Isaac Asimov wrote of the human brain: “It’s the most complicated organization of matter that we know.”
Starting with the trailblazing anatomical drawings of the human brain by Leonardo da Vinci in 15th century Renaissance Europe—who, in a sense, can be said to be the world’s first neuroscientist—researchers have been able to understand the workings of the “three-pound universe” (a phrase coined by Judith Hooper) at many levels.
But attempts to replicate its unfathomably complex circuitry had never been made until a few years ago.
One effort, known as the Blue Brain Project—a tie-up between I.B.M. and the Brain Mind Institute at the École Polytechnique Fédérale de Lausanne in Switzerland—tries to map the brain’s biological wiring down to its most granular level.
Vast quantities of data recorded by neurologists are fed into the I.B.M. Blue Gene supercomputer (after which the project takes its name) to recreate a highly realistic simulation of the brain’s functioning. They began by creating a digital replica, of sorts, of a tiny section of a small brain: the neocortical column of a rat brain, a 0.2 by 0.5 millimeter mini-cylinder, which in essence, is the building block of every mammalian brain.
The cerebral cortex, the “gray matter,” which makes up 80 percent of the brain, is nothing but a multiple replication of these little structures.
Simulating the real-time activities of the human brain with the aid of software would require a supercomputer with a memory of 3.2 petabytes (10 followed by 15 zeros), a magnitude that the industry isn’t expected to reach in the next three years.
But a European team is already taking this to the level of hardware by redesigning the brain’s cellular structure on a thin chip. To hardwire an inconceivably crowded circuit of 10 billion neurons and 220 trillion synapses—that would make the density of the Amazon rainforest appear to be child’s play—on a miniscule scale would be an incredible engineering feat.
To put that in perspective: each neuron is connected to the others through a forest of 8,000 synapses, and it takes about 20 transistors to reproduce just one synapse. But so far, 200,000 neurons, connected by 50 million synaptic junctions, have been fitted onto a standard, eight-inch silicon wafer. Even though that’s only a small fraction of the neural network, it’s, without doubt, a big milestone.
More than any other machine in existence today, this chip comes closest to mimicking the brain’s ability to learn. It’s hoped that this will provide valuable insight in developing a new generation of “massively parallel, powerful computers.”
In theory, computers run much faster than neurons, but the reason they appear to be slower is that they’re serial machines, which cannot do many different things simultaneously.