In Search for Intelligence, a Silicon Brain Twitches

By GAUTAM NAIK

For the last four years, Henry Markram has been building a biologically accurate artificial brain. Powered by a supercomputer, his software model closely mimics the activity of a vital section of a rat's gray matter.

Dubbed Blue Brain, the simulation shows some strange behavior. The artificial "cells" respond to stimuli and suddenly pulse and flash in spooky unison, a pattern that isn't programmed but emerges spontaneously.

"It's the neuronal equivalent of a Mexican wave," says Dr. Markram, referring to what happens when successive clusters of stadium spectators briefly stand and raise their arms, creating a ripple effect. Such synchronized behavior is common in flesh-and-blood brains, where it's believed to be a basic step necessary for decision making. But when it arises in an artificial system, it's more surprising.

Photos: Artificial Intelligence

See images of artificial neurons in Blue Brain, a biologically accurate artificial brain powered by a supercomputer.

View Slideshow

Blue Brain Project

• More photos and interactive graphics

 

Scientists Create Artificial Brain

3:41

Meet Blue Brain, a "brain" made up entirely of silicon and housed inside an IBM supercomputer. An astonishing advance, the artificial brain may be the first step toward manmade higher behavior, WSJ's Gautam Naik reports.

Blue Brain is based at the École Polytechnique Fédérale de Lausanne in Switzerland. The project hopes to tackle one of the most perplexing mysteries of neuroscience: How does human intelligence emerge? The Blue Brain scientists hope their computer model can shed light on the puzzle, and possibly even replicate intelligence in some way.

"We're building the brain from the bottom up, but in silicon," says Dr. Markram, the leader of Blue Brain, which is powered by a supercomputer provided by International Business Machines Corp. "We want to understand how the brain learns, how it perceives things, how intelligence emerges."

Blue Brain is controversial, and its success is far from assured. Christof Koch of the California Institute of Technology, a scientist who studies consciousness, says the Swiss project provides vital data about how part of the brain works. But he says that Dr. Markram's approach is still missing algorithms, the biological programming that yields higher-level functions.

"You need to have a theory about how a particular circuit in the brain" can trigger complex, higher-order properties, Dr. Koch argues. "You can't assemble ever larger data fields and shake it and say, 'Ah, that's how consciousness emerges.'"

Despite the challenges, the push to understand, replicate and even re-enact higher behaviors in the brain has become one of the hottest areas of neuroscience. With the help of a $4.9 million grant from the U.S. Department of Defense, IBM is working on a separate project with five U.S. universities to build a tiny, low-power microchip that simulates the behavior of one million neurons and ten billion synapses. The goal, says IBM, is to develop brainy computers that can better predict the behavior of complex systems, such as weather or the financial markets.

The Chinese government has provided about $1.5 million to a team at Xiamen University to create artificial-brain robots with microcircuits that evolve, learn and adapt to real-world situations. Similarly, Jeff Krichmar and colleagues at the University of California, Irvine, Calif., have built an artificial-brain robot that learns to sharpen its visual perception when moving around in a lab environment, another form of emergent behavior, a form of spontaneous self-organization. And researchers at Sensopac, a project backed by a grant of €6.7 million ($9.3 million) from the European Union, have built part of an artificial mouse brain.

The scientists behind Blue Brain hope to have a virtual human brain functioning in ten years -- a lengthy time period that underscores the scientific challenge. The human brain has 100 billion neurons that send electrical signals to each other via a network of at least 100 trillion connections, or synapses. How could this dizzying complexity ever be recreated in a virtual model?

Dr. Markram has adopted a systematic, if painstaking approach. He decided to work out the blueprint of its wiring and then use that map to rebuild the brain in an artificial form. He focused on a rat's neocortical column, or NCC, an elementary building block of the brain's neocortex, which is responsible for higher functions and thought. In a rat's case, that includes planning to obtain food.

A rat's NCC, comprised of about 10,000 neurons and their 10 million connections, functions much like a computer microprocessor. All mammals have NCCs, and the ones in humans aren't all that different from the ones in rats. However, humans have far more NCCs, which means far greater brain power. Dr. Markram figured that if a rat simulation did a good job of correctly mimicking activity in a real rat's brain, he could use the same model as a road map for simulating the human brain.

Dr. Markram began by collecting detailed information about the rat's NCC, down to the level of genes, proteins, molecules and the electrical signals that connect one neuron to another. These complex relationships were then turned into millions of equations, written in software. He then recorded real-world data -- the strength and path of each electrical signal -- directly from rat brains to test the accuracy of the software.

At the Lausanne lab one recent afternoon, a pink sliver of rat brain sat in a beaker containing a colorless liquid. The neurons in the brain slice were still alive and actively communicating with each other. Nearby, a modified microscope recorded some of this inner activity in another brain slice. "We're intercepting the electro-chemical messages" in the cells, then testing the software against it for accuracy, said Dr. Markram.

The rat's NCC has 10,000 neurons, and it takes the power of one desktop computer to mimic the behavior of a single neuron. To model the entire NCC, Dr. Markram relies on an IBM computer that can perform 22.8 trillion operations a second. This enables the simulation to be rendered as a three-dimensional object. Thus, when Blue Brain is running, its deepest inner workings are seen in astonishing detail, in the form of a 3-D simulation that unfolds on a computer screen.

In a darkened room, Blue Brain displays a virtual NCC as a column-like structure, its blue color signifying a state of rest. When zapped by a simulated electrical current, the neurons start to signal to each other and their wiring progressively sparks to life different colors. Tests indicate the same areas light up in the model as do in a real rat's brain, suggesting that Blue Brain is accurate, says Dr. Markram.

More complex things start to happen. First there's a burst of red, then white, then red again, as the NCC's wiring fills up with a cascade of myriad signals. There are so many connections, the NCC looks like an incredibly dense tangle of undergrowth.

Then, two successive waves of yellow color suddenly race through Blue Brain. It's a sign that the neurons have synchronized their behavior on their own. "The cells start to take on a life of their own," says Dr. Markram. "That's what your brain is [and when such patterns become sophisticated] it becomes your personality."

If Blue Brain ever gets sophisticated enough to closely mimic the human brain, will it exhibit consciousness? Says Dr. Markram: "If it does emerge, we'll be able to tell you how it emerged. If it doesn't, we'll know that it's the result of more than just 100 billion neurons interacting."

Write to Gautam Naik at gautam.naik@wsj.com

Corrections & Amplifications:
There are 10 million neuronal connections in the neocortical column of a rat brain. A previous version of this article incorrectly said there were 10 billion such connections.