Computerworld

IBM aiming to bring brain power to computers

Far-reaching research could change whole basis of computing
  • Agam Shah (Unknown Publication)
  • 30 November, 2008 22:00

IBM has publicised work it is doing to bring the brain's processing power to computers, in an effort to make it easier for PCs to process vast amounts of data in real time.

Researchers at the company want to put brain-related senses such as perception and interaction into hardware and software so that computers are able to process and understand the data more quickly while consuming less power, says Dharmendra Modha, an IBM researcher. He and the other researchers are bringing the neuroscience, nanotechnology and supercomputing fields together in an effort to create the new computing platform, he says.

The goal is to create machines that are mind-like and adapt to changes, which could allow companies to find more value in their data. At present, a majority of information's value is lost, but relevant data can allow businesses or individuals to make rapid decisions in time to have significant impact, Modha says.

"If we could design computers that could be in real-world environments and sense and respond in an intelligent way, it would be a tremendous step forward," he says.

There is a problem in the current core philosophy of computing and a new approach is needed, he says. Today's model first defines objectives to solve problems, after which algorithms are built to achieve those objectives.

By contrast, "The brain is the opposite. It starts with an existing algorithm and then problems [are] second. It is a computing platform that can address a wide variety of problems," Modha says.

For example, the new approach could help efficiently manage the world's water supplies through real-time analysis of data that could help discover new patterns, he says. A network of sensors could monitor temperature, pressure, wave height and ocean tide across the oceans. "Imagine streaming this data to a global brain that discovers invariant patterns and associations that no algorithms of today can do," Modha says.

It will also be able to sense the world's markets, such as stocks, bonds and real estate, extracting patterns and associations in the way the brain extracts information from those environments.

The research is not about concrete applications yet, but about understanding what the brain does and its implementation in computing, Modha says. The research includes work on nanotechnology, which has made it feasible to realise the brain function in cognitive computing chips that rival the low-power and small space of the brain, he says. Neuroscience has also matured, and supercomputing technology has progressed enough for IBM to undertake large-scale simulations to test a wide variety of hypotheses, he says.

It's a long and arduous research project that may lead to a number of technological breakthroughs, Modha says. He didn't provide a timeline for implementation of the platform.

If IBM succeeds in making this platform, it will lead to an entirely new computer architecture and programming paradigm that could overwrite the traditional ways of computing, Modha says.

For the research IBM is working with DARPA (the Defense Advanced Research Project Agency) and universities including Stanford, the University of Wisconsin in Madison, Cornell University, Columbia University Medical Centre and University of California at Merced.