Rat Brain Robots Attack
Think about what has to happen. They have figured out how to take rat brain cells and disconnect them from rats body, yet keep them alive. For extended periods of time too. DeMarse's paper A New Approach to Neural Cell Culture for Long-Term Studies (PDF) says that they have kept brains cells alive in jars for over a year. How do you feed it? The paper doesn't say, mostly it details how to keep it clean.
The second step is figuring out how to interpret the signals coming from the brain and how to provide it with proprioceptive feedback. This part is explained in the paper The Neurally Controlled Animat: Biological Brains Acting With Simulated (PDF.)
The most interesting part is also the part I can find the least information about. Once you have this rat brain working and you are interpreting its signals then you have to figure out how to train it to do what you want. The brain isn't going to know that up is up and down is down and flying up is better than flying down. It's just a brain, no sensors like eyes or ears or equilibrium. It has no prori knowledge like crashing is bad or flying is better right side up. These things have to be represented in the signals you send back to the brain. How do you spank a brain cell to tell it that it did a bad job. How do you give it a treat to say, "Nice work?" I sent DeMarse an email asking these questions, but as of yet, no response. Maybe his rat brain will be able to drop me a line next year.
Being an AI games programmer this is so totally amazing to me. My job, what I try to do everyday, is inherently coded into a blob of goo at the center of DeMarse's experiment. We're starting to be able to interact with that goo, figure out what it does, how it works, and then design machines to do the same thing. We don't want to go shipping blobs of biological goo with our video game systems. Rat memory cards don't fit into my Playstation.
Soon, once again, what seems like magic will be explained.
No comments:
Post a Comment