The map is not the territory, but this one folds up the same way.

Apr 30, 2007

Researchers from the University of Nevada and IBM's Almaden Research Lab have used the BlueGene L supercomputer to run a heretofore unprecedented simulation of about one-half of a mouse's brain. It's not easy to keep an organic brain going outside of a living body so they did the next best thing, which was write a program that emulates the organic brain as closely as they could. This isn't as easy as it sounds because neural networks more advanced than those of worms have so many interacting factors that taking them all into account is a gargantuan task. It is also all of those factors that, when you stack them up, go into making brains what they are. The simulation involved emulating about eight million neurons, each of which had up to six thousand, three hundred synapses, or connections with other neurons. Even after scaling it down from the theoretical maximum of eight thousand possible synapses per program-neuron, they were only able to run it for ten wallclock seconds, and from the simulation's point of view (I use the phrase metaphorically) it ran at one-tenth the calculated speed of a real mouse's right hemisphere.

During a number of runs of the simulation, the researchers discovered some amazing things: The simulated neurons spontaneously organized themselves into operational clusters which exhibited much the same behaviour as clusters of real neurons that were connected. The patterns of communication between the simulated neurons also mimicked the patterns of electrical impulses found in real mouse brains. In other words, the simulation was complex enough to exhibit emergent behaviour within medically expected parameters.

Their next step is to optimize the simulation such that it'll run closer to the speed of a real mouse brain. Then, possibly, they could start interacting with it on some level.