Friday, August 19, 2011
IBM unveils brain-like chip
By John Roach, contributing writer at msnbc.com.
Computer chips with worm-like intelligence were unveiled today by researchers at IBM, a breakthrough, they say, on the road to creating computers that function like the human brain.
For now, achieving the goal of human-like intelligence in a computer with the size and power needs of our brains is a long ways off, Dharmendra Modha, the researcher leading the project, told me, but the chips he held as we spoke were proof that a "new generation" of computers are in the offing.
"It is IBM's first cognitive computer core that brings together computation in the form of neurons, memory in the form of synapses and communication in the form of axons," he said.
Such chips, he said, could form the basis of computers that are able to monitor real-time traffic-light cameras, notice an anomaly and dispatch an ambulance in time to save lives.
Other potential applications include lining the ocean with sensors for everything from temperature, humidity and wave height to acoustics and turbidity. The computer would constantly monitor all that data and detect patterns such as rogue waves that could interrupt shipping or a tsunami that could wipe out coastal villages.
A glove instrumented with sight, smell, temperature and other sensors and put on the hand of produce handlers at the grocery store could identify fruits and veggies that are contaminated, again saving lives.
The chips do this by integrating memory and processing, unlike today's computers, which separate the functions. It's a difference, he said, between growing food in one part of the world then eating it in another and a farmers market where you buy and eat locally grown food.
The silicon cores unveiled today aren't at the level of the human brain yet. They have 256 neuron-like nodes. One has what the company calls 262,144 programmable synapses, the other contains 65,536 synapses. They can drive a car through a simple maze and reconfigure a triangle from just a fragment, Modha said.
It can also play Pong, the 1970s arcade game. "It might beat you, I don't promise, but it might," Modha said.
The next step is to take these tiny brain-like circuits and weave them into a system that eventually has 10 billion neurons and 100 trillion synapses that consume just 1 kilowatt of power and occupy the same volume as a shoebox.
The technology, Modha added, is a departure from the way computers have evolved over the past 100 years and allowed the development of machines such as the question-answering maverick Watson that won a well-publicized game of "Jeopardy" earlier this year.
"Watson represents the epitome of artificial intelligence today, I would say," Modha said. "And we are trying to emulate the brain. They are yin and yang, salt and pepper, they may work together and complement each other, but they are not the same."
The project has keen interest from the U.S. government. As IBM unveiled the chips, they also announced $21 million in new funding from DARPA for the project, which is named SyNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics).
The goal of the project is a system that not only analyzes complex information from multiple sensory modalities at once, but also dynamically rewires itself as it interacts with its environment — all while rivaling the human brain's compact size and low power usage.
We are still years away from such a computer, Modha said, and when it arrives it is likely to complement other computers and humans, not outsmart us or our machines.
"The human brain is so darn bloody awesome it shames me," he said. "The thing is, today's computers are so darn bloody bad in terms of power, space and functionality ... you pick a problem with so much room to play that you improve it slightly, you look like a hero."