Making computers act smarter through fast pattern recognition

Programs can seek out related information, or track callers' emotions

April 10, 2005|By Jon Van | Jon Van,CHICAGO TRIBUNE

Donald McLellan has a pretty smart computer.

It watches what he reads and writes, and can go online for information it thinks he might need.

"If I didn't have it, I'd have to hire a research analyst to sit next to me," said the corporate vice president at Motorola Inc.

McLellan uses software called Watson, developed at Northwestern University and marketed by Chicago's Intellext Inc., which is part of a new wave of programs that provide computers with something akin to intelligence. But these programs do not think for their users.

"Computers haven't gotten more deep cognitively, but they have gotten a lot faster," said Steve White, a senior research manager at International Business Machines Corp. "What we can do is use that speed to do brute force calculations to solve problems."

Intellext's Watson is one example of software that takes advantage of more powerful computers. Another is a program that tracks the emotions of people talking on the phone, created by a company that monitors call-center conversations.

Computers have long been likened to human brains, sparking fears and hopes that someday a collection of silicon and wires might think like a person. But even today's most powerful units are not smart enough to tie a shoelace or do anything most human 4-year-olds accomplish without thinking.

Even so, increasing computing power enables machines to recognize patterns and operate in ways that seem eerily intelligent.

When Watson sees a technical term McLellan types, it has no idea of the meaning. But by using pattern recognition, it can put the term into a context, enabling it to find relevant documents created by others at Motorola.

McLellan directs Motorola's strategic transactions group, and he receives many proposals from people outside the Schaumburg, Ill., company. With Watson, he is able to quickly find the appropriate people within the company to help evaluate those proposals.

As McLellan delves into something Watson finds, the program takes note of what interests him and explores further to get additional related material. Watson is a remarkably useful tool, he said.

The cost of Watson, which was unveiled in February, ranges from $99 for an individual user to thousands of dollars for large companies.

Kristian Hammond, a professor at Northwestern and co-founder of Intellext, was active in the artificial-intelligence branch of computer science for years at Yale University and the University of Chicago before joining Northwestern.

Hammond says he no longer embraces the notion of intelligence commonly shared by artificial-intelligence researchers.

"That model is that people have a clear, crisp idea of what they're thinking," Hammond said. "Our model is that there's never a clear idea; often it's just a collection of ideas in a context. You change the context and you change the intelligence."

A similar philosophy is at work at NICE Systems Inc., a Rutherford, N.J., company that records call-center conversations to monitor for quality. Its software can determine when a caller becomes emotional and can recognize specific words.

Useful remarks

"We didn't start with artificial intelligence in mind, but that's the direction we've gone," said Eyal Danon, the company's marketing director.

NICE systems record 500 calls a second for clients around the world. Managers listen to fewer than 1 percent of those calls, and the ones they hear are usually bland and provide few valuable insights about customers.

Identifying conversations in which customers get emotional can make a big difference. Those are generally more interesting and useful, Danon said.

About 300 scientists worked for three years to develop 26 algorithms that make the NICE monitoring system work.

"It's similar to a lie detector," Danon said. "It looks at the pitch, volume, tone, speed and tempo of the caller's voice, and watches how those change over time. It looks for words like `cancel' and any mentions of the company's competitors."

Sharon Whitwam, vice president of member services for WPS Health Insurance in Madison, Wis., said the system is helpful.

Tapping into calls during which the caller is emotional has saved WPS many headaches, Whitwam said, adding, "If we see a high level of emotion in a call, management will listen to it and make a call back to the customer if necessary."

Complicated problems are being solved by using approaches having little to do with artificial intelligence, IBM's White said.

Contrary to a once fairly widespread belief, White said, computers don't work the way human brains do. Adding speed makes computers more useful but doesn't endow them with intelligence.

"We need to better understand the brain's architecture - bundles of interconnected neurons," White said.

"They're very slow as computing elements, but their connections make them very powerful and intelligent. We might build a different kind of computer modeled after that."

Some researchers are trying to do just that.

Jeff Hawkins, a computer architect in California's Silicon Valley who started the high-tech companies Palm Computing and Handspring, is developing a computer that mimics the human brain.

`Can be done'

Hawkins founded the Redwood Neuroscience Institute in Palo Alto, Calif., to further his research. A working computer that demonstrates humanlike intelligence will be ready by the end of this decade, he predicts.

"It can be done," Hawkins said. "We are over the hump."

Such an intelligent computer could be taught to look at random images and tell the difference between a cat and a dog or a duck and a chicken, much as young children learn to do. The machine might also learn to walk down a hill or drive a car, Hawkins said.

The Chicago Tribune is a Tribune Publishing newspaper.

Baltimore Sun Articles
|
|
|
Please note the green-lined linked article text has been applied commercially without any involvement from our newsroom editors, reporters or any other editorial staff.