tags: learning, rat brain, airplane, streaming video
This research shows how brains learn at the cellular level. By placing neurons from a rat into a cell culture that is contected to a flight simulator, the neurons learn how to fly an airplane. And they learn it by trial and error. [3:25].
- Log in to post comments
OMG - Astonishing!!!! WOW - GeeWiz, I'm speachless... Forget the robots and harness neurons to run anything man desires. My head is swimming with ideas. What an amazing piece of work. Could these neurons be the "brain" the robots need?! Hmmmmm,I will be thinking about this all day.
GREAT post,GrrlScientist,- Thanks!!
This is truly amazing! Thanks for sharing this with us. I wish the video went a little deeper into the science of this. Specifically, I wonder how they were able to convey negative actions to the neurons as opposed to positive ones? I mean, how do you convey to a bunch of neurons that flying a plane straight and steady is the right thing to do while crashing it into the ground is the wrong thing? Anyone have a link to more info on this experiment?
The experiment is actually about 3 years old - there is lots of info on it in Google
http://www.google.co.uk/search?hl=en&q=rat+neurons+fly+plane&btnG=Googl…
When I first saw this, about 3 years ago, it made me think of a future in which those desiring new skills quickly, would, rather than resorting to books or classes or practice, pay to have pre-trained rat neurons surgically implanted in their brains.
You can find a paper about it here; the authors trained the network by selecting two electrode sites, one to represent pitch and the other to represent roll. They then deliver high or low frequency stimulation, which they suggest dampens or increases activity in the network respectively, until the airplane is flying straight and level. Then they remove the high frequency stimulation which they say leads to "over corrections."
Based on the description in that paper, I do not understand how this method has done anything except get the network to act as a circuit with a particular amount of resistance. Essentially, the network is being trained to provide a specific value of pitch and roll, and that's it (if I understand correctly). No dogfighting, obstacle avoidance, etc (claims in the video notwithstanding).
This project also seems defunct now - he's working on other stuff.