Smart Takes

Scientists build biggest artificial brain of all time; 1.6 billion neurons, as smart as a cat

Posting in Technology

Scientists have built the biggest artificial brain of all time using a supercomputer powered by 147,456 processors, 150,000 gigabytes of memory and millions of watts of electricity that's about as smart as a house cat.

Scientists have built the biggest artificial brain of all time using a supercomputer powered by 147,456 processors, 150,000 gigabytes of memory and millions of watts of electricity that's about as smart as a house cat.

Scientists at IBM's Almaden research center created the brain simulation, which is said to run at one-eighty-third the speed of a human brain, to make advances in neuroscience and computer processing.

The artificial brain simulation emulates the human visual cortex, with 1.6 billion virtual neurons connected by 9 trillion synapses. That's an advance over the previous record, the same team's simulated rat brain with 55 million neurons from two years ago.

The feat was announced at the Supercomputing 2009 conference in Portland, Ore.

The human brain's visual cortex -- it's the wrinkly outer layer -- is responsible for handling visual motion, which includes speech and face recognition, as well as hand-eye coordination.

The researchers' simulation uses neuroscientific data from rats, cats, monkeys and humans in an attempt to better understand how the brain works and handles such distinct functions.

There's a quid pro quo, too: the experiment can also help build better computers that, with a little reverse engineering, could understand biological feedback such as sight, hearing and touch. Add the radar-and laser-sensing capabilities that a computer also has, and you've got one potentially intelligent machine on your hands.

"Imagine peppering the entire surface of the ocean with pressure, temperature, humidity, wave height and turbidity sensors," Modha said to Popular Mechanics. "Imagine streaming this data to a reverse-engineered cortex."

In other words, a truly "smart planet."

But the IBM simulation isn't yet capable of that, and can't develop neural patterning that a real brain can during its lifetime. For now, it's a brain without a body -- a crucial part of how the brain wires itself.

The simulation was run by one of the fastest supercomputers in the world: Dawn, a Blue Gene/P supercomputer at Lawrence Livermore National Laboratory in California.

Don't believe in nature's awesome power? Consider that it took Dawn's acre-size footprint packed with 147,456 processors, 147,000 gigabytes of memory, 10 rows of racks, miles of cable, a million watts of electricity and 6,675 tons of air-conditioning equipment spouting 2.7 million cubic feet of chilled air to simulate the biological wiring found in your 10 lb. pet cat.

And that's still one-eighty-third the amount of power your own brain can muster.

As you might suspect, the team's ultimate goal is to simulate the entire human cortex, which consists of about 25 billion neurons. That feat will require 1,000 times the computing power than the best supercomputers have right now.

Cognitive computing engineer Dharmendra Modha explains more on his blog:

We are also interested in using our technique to make measurements on the projectome and communication between brain areas that can generate hypothesis about brain function that may be validated with behavioral results or perhaps functional imaging and can be integrated with large-scale simulations.

With consideration to advances in supercomputing, that kind of power could be available by 2019 -- but it could require up to a gigawatt of electricity -- more electricity than your house will use in a century.

In contrast, the human brain is efficient enough to crunch information on just 20 watts.

Update: There seems to be some reader confusion about the use of the term "smart"  in this article. The artificial brain is built to imitate the structure and wiring of a biological brain, and is merely a simulation on the scale of a cat's cortex. It is not actually acting on its own. Some of the language has been modified to allow for more clarity.

Share this

Andrew Nusca

Editor Emeritus

Andrew Nusca is editor of SmartPlanet and an associate editor for ZDNet. Previously, he worked at Money, Men's Vogue and Popular Mechanics magazines. He holds degrees from the Columbia University Graduate School of Journalism and New York University. He is based in New York but resides in Philadelphia. Follow him on Twitter. Disclosure