Get Smart : Computers Push the Limits of Artificial Intelligence
In science fiction’s chamber of horrors there has always been a special place for the machine that thinks. The impersonal, soulless collection of circuits that out-reasons--and therefore can destroy--man had become a favorite villain even before the great HAL-versus-Dave battle in “2001: A Space Odyssey.”
It took the computer revolution of the ‘80s to persuade us that the machines were not taking over, that they were actually nothing more than highly sophisticated dray horses, electronic beasts of burden designed to do the grunt work and do it fast; they became friends.
But now Rui de Figueiredo Ph.D., professor of electrical and computer engineering and mathematics at UC Irvine, says these electronic pals are about to get smarter.
A lot smarter.
Within the decade, he predicts, they will stop working like donkeys and start thinking like humans--analyzing critically, making intelligent choices, learning, even programming themselves. Becoming the most capable servants mankind has seen.
In the next few years, such “smart” computers will appear, presaging what De Figueiredo believes will be, in less than two decades, a technological upheaval tantamount to the Industrial Revolution of the 19th Century.
In his Laboratory for Intelligent Sensors and Systems at UCI--a small partitioned room filled with computer monitors and keyboards--De Figueiredo and a group of his students are pushing the limits of what has come to be known as “artificial intelligence,” the study of machines that behave like humans. In a series of experiments in the lab, they are creating computer-based machines that are taking steps toward sophisticated human-like reasoning.
The UCI researchers are among a vanguard of scientists and academics throughout the country who are trying to forge a kind of synthesis between the human mind and the computer and who are employing a variety of disciplines in their quest: biology, electronics, mathematics and computer and related sciences.
At USC, for instance, two centers study artificial intelligence: the Center for Neural Informational and Behavioral Sciences, which studies the brain and how it processes information through biological experiments, and the Center for Neural Engineering, which studies the brain model in order to build thinking computers.
A team of UCLA researchers has tackled brain/machine experiments involving, among other things, natural language understanding, robotics and logic programming. UC San Diego and Caltech also are noted for artificial-intelligence work.
At UCI, the emphasis is on using optics and “artificial neural networks” that behave like the brain’s biological neurons. The resulting human-like characteristics enable the machines to see via camera lenses and, having seen, to interpret.
The combination of the two abilities has enabled De Figueiredo and his researchers to produce:
* A “find-the-eagle” machine, designed to locate objects obscured by extreme clutter. With a camera lens pointed at a table covered with wooden blocks that mostly obscure a small cardboard cutout of an eagle, this machine can find the eagle by, in effect, selectively eliminating everything that doesn’t look like the eagle. The computer has “seen” the eagle before and, based on its knowledge of what the eagle looks like, it can locate the bird’s shape in seconds--with as few as three edges visible.
* A “seeing-eye dog” computer. Again using a camera lens, this machine detects and locates objects and determines their nearness and their positions relative to their surroundings. The purpose is object avoidance rather than recognition, with possible applications for the blind and for automatic spacecraft docking.
* An optimal lighting machine. “To my knowledge,” says De Figueiredo, “there is no good scientific technique on how to light a scene on a movie or TV set or a stage, how to place the lights, orient them and control intensity and color. We use actual cameras as light meters. A TV camera looks at something and the image is translated into the computer, and you keep on moving the lights until a particular criterion goes up.”
* “Photometric stereo,” a technique that makes it possible to obtain an object’s three-dimensional shape from a remote location. Although it is not a new technique, De Figueiredo and his researchers have substantially improved it.
The technique involves the computer’s ability to look at a three-dimensional object through a camera lens, analyze the light reflected off it from lights placed at different sources and reproduce a three-dimensional rendering of the object on a computer screen or paper. Possible uses include locating, identifying and charting objects in hostile environments, like space, the sea and nuclear plants, or in such dangerous situations as bomb detection and fire fighting.
These experiments, says De Figueiredo, are among the first steps toward a world in which man and machine will become more than partners; they will probably develop an almost symbiotic relationship.
“In the next century, we’re going to think of both (computers and humans) as parts of the same system, as one system,” he says.
“Machines are going to have almost the same status as humans in the factory, for example, and you’ll think of the machine as sort of your companion, which is one of the functions of an intelligent machine.
“Machines can help the handicapped and the aged and the disabled perform functions, to enjoy life and have a viable life,” he says. “An intelligent machine can work in an unstructured environment like your house, and it can eventually learn your habits and your style of life.”
De Figueiredo, who went to UCI last October from Rice University, where he headed the program on intelligent systems and robotics, admits that this can sound like science fiction, even to scientists. The idea of the human-like computer can still carry unsettling overtones. But the basis for the theory--and, says De Figueiredo, the eventual and inevitable fact--lies in one of the most pragmatic of disciplines: mathematics.
In one of the major thrusts of his research, De Figueiredo is developing a mathematical theory of artificial neural networks, the sophisticated guts of the computer that he says can most closely simulate the function of the biological neurons of the human brain.
The structure of the artificial neural network is unfixed, allowing for constant growth and evolution. The mathematical principles that govern them, says De Figueiredo, are the same as those applied to understanding photons of light. He can, he says, explain the “thinking” structure of intelligent machines by using mathematics alone.
Still, when one starts to consider the possibilities, the discussion turns not only to mathematics, but to logic and philosophy. De Figueiredo draws a comparison between man’s understanding of the solar system and a computer’s ability to reason:
“The first try at an explanation (of the solar system) was made by the Egyptians during the time of the Pharaohs. They described the behavior of the stars and planets. They described how things are done rather than why things happen. It wasn’t until Newton that people really tried to understand the simple motion of the planets and stars.”
So far, he says, electronic braininess has been limited. In the battle of wits, man is still king of the hill.
“You look at the human brain,” says De Figueiredo, “and you see that it doesn’t work very rapidly like the Cray (supercomputer). The neurons are very slow. They operate in milliseconds, but the Cray machine can operate at nanoseconds, extremely rapidly. But we can think much more rapidly than a machine can. We can arrive at a decision faster than a machine. Computers are very rapid, but they’re very dumb. They’re sophisticated calculators. Somebody has done the thinking before, and the computer calculates.”
But the machine is starting to catch up. De Figueiredo points to the year 2010 as a watershed year for the “smart” computer--25 years after the personal computer revolution of 1985. The sophistication of artificial neural networks may one day even allow the development of machines so intelligent, and so small, that they can be injected into the human body and programmed to selectively attack viruses or cancer cells.
In the end, however, any machine, however sophisticated, still is a product of human design, he says; rather than supplanting man, the new machines are more likely to help us better know who we are as humans.
“I think people are insecure in dealing with machines, and that’s what brings fear, but I think that fear is going to vanish like it did with the (personal) computers,” says De Figueiredo.
“They can help you rather than cause problems. And, absolutely, I think that machines can help us in understanding one another. They can make our lives rich.”
