Making machines more like us

Innovations

A holiday shopper extracts cash from an ATM machine that checks his facial structure, not his magnetic card, to verify his identity.

A chemical engineer walks through a plant with a small device that smells hundreds of different odors to ensure nothing is amiss.

A cell phone recognizes a business traveler's voice, allowing him to check his e-mail without dialing one number.

A mere decade ago, devices like these remained in the realm of science fiction. Today, they are real and improving rapidly.

Although the prospect of talking, thinking machines lies far in the future - or only in the imagination - these devices represent a sea change in how humans interact with machines. The first steps toward imbuing silicon with the five senses also presage a new level of communication between man and machine, and possibly a new and exotic amalgam of the two.

"We work with machines that force us to be at arm's lengths from them - we have to communicate with them via a keyboard or a number pad," says Joseph Atick, the head of Visionics Corp., which makes the facial-recognition software. "So I am more enslaved by it than the other way around. That is changing rapidly...."

Among the advances:

*Facial-recognition software from a New Jersey company can sift through a million databased faces per second.

*New voice-recognition software allows computers to pick out words from background noise many times louder than the speaker's voice and more efficiently than the human ear.

*A desktop digital-scent device called iSmell can read computer code and translate it into discernible aromas like orange peels and cedar.

*A new hand-held electronic nose mimics the smelling activity of the brain and can be programmed to teach itself to sense odors far beyond human discernment.

The dream of machines that interact seamlessly on a sensory level with people goes back to the dawn of computers when films like "2001: A Space Odyssey" portrayed machines so responsive that they took on human characteristics - at times to sinister effect.

But scientists pursuing this dream have achieved only limited success. Early attempts at voice-recognition software in the 1970s and 1980s produced garbled mumbo-jumbo or devices so sensitive that most background noise rendered them inoperable. Facial recognition proved even more difficult.

In the wake of their failures, researchers began to realize the complexities of everyday functions of the human brain and the difficulties of replicating the brain's actions. As a result, many researchers concluded that computers would never adequately perform the most basic human communicative functions - facial recognition, olfactory recognition, and voice recognition.

Recently, however, scientists have grown more adept at writing computer algorithms that enable the machines to learn on their own and autonomously build upon a body of knowledge in a manner like the brain's.

They have also developed and improved upon a computer technology called neuronal networks, which rely on multiple processors or systems of nodes to perform computing tasks. It's an attempt to mimic how the brain's asymmetrical and intricately linked mass of neurons functions. The rapidly increasing speed and decreasing size and price of computer chips has also helped.

Now scientists envision in coming yearsdense networks of tiny computers embedded throughout society that can sense a human's presence and identity without interaction and respond accordingly.

A person will walk into a room, for example, and the sensor will say, "This is Joseph. Joseph likes the lights to be low."

But great difficulties still lie ahead in mastering interaction. While computers might learn to use the basic sensory functions and simple preferences, teaching them contextual cues remains problematic. "Let's say I have a verbal remote control for my television," says Gene Frantz, a senior fellow at Texas Instruments. "I am watching the news and ... the newscaster says the Dow Jones Average was off 100 points today, so the television turns itself off. How do you teach a computer the difference?"

Indeed, Dr. Frantz says computers have a long way to go. "There aren't many tasks that a computer can do better than a human," says Frantz. "It will be many years before a computer can overtake a human."

At the next level, scientists remain largely unable to create even the impression of significant intelligence in a machine. "We are at the verge of having a conversation with a machine, [but] I don't think the scientific community today knows how to solve the free-form problem of having a conversation with a machine where the subject matter is open," says David Nachmud, a research manager at IBM Laboratories.

But Dr. Nachmud and most other scientists say more human-like machine interfaces will inevitably arise. With them will arrive a whole new set of moral and societal issues and complexities. "As the complexity of the knowledge base available to the computer becomes deeper and deeper, our ability to predict what the computer is going to do will diminish," says Dr. Atick. "There isa danger that the human evolution cannot keep up with the speed at which machines evolve."

(c) Copyright 1999. The Christian Science Publishing Society