“We’re not trying to replicate the brain. That’s impossible. We don’t know how the brain works, really,” says the chief of IBM's Cognitive Computing project, which aims to improve computing by creating brain-like computers capable of learning in real-time and consuming less power than conventional machines. No one knows how the brain works, but have the folks at IBM tried to figure it out? It seems strange to say that it's impossible to replicate the brain, especially coming from a man whose blog's caption reads, "to engineer the mind by reverse engineering the brain." Perhaps I'm picking at his words - replicating and reverse engineering are totally different things; to replicate is to copy exactly, while reverse engineering isn't as strict, since it's concerned with macroscopic function rather than microscopic structure. But of all the things that seem conceptually impossible today, it's the "engineer the mind" that's the winner, especially if one can't "replicate the brain." The chances of engineering a mind are greater the closer the system is to the brain; that's why my MacBook, to my continual disappointment, does not have a mind.
These little trifles haven't stopped Darpa from funding IBM and scientists elsewhere. IBM now boasts a prototype chip with 256 super-simplified integrate-and-fire "neurons" and a thousand times as many "synapses." This architecture is capable of learning to recognize hand-drawn single-digit numbers. Its performance may not be optimal, but still impressive considering the brain likely allocates far more neurons (and far more complicated neurons) to the same task. On another front, the group reported using a 147,456-CPU supercomputer with 144TB of main memory to simulate a billion neurons with ten thousand as many synapses. Now if only they could combine these two efforts and expand their chip from two hundred to a billion neurons.