disembodied rat neurons are thinking about flying an F-22
Posted: Mon Oct 25, 2004 6:15 am
Ran across this, thought it was way cool, and so I am sharing. Does this raise ethical, "terminator" and "Skynet" types of questions?
By Lakshmi Sandhana
02:00 AM Oct. 23, 2004 PT
Somewhere in Florida, 25,000 disembodied rat neurons are thinking about flying an F-22.
These neurons are growing on top of a multi-electrode array and form a living "brain" that's hooked up to a flight simulator on a desktop computer. When information on the simulated aircraft's horizontal and vertical movements are fed into the brain by stimulating the electrodes, the neurons fire away in patterns that are then used to control its "body" -- the simulated aircraft.
"It's as if the neurons control the stick in the aircraft, they can move it back and forth and left and right," said Thomas DeMarse, a professor of biomedical engineering at the University of Florida who has been working on the project for more than a year. "The electrodes allow us to record the activity from the neurons and stimulate them so we can listen to the conversation among the neurons and also input information back into the neural network."
Currently the brain has learned enough to be able to control the pitch and roll of the simulated F-22 fighter jet in weather conditions ranging from blue skies to hurricane-force winds. Initially the aircraft drifted, because the brain hadn't figured out how to control its "body," but over time the neurons learned to stabilize the aircraft to a straight, level flight.
"Right now the process it's learning is very simplistic," said DeMarse. "It's basically making a decision about whether to move the stick to the left or to the right or forwards and backwards and it learns how much to push the stick depending upon how badly the aircraft is flying."
The seed idea for DeMarse's autopilot came out of earlier work with Steve Potter on the Animat project, where researchers used living rat neurons to control an animated object in a virtual world. They also connected the neurons to a robot and tried to teach the brain to track and approach objects.
The bigger goal is to figure out how neurons talk to each other. MRI scans, for example, show millions of neurons firing together. At that resolution, it is impossible to see what's happening between individual neurons. While scientists can study neural activities from groups of cells in a dish, they can't watch them learn and grow as they would within a living body unless the neurons have some kind of body to interact with.
By taking these cells and giving them back a "body," the researchers hope to uncover how the neurons communicate with each other and eventually translate that knowledge to develop novel computing architecture.
"Granted, this is just a handful of neurons in a dish," said Potter, an assistant professor at Georgia Tech's neuroengineering laboratory. "It isn't a full-blown brain. It doesn't have a real body. But with this kind of system you can literally watch these things compute and you have a chance to learn how the brain does its computation."
DeMarse plans to make the autopilot more competent by having the brain use a horizon to judge how it controls the plane. But the true breakthrough will come about when the researchers detect how neurons communicate in a network.
"We know some of the rudimentary rules," said DeMarse. "We just don't quite understand the language that they use to do their computations. We can extract the general features from it to control the aircraft but there's a lot more information buried in the signals that they are using, and we simply don't know what that is. So there's a lot more to do in terms of understanding the language of the network."
By Lakshmi Sandhana
02:00 AM Oct. 23, 2004 PT
Somewhere in Florida, 25,000 disembodied rat neurons are thinking about flying an F-22.
These neurons are growing on top of a multi-electrode array and form a living "brain" that's hooked up to a flight simulator on a desktop computer. When information on the simulated aircraft's horizontal and vertical movements are fed into the brain by stimulating the electrodes, the neurons fire away in patterns that are then used to control its "body" -- the simulated aircraft.
"It's as if the neurons control the stick in the aircraft, they can move it back and forth and left and right," said Thomas DeMarse, a professor of biomedical engineering at the University of Florida who has been working on the project for more than a year. "The electrodes allow us to record the activity from the neurons and stimulate them so we can listen to the conversation among the neurons and also input information back into the neural network."
Currently the brain has learned enough to be able to control the pitch and roll of the simulated F-22 fighter jet in weather conditions ranging from blue skies to hurricane-force winds. Initially the aircraft drifted, because the brain hadn't figured out how to control its "body," but over time the neurons learned to stabilize the aircraft to a straight, level flight.
"Right now the process it's learning is very simplistic," said DeMarse. "It's basically making a decision about whether to move the stick to the left or to the right or forwards and backwards and it learns how much to push the stick depending upon how badly the aircraft is flying."
The seed idea for DeMarse's autopilot came out of earlier work with Steve Potter on the Animat project, where researchers used living rat neurons to control an animated object in a virtual world. They also connected the neurons to a robot and tried to teach the brain to track and approach objects.
The bigger goal is to figure out how neurons talk to each other. MRI scans, for example, show millions of neurons firing together. At that resolution, it is impossible to see what's happening between individual neurons. While scientists can study neural activities from groups of cells in a dish, they can't watch them learn and grow as they would within a living body unless the neurons have some kind of body to interact with.
By taking these cells and giving them back a "body," the researchers hope to uncover how the neurons communicate with each other and eventually translate that knowledge to develop novel computing architecture.
"Granted, this is just a handful of neurons in a dish," said Potter, an assistant professor at Georgia Tech's neuroengineering laboratory. "It isn't a full-blown brain. It doesn't have a real body. But with this kind of system you can literally watch these things compute and you have a chance to learn how the brain does its computation."
DeMarse plans to make the autopilot more competent by having the brain use a horizon to judge how it controls the plane. But the true breakthrough will come about when the researchers detect how neurons communicate in a network.
"We know some of the rudimentary rules," said DeMarse. "We just don't quite understand the language that they use to do their computations. We can extract the general features from it to control the aircraft but there's a lot more information buried in the signals that they are using, and we simply don't know what that is. So there's a lot more to do in terms of understanding the language of the network."
Pay attention