|Traffic | Weather | Your account||Movies | Restaurants | Today's events|
Lab works on program that emphasizes context in Web searches
CHICAGO — Many viewers were probably impressed when a character on "Star Trek" asked a computer for a cup of tea and it was produced immediately.
Not Kristian Hammond.
"I wondered why he had to ask," said Hammond, co-director of Northwestern University's intelligent-information computer lab. "A truly intelligent machine would anticipate that its operator wanted tea."
That's the kind of smarts that Hammond and his colleagues put in computers — machines ready to answer questions you haven't yet formed. To Hammond and Larry Birnbaum, the lab's other co-director, too many scientists working with artificial intelligence have spent too much time on esoteric rather than practical pursuits.
"To be useful, anything you build has to be scalable," Birnbaum said, so that one solution can be applied to many problems. Taking years to build a machine that can do one nifty thing really well just won't cut it.
Northwestern's lab specializes in guiding computers through the mountains of information that reside on the Internet and in other databases, plucking out gems a person might use. The secret is context, letting the machine know its user's immediate interests.
The first commercial product to emerge from the lab is Watson, a program that watches what you're doing on the computer and quietly presents links to information you might want. This gives Watson an advantage over traditional search engines, Hammond said.
Google, the most successful search engine, is based broadly on popularity. Type the term Armstrong into a Google search and you will get links to a floor-products company, a bicycle racer, a state university in Georgia and the bicycle racer's foundation before there's a mention of the famed musician.
But if your computer has opened a Power Point presentation about the impact of jazz on American culture when Watson is given the term "Armstrong," every item displayed relates to trumpet player Louis Armstrong.
"Context is the Achilles' heel of information systems," Hammond added.
The Watson product, marketed by the Chicago-based firm Intellext, is only the beginning of the goodies Northwestern's computer lab has in its pipeline.
The team has written programs that search for information about books a person checked out from a library or purchased from a store, about TV shows he watches and music he listens to. The information is voluminous and varied, and chances are most of it would be ignored.
But maybe the fact that Amazon.com has an out-of-print book by an author you just enjoyed would be welcome. Knowing there's a Web site where do-it-yourself plumbers give each other advice installing sink traps might interest someone who's been watching a home fix-up program on television.
When you buy an item, it might be helpful to get reviews from others who've purchased it — say a Consumer Reports rating relative to competing brands and information about buying the item at a better price.
"If you want to buy an over-the-counter medication for a cold, you could scan its bar code and the search engine could find out if it could interfere with any of the other medications you take," Hammond said.
More than a PC
While Watson and related technologies require exhaustive searches of computer databases and the Internet, they don't require a user to be tied to a computer. The TV application, for instance, could be built into the remote control.
"As you watch a show, it might raise a question in your mind," Hammond said. "You just push the button on the remote to start the search. The information could come to you now, maybe to your cellphone or Palm Pilot. But it could just as well come as an e-mail to your computer for you to see later."
Some may fear that giving computers so much data raises the specter of Big Brother, but Hammond said Watson and other potential products must reside as software on the user's own machines. "You have to have it on machines that are close to you."
The goal is for your computers to know enough about you to anticipate your needs but keep that information private as they do your bidding, he said.
Some applications at the lab go beyond giving machines insights about operators' desires.
The researchers created a search engine that scans Web logs for comments that demonstrate strong emotion. This provides a window on what bloggers are writing about — say, Wrigley's gum, for example — which could be useful to marketers, Hammond said.
"There are all these people out there telling stories," he said. "It's a great resource just waiting to be mined."
Copyright © 2006 The Seattle Times Company