The Banana Peel Project

Computational neuroscientists have used IBM’s Blue Gene L supercomputer to model an entire “biologically accurate” mammalian neocortical column–that’s 10,000 neurons and 3,000,000 connections.

Models like these can be used to test all kinds of things, like drugs and therapies. What happens when scientists run experiments not on real brains or real neuronal structures, but on models of them built from the bottom up? Even if the model “works,” what are we really dealing with here? When do we know the simulation is “good enough?”

One researcher makes a great point: these models are pretty good at the cellular level, but what happens at the molecular level? As any good neuroscientist will tell you, neurons are not black boxes. Lots of information is conveyed directly from axon to axon for example, at a level much smaller than would be captured by modeling axonal connections alone.

But does that mean that the model is wrong, or doesn’t work? I’m still wondering.

[via information aesthetics]