Computers seeking to emulate the human brain will have to abandon current structures and become more organic, scientists and researchers said at IBM’s annual Almaden Institute conference in San Jose, California.

The theme of this year’s conference, cognitive computing, had experts declaring Wednesday that traditional software programs emulating behavior should be tossed away. Computers based on neuroscience and psychology more accurately reflect the way the brain works, they said.

Real intelligence in computers and other gadgets could mean improved devices that appeal to consumers and professionals alike. Such possibilities could be worth the $4 billion the National Institute of Standards and Technology urges academia and companies to invest in new computing theories over the next decade.

The research justifies spending, scientists said.

“The brain isn’t like a [current] computer. It’s more like an evolutionary jungle,” said Nobel Prize winner Gerald Edelman, director of the Neurosciences Institute, which devises and tests theories on how the brain works.

“They learn by making mistakes, just like we do,” said Dr. Edelman. He believes cognitive computing focuses on meeting a goal, while current artificial intelligence technology is concerned too much about following software instructions and can’t learn from errors.

Though his organization focuses on theory, Dr. Edelman is also involved in practical applications—such as creating robots like Darwin X and BrainWorks. They can learn similar to the way humans or animals do. BrainWorks won the 2005 RoboCup, a soccer-like event for robots.

Artificial Neurons

Meanwhile, halfway around the world, the École Polytechnique Fédérale de Lausanne (EPFL) in Switzerland has used supercomputers—both IBM and SGI machines—to create artificial neurons and synapses.

Showing off the visual difference between the zeros and ones of binary code versus the swirling, colorful images of computer-modeled neural columns, EPFL’s Blue Brain director Henry Markram noted, “When you zoom in, you see it’s not just an object, like a chair—it’s a whole universe.”

Some ideas are close to market.

“This is an actual project… that can be done now,” said Dr. Robert Hecht-Nielsen, director of the confabulation laboratory at the University of California, San Diego. He showed a presentation for Chancellor, a cat food dispenser that talks to owners about ordering more food.

Dr. Hecht-Nielsen said Chancellor uses technology that allows a machine to predict language. While it demonstrates grammatically correct English, he said it has shown similar proficiency in Chinese, Arabic, and Spanish.

Dr. Hecht-Nielsen is also vice president of research and development for the credit-scoring software developer Fair Isaac, which is using his confabulation theory to create vocally interactive systems.

Neural Prosthesis

Further off—but arguably more compelling—than the pet feeder is a device that can act as a “neural prosthesis” for patients who have lost cognitive or functional brain abilities due to disease or injury.

Ted Berger, director of neural engineering at the University of Southern California, said the device would be attached to a patient’s head and act in place of his or her damaged hippocampus.

“The hippocampus acts like a set of parallel processors,” said Dr. Berger. The artificial replacement would do the same.

Unfortunately, Dr. Berger’s device is not likely to be ready for more than a decade. Besides clinical trials, brain-based computers and gadgets face today’s technology limitations. Even supercomputers that can process more than a teraflop, or trillions of floating calculations per second, are too slow to develop an artificial mind.

“Even a machine like Blue Gene [the world’s fastest] is not going to simulate a full human brain,” said Dr. Markram of the EPFL. “[It] could maybe simulate a mouse brain.”

Meanwhile, some attendees questioned whether the cognitive computing theories were much different from current theories or technologies.

Learning from Scratch

But proponents noted their systems aren’t based on software that anticipates what will happen and programs the right response. Instead, they emphasized the ability to learn from scratch.

Cognitive computing doesn’t depend so much on new, faster processors as on new ways to use existing technology.

“I think with current technology and enough money, you can build a computer than can simulate a full human brain,” said Dr. Markram.