January 29, 1996


In a Sea of Web Data, a Sudden Flash of Light

With some 16 million pages on the World Wide Web, a gazillion entertainment titles, and oceans of data on big-ticket products like cars, soon we will need the rest of our lives to find what we want.

For many years, performing this hunter-gatherer function has been the promise of software agents. Researchers have long promised that these chunks of computer code, imbued with your preferences and the authority to make decisions about them, will be able to comb computer networks and data to deliver to you the information, products or services you want, without requiring your active involvement.

These agent-based services, now coming to market, could radically increase the utility of the Net if they work as promised. And on Monday one of the first commercial applications is making its official debut on the Web.

Agents Inc., a Cambridge, Mass., commercial spinoff of the Media Lab at the Massachusetts Institute of Technology, will bring out its Firefly consumer service on Monday. Firefly uses MIT's agent technology to automate word-of-mouth recommendations via the Web. Firefly's debut product category is music, already a hot Web genre.

When you register at Firefly, you are asked to begin ranking which artists you like and dislike. As soon as you begin the process, Firefly begins to compare your preferences with the 30,000-plus people who are testing the system today.

What Firefly is looking for, as it examines your preferences, is what Agents calls your "nearest neighbors" -- that is, those people whose taste in music most closely matches yours. These "nearest neighbors" actually change dynamically -- Firefly continues to update your profile, and theirs, from minute to minute as each user continues to log preferences with the system.

And because there are so many other users to compare you with, Firefly immediately begins to suggest artists you might like.

Pattie Maes, a co-founder of Agents and the associate professor at MIT who is most associated with software agents, says the Firefly concept is especially useful in today's music market.

"The whole notion of genre is exploding," Dr. Maes said. "People don't collect just jazz records now, or pop. Our similarities engine takes that into consideration and leverages it -- if you are passionate about the Beatles, the system can help determine what else you might be passionate about -- yes, John Lennon, but also sometimes Mozart."

Dr. Maes said that systems like Firefly might make more entertainment products commercially viable. "It will become easy to find even a very small, boutique audience for quality stuff," she said.

Firefly's preference-matching is impressive, but the service gets its edge from the community environment it creates: you can find out who likes the same music as you, which nowadays can be a critical prerequisite for a good relationship.

All Firefly members -- their privacy protected in many ways -- get a home page listing their favorite artists., and a place for others to leave messages. They can search for "neighbors" and talk to them directly., chat in groups, or whisper messages directly to each other while they are using the service.

"We're trying to counter the neutron bomb effect on the Web -- all information and no people," said Nicholas Grouf, the chief executive and co-founder of Agents.

Though other companies with agent products are less concerned with community, they may bring a new level of utility to data, whether found on the Web or elsewhere.

One promising start-up is a La Jolla, Calif., company called Reason Line. Its president, Steve Tomlin, the former vice president and general manager of interactive media for the home-shopping channel QVC Inc., calls Reason Line's service a "decision processor" that will eventually help people with everything from buying a car to filtering news services.

"We don't ask you what cars you like, we ask you, for example, 'How do you feel about safety?"' Tomlin said. The system lets you fix your response -- "I care a lot" or "I don't care at all" -- on a sliding scale.

This ability to capture nuance statistically, Tomlin said, is one way that Reason Line's software differs from the CD-ROM products available today for selecting colleges or cars.

Reason Line is teaming with data collection companies for a number of different products, Tomlin said, and is likely to start its service before fall.

And General Magic Inc. -- the Silicon Valley company that popularized the idea of software agents -- is getting its act together and taking it on the Web.

The company is to announce on Monday that it will begin shipping a free, pre-release version of Active Web Tools, a software kit for developers based on General Magic's Telescript, a programming language to create agent-based network services.

For example, a Telescript agent could visit a variety of Web sites several times a day to check on financial news and stock quotes, said Tom Hershenson, a marketing executive at of General Magic. "When your portfolio value changes, Telescript will page or fax you, or initiate some action to contact you," he said.

Software agents that really work as Hershenson, Tomlin and Dr. Maes describe could easily galvanize the on-line services business. For one, their ability to deliver custom information could be the beginning of the end for services like America Online that try -- and often fall short -- to aggregate the perfect mix of information for a mass market.

What's more, agents could catalyze a whole new surge of growth on the Net. People are already annoyed about wading through tons of Web junk to get the information they want. Software that saves them the trouble would be a huge step toward making them more productive without cramping the unwieldy growth that defines the Internet.

The Digital Commerce column is published weekly, on Mondays. You may view previous columns.

Home | Sections | Contents | Search | Forums | Help

Copyright 1996 The New York Times Company