Interesting article, however I would say that five lines of C++ code per neuron is pretty optimistic considering the amount of work going into modeling a neuron in the blue brain project; http://seedmagazine.com/designseries/henry-markram.html
Also I remember reading in Code Complete that the space-shuttle software had zero defects per 500,000 lines of code by combining formal development methods, peer reviews, and statistical testing. Not really sured how they 'proved' there were zero defects, but I'm sure you could get the number far lower than 1 per 100 lines.
The basic premise of writing a model for the entire human brain in software is bunk. The general idea, as I understand it, is to write a piece of software that simulates the biochemical processes that happen in the brain, with the neurons themselves being represented by some sort of graph.
Basically, the 'brain-in-a-jar', except the jar is virtualized.
I somehow managed to say the exact opposite of what I wanted.
What I meant to say is that physically coding every bit of what every neuron does in software makes zero sense. As in, having X lines of code for neuron one, X for neuron two, etc. Which is what the original article seems to apply with the 'lines of C++ code' estimation -- that you would have to write unique code for every neuron.
What makes sense to me, is to build a software model of a brain, and let that grow in the same way brains grow naturally, just in a virtualized environment.
Also I remember reading in Code Complete that the space-shuttle software had zero defects per 500,000 lines of code by combining formal development methods, peer reviews, and statistical testing. Not really sured how they 'proved' there were zero defects, but I'm sure you could get the number far lower than 1 per 100 lines.