In the 1950s, the pioneers of artificial intelligence (AI) predicted that, by the end of this century, computers would be conversing with us at work and robots would be performing our housework. But as useful as computers are, they’re nowhere close to achieving anything remotely resembling these early aspirations for humanlike behavior. Never mind something as complex as conversation: the most powerful computers struggle to reliably recognize the shape of an object, the most elementary of tasks for a ten-month-old kid.
A growing group of A1 researchers think they know where the field went wrong. The problem, the scientists say, is that Al has been trying to separate the highest, most abstract levels of thought, like language and mathematics, and to duplicate them with logical, step-by-step programs. A new movement in Al, on the other hand, takes a closer look at the more roundabout way in which nature came up with intelligence. Many of these researchers study evolution and natural adaptation instead of formal logic and conventional computer programs. Rather than digital computers and transistors, some want to work with brain cells and proteins. The results of these early efforts are as promising as they are peculiar, and the new nature-based Al movement is slowly but surely moving to the forefront of the field.
Imitating the brain’s neural (神经的)network is a huge step in the right direction, says computer scientist and biophysicist Michael Conrad, but it still misses an important aspect of natural intelligence. "People tend to treat the brain as if it were made up of color-coded transistors," he explains, "but it’s not simply a clever network of switches. There are lots of important things going on inside the brain cells themselves." Specifically, Conrad believes that many of the brain’s capabilities stem from the pattern-recognition proficiency of the individual molecules that make up each brain cell. The best way to build an artificially intelligent device, he claims, would be to build it around the same sort of molecular skills.
Right now, the notion that conventional computers and software are fundamentally incapable of matching the processes that take place in the brain remains controversial. But if it proves true, then the efforts of Conrad and his fellow AI rebels could turn out to be the only game in town. The author says that the powerful computers of today （）
A．are capable of reliably recognizing the shape of an object
B．are close to exhibiting humanlike behavior
C．are not very different in their performance from those of the 50's
D．still cannot communicate with people in a human language