Imagine an airplane piloted by a cluster of brain cells growing in a little glass dish. The scenario sounds unlikely, but in Thomas DeMarse's lab, the brain cells are already in flight school.
DeMarse, an assistant professor of biomedical engineering at the University of Florida, grows rat brain cells-more specifically, neurons-in a dish connected via electrodes to a flight simulation program on a computer. The neurons have been successfully trained to fly the plane straight.
But this research will not be replacing human airplane pilots any time soon; the main goal of this study is to decipher how neurons communicate with each other and how learning takes place in the brain.
\This is a case where the neural network starts out not being able to fly, but over time it adapts [so it can],"" DeMarse said.
Although it is tempting to say the neurons ""learned"" how to fly, DeMarse shied away from that word. ""Learning isn't the right word. I'd say it's more like adaptation,"" he said. ""[The brain cells] have no sense that they are connected up to anything. They don't know what they're doing.""
Rat neurons are grown in a special dish with a grid of 60 electrodes on the bottom. The neurons grow on and around the electrodes, interconnecting to form a neural network. The neurons communicate by emitting and receiving electrical pulses, forming a kind of electrical circuit.
The computer shocks the neurons through one or more electrodes, thereby transmitting information to the neurons about the flight path of the simulated airplane. The neurons respond to the ""input"" signal with a knee-jerk electrical signal of their own, which the electrodes also record. This ""output"" signal is translated into directions to adjust the flight path of the plane.
When neurons give the correct signal to fly the plane, they are rewarded with electrical stimuli that encourage the neurons to strengthen their connections. When the neurons give the wrong signal, perhaps by making a descending plane descend even more, they are punished with stimuli that compel the neurons to dissociate from one another. This feedback causes the neural network to rewire itself so it responds correctly to the input information, keeping the plane flying level.
Although it has been hyped in the media for a year, DeMarse's research has not yet been published for scientific scrutiny. DeMarse declined to name a publication date.
""It's hard to evaluate what they did, because all [the technical] details are left out,"" said Justin Williams, UW-Madison assistant professor of biomedical engineering and neurosurgery.
Yet the technology's potential applications are staggering, from a medical cure for epilepsy to unmanned military reconnaissance flight missions.
""We might be able to use it to rewire circuits in real brains that are malfunctioning,"" mused Matt Jones, a UW-Madison assistant professor of physiology. ""You can come up with incredibly useful fantasies about what you can do with [this technology], and nightmares.\
