that enables it to form general concepts from a few specific lessons. So it can learn by experience as it gets older. What we’re doing is developing ways of providing that kind of basic programming for a machine.”‘
“It’s called an IQ transplant,” Chris murmured from his console without looking up.
“You mean like a kid doesn’t need to go round burning itself on everything in the house to get the message that hot things hurt?” Laura offered after a moment’s reflection.
Dyer nodded. “That kind of thing and more basic stuff too.”
“How do you mean, more basic stuff?” Laura asked.
“The kind of thing that you have to know before you can even stretch your hand out to touch something,” Dyer said. “All the things that are so obvious that you don’t even realize you had to learn them once. But to a computer they’re not obvious at all. We’re finding out how to teach it.” Laura was staring at him suspiciously. He went on, “Even a child of two has a mental model of objects occupying a three-dimensional space, and of itself being one of them. It can interpret visual patterns on its retina in terms of that space. It knows that objects fall if they’re unsupported, that two of them can’t be in the same place at the same time, that they continue to exist when you can’t see them . . . that hard things can break and soft things can bend . . . things like that. Those things go together to set up a child’s pattern of basic knowledge of the world around it. When it’s given a problem to solve or when it sets out to perform some task, it automatically applies constraints, based on what it’s learned, that enables it to separate the possible approaches that make sense from the ones that don’t.”
“Any problem at all is simple to solve when you take away all the constraints,” Ron chipped in.
“Any problem at all?” Laura sounded distinctly skeptical.
“If the cat’s got fleas, one guaranteed way of getting rid of them is to throw the cat in the incinerator,” Chris came in. “Intense heat is a fail-safe way of killing fleas.”
“The problem is it kills cats too,” Ron said. “But when you gave me the problem you didn’t tell me I wasn’t supposed to do that. You assumed my own common sense would tell me that part of it. Except I happen to be a computer. I don’t have any common sense.”
“The solution is quite simple until you start applying commonsense constraints to it,” Dyer summarized. “The more common sense you have, the more you’ll constrain the acceptable solutions. So the decisions get tougher but the answers are more effective.”
Laura traced a long, red-painted nail slowly along the glass top of the tank while she digested what had been said. Then she looked up and tossed her hair from the side of her face in the same motion.
“Okay, I think I can see what you’re getting at,” she said. “So how do you begin getting a machine to think like that?”
“By making it do exactly what a baby has to do,” Dyer told her. “We give it a world to grow up in and learn from.” He caught Laura’s puzzled look and turned toward Chris. “How’s Hector today?”
“Oh, he’s feeling okay,” Chris replied. “We had him running earlier. Want a demo for Laura?”
“Why not?” Dyer answered. Despite herself, Laura was becoming intrigued. She watched as Chris exchanged a brief dialogue with the console. Then a white iridescent glow appeared suddenly, pervading the entire volume of the tank. Laura jumped back instinctively with a squeal. Dyer grinned. After a few seconds the glow condensed into patches of color that quickly coalesced and stabilized into a vivid and detailed holographic image.
The image was a miniature representation of a one-story house, looking to all intents and purposes like a real, solid children’s doll house, complete with fittings and furnishings. When Laura approached it again and studied it more carefully, however, she
Suzanne Woods Fisher, Mary Ann Kinsinger