, the neural network he is exploring, along with many other scientists, is setting the stage for the creation of hybrid computers that are based largely on biological systems and could potentially dwarf the most powerful supercomputers today. And if they get too smart, as Utah State's deGaris notes, we might want to be sure they learn to love us.

Lee Dye's column appears weekly on A former science writer for the "Los Angeles Times," he now lives in Juneau, Alaska.

-- This embed didnt make it to copy for story id = 198839.
  • 1
  • |
  • 2
Join the Discussion
blog comments powered by Disqus
You Might Also Like...