Web May Hold the Key to Achieving Artificial Intelligence

If you ran into him online, you might first be struck by the kid’s prodigious memory. He calls himself "SmarterChild" and can recite a litany of facts — this season’s entire baseball lineup, every word in the dictionary, and the weather in major cities across the country.

By Ariana Eunjung Cha
Washington Post Staff Writer

But other queries provoke odd responses.

A question about SmarterChild’s age returns, "One year, one month, 11 days, 16 hours, 7 minutes, 47 seconds!" Asking where he lives gets, "In a clean room at a high-tech hosting facility in California."

SmarterChild, a computer program, is part of a new species of "chatterbots" that are renewing debate about the extent to which computers can achieve intelligence.

The electronic personalities of this generation use the vast repository of information on the World Wide Web as their memory bank, not just some rigid database. To answer questions about baseball, for instance, SmarterChild scours the Web site of SportsTicker Enterprises LP; for spelling, it goes to the American Heritage Dictionary online; for the weather, it visits Intellicast.com.

The company that conceived SmarterChild, Active Buddy Inc., created the bot as a marketing tool that would engage people in conversation and then tell them about various products or services.

Other companies have begun using these systems to help with customer service or Web searching. Eventually, however, some believe that technicians will be able to turn programs like SmarterChild into more intelligent systems. That is, the network will naturally begin to evolve into a sort of global brain, one made up of the constellation of the roughly 1 billion computers comprising the Internet.

Such a system might automatically offer advice on city planning based on demographic patterns or recommend that printing cease on a novel that hasn’t sold a copy in weeks. It might even pinpoint the outbreak of a disease based on the health complaints people are searching for information about online.

The idea that computers might serendipitously comb through troves of data to produce useful bits of information faces numerous political, economic and social hurdles, such as privacy concerns, not to mention enormous technical obstacles. And skeptics abound.

Push Singh, who runs an artificial intelligence project for the Massachusetts Institute of Technology’s fabled Media Lab, scoffs at the notion that such AI systems are likely to develop any time soon.

"Intelligence," he said, "is not a simple thing, and it’s not going to arise accidentally."

Scientists have worked to create an artificially intelligent agent for as long as there have been computers. Yet every revolution in power and processing speed has only pushed AI further into the future as science smacks up against the complex biology of intelligence.

But the infinite nature of the Web echoes the infinite mystery of the brain, raising the possibility of success with artificial intelligence at some level.

Singh said he recognizes that. "The Web as it stands is not the future," he said. "There will be something that comes after the Web, something that I’m sure will be built on AI technologies."
Virtual Boy Scout

Created by engineer Timothy Kay, SmarterChild began popping up in instant messaging systems last summer. Since then, close to 9 million people have talked to him.

Chatterbots, which converse with people through real-time text messages, have existed on the Internet for years. Underneath their friendly exterior, they are basically databases built by humans that link typical questions to stock responses.

SmarterChild is different. Its database is limited only by the reach of the Web. Scientists are beginning to capitalize on the way the global network converts "knowledge," or at least reams of data, into a digital language computers can understand.

"The Internet starts to make things possible again," said Michael Kearns, a computer science professor at the University of Pennsylvania and former director of artificial intelligence at AT&T Labs.

To be sure, SmarterChild often spits out gibberish and non sequiturs just like its predecessors. But its ability to access and digest online information represents a major step for artificial intelligence.

So potent are the possibilities that researchers at a diverse group of academic, nonprofit and government-backed and corporate centers such as MIT, the World Wide Web (W3) Consortium, the Los Alamos National Laboratory and Microsoft Corp. are embarking on projects to tap information already available on the network.

For the most part, bots like SmarterChild are able to talk only about certain established topics. But some have been able to reach a touchstone of artificial intelligence — passing the Turing Test, in which researchers ask humans to guess whether they are communicating with a person or a machine. If people can’t tell the difference, the machines are deemed to have passed the test.

Some scientists believe that by fusing the many systems of the Internet, an artificial being with the combined knowledge of, say, Albert Einstein, Richard Nixon and Britney Spears could be born.

But before that happens, the AI community must overcome two huge barriers.

The first is that computers have a hard time reading Web pages because the files are labeled in different ways, some more unconventional than others. That’s why Active Buddy programmers need to tell SmarterChild where to look for the weather; it would be a significantly more difficult task to let him find it.

A group led by Tim Berners-Lee, creator of the Web and director of the W3 Consortium, hopes to fix some of that by assigning keywords or tags to text, sounds and images. The task of renaming pages, however, must be done manually and will take years to complete.

Another wall that AI projects have hit is that while online entities like SmarterChild can regurgitate and process information more accurately and faster than any human, they lack common sense, a basic grounding of knowledge that is obvious to any young child. The computer mind, for instance, has had difficulty understanding concepts like "once people die, they stop buying things" or "trees don’t grow in cars."

MIT’s Singh and others are trying to create a "knowledge base" that can be implanted into AI projects by using human volunteers. People who want to help the project, called the Open Mind Initiative, can go to its Web site and type whatever comes to mind (and makes sense) when they are flashed certain photographs, diagrams or sentences.
A Global Brain

Another project, led by researchers at the Free University of Brussels in a loose collaboration between nearly a dozen scientists, psychologists and biologists around the world, attempts to help computers understand relationships between people, objects and ideas by studying how humans access information online.

It all began one evening in 1999, when a graduate student named Johan Bollen created an early version of software that gives Web sites the ability to automatically reorganize the content on their pages.

Using the "cookies" that sites use to identify and track Web users, the program analyzes the routes people take to get information and tries to simplify them. The software mimics the human brain, strengthening, dissolving and even creating hyperlinks on a page based on patterns of use; the Web pages act like neurons, and the links act like the synapses between them.

If it finds that people often go from A to B to C, it will create a path directly from A to C. For instance, if many people are hopping from the main Yahoo page to the Finance section and then to the page about WorldCom Inc., the program might create a new link from the main Yahoo page to the telecommunications company’s Web site.

"It’s about helping people find the connections between information," said Bollen, 30, now an assistant professor at Old Dominion University in Norfolk. "You have so much junk on the World Wide Web there’s no guarantee that the information is good and fits what you desire. What I’m talking about is a Web that bends itself to the actions of its users."

Bollen’s technology is already being used as part of a library search engine at Los Alamos called the Active Recommendation Project. The program can offer people a list of links that may include relevant material, even if the links don’t contain the word the user entered. The more people use the system, the smarter it becomes.

One of the central ideas of researchers who believe in the vision of a "global brain" is that the earth can be seen as a single organism with many complementary parts that must work together to succeed.

Francis Heylighen, a professor at the Free University who oversaw Bollen’s initial project, likes to use insects such as ants, bees and termites as examples. "Individually dumb, but capable of surprisingly intelligent behavior when functioning as a group," he said.

The ant analogy is exactly what frightens some of his peers.

They worry that a "hive mind" might stifle freedom and individuality. Already, some efforts to reorganize Web sites based on the preferences of the majority end up drowning out the voices of the minority.

Others are concerned about privacy issues, that computer networks will become all-knowing. Still others worry about the Internet becoming all-powerful.

Los Alamos scientist Luis Rocha, who is heading up the digital libraries project, said he doesn’t know whether the Internet could ever become a malevolent, intelligent, self-aware being.

Still, he said: "A lot of times science is moved by far-fetched goals. You aim for the moon and hit London. And a lot of times, that’s somewhere you haven’t gone before."

Staff researcher Richard A. Drezen contributed to this report.

© 2002 The Washington Post Company

http://www.washingtonpost.com/ac2/wp-dyn/A43363-2002Sep5?language=printer