Smart MACHINES.

Machines that walk, talk, and show emotions aren't science fiction anymore

Kismet loves attention. Stop to chat and Kismet will lean forward and gaze intently into your eyes. Keep talking and Kismet will look interested, surprised, or pleased--and even talk back. But get too close and Kismet will back away.

These responses seem so lifelike that visitors almost forget something: Kismet is a machine, not a person. Kismet, a robotic head powered by a bank of about 15 computers, is programmed not just to respond to people, but to learn from them and their body language.

"The goal is to build a socially intelligent machine that learns things as we learn them, through social interactions," says Kismet's creator, Dr. Cynthia Breazeal of the Massachusetts Institute of Technology (MIT).

In his film A. I.: Artificial Intelligence, director Steven Spielberg pushes the technology envelope further-much further. David is a handsome boy who happens to be a computer. Programmed to love, he is adopted by human parents. David desperately wants their love--but they only see him as a machine.

Can a machine ever have such human traits as curiosity, warmth, intuition (insight), and self-awareness? Can we build an artificial being that thinks and feels? Scientists in the field of artificial intelligence hope so.

What is Artificial Intelligence?

Artificial intelligence (AI) is the science of trying to create intelligent machines and software that can think independently, the way a person can.

The first electronic computers, built in the 1940s, could only do simple calculations. In the 1950s, experts began to design computer programs, based on a vocabulary of symbols. Soon, computers could look for patterns in language, thought, and images. Later they could use logic (reasoning) to solve problems.

People were so excited by these achievements that in 1970, Al expert Marvin Minsky made a bold prediction. Within eight years, he told Life magazine, "we'll have a machine that will be able to read Shakespeare [or] grease a car."

Today, AI technology is part of everyday life, though not in the way Minsky imagined. AI technology helps investment companies find the best places to invest money. It also alerts doctors and nurses to patients who could have a dangerous reaction to certain medications. Some software even recognizes and responds to human voices.

In 1997, an AI computer named Deep Blue made headlines when it defeated the world's top chess champion, Garry Kasparov. It was the first time a computer had defeated a chess grand master.

The Next Step

But scientists in the field of AI want to do more. They want to create machines that think and feel and learn like people.

Douglas Lenat has been working for 17 years to give a computer named Cyc (syke) human awareness. Lenat and a small team of engineers have fed Cyc millions of bits of data (information) about people and what they know, such as "infants don't drive cars."

Lenat thinks that if Cyc (short for encyclopedia) can work out the consequences of all that information, it will be able to understand the world. "Intelligence," says Lenat, "is 10 million rules."

He may not be far off. In one recent demonstration, an engineer asked Cyc about anthrax. Its response: "Do you mean Anthrax the heavy-metal band, anthrax the bacterium, or anthrax the disease?"

Dr. Rodney Brooks, director of the AI Lab at MIT, favors a different approach. He says that intelligence must be learned "from the ground up." Cog, the robot that Brooks has built, is programmed to learn the way the human brain learns. "We simulate [artificially copy] layers of the brain," Brooks explains. "One for sound, one for vision, etc."

Brooks thinks that the robot must interact with the physical world in order to gain human insights. So Cog is built with a head, torso, and arms. It can follow movements with its eyes, recognize faces, grasp objects, and even play the drums. Cog is designed to grow and develop the way a baby does, through physical interactions and movement.

Promise or Peril?

We probably won't know for decades whether or not a machine can develop human intelligence--or even come close. "The problem of how intelligence works and how it can be duplicated artificially is tremendously difficult," says Stuart Schieber of Harvard University.

He believes that development of such a machine is highly unlikely. The brain, scientists now realize, works in a more complex (complicated) way than previously thought.

But some scientists don't rule out the idea of an "intelligent" robot. They say that machines will gain more and more reasoning power--especially as researchers from different fields share their advances.

"As this enormous computing power is combined with... advances of the physical sciences ... enormous power is being unleashed," says Bill Joy of Microsoft. "These combinations open up the opportunity to completely redesign the world, for better or worse."

"By 2040," Dr. Hans P. Moravec of Carnegie Mellon University predicts, "the robots will be as smart as we are."

Few scientists will go that far. But, as Robert Epstein of San Diego State University told Congressional Quarterly magazine, "There's going to be a day when a philosopher who has doubts [about a machine's intelligence] is going to argue the point with a machine, and that's when the argument will be over."