“This question that has been lurking around almost since artificial intelligence and computation began,” says Randall Davis, professor of Electrical Engineering and Computer Science at MIT. The first step to an answer, he says, is to define what we mean by artificial intelligence: AI “is the attempt to program computers to do the kinds of tasks that represent acts of intelligence in people.” Early examples, Davis explains, were very computational in nature – “proving theorems in geometry and playing checkers and chess.” Other tasks, like driving a car and navigating through traffic, doing scientific experiments, or diagnosing medical conditions, came later – but all these tasks, he adds, can now be managed, to some degree at least, by computer programs.
But this doesn’t get us to the essence of the question: Are machines limited to doing what we tell them to do, or can they come up with a new idea on their own? You’ve probably heard that a computer can do only what it has been programmed (by humans) to do. To a large extent this is correct – but it doesn’t necessarily follow that the outcome is something we can always predict; computers can surprise us.
“Just because I program a machine doesn’t mean I can predict what it will do,” says Davis. “Sometimes the combination of the instructions we provide and the input from the real world is so complex that we can’t tell what the machine will find. For example, I can program a chess application with the criteria for identifying a good move in a given situation. But by virtue of its processing capabilities, the program can analyze millions of possible moves and may come up with something I can’t predict.”
But is this creativity, or just computational superiority? Psychologists have studied creativity for years and it turns out there are ways to boost creativity. One of them is to try looking at the opposite of what you’re seeking. As long as 30 years ago, a friend and colleague of Davis’s was working on a program designed to uncover interesting avenues of study in elementary number theory. One area the program found “interesting” was prime numbers, which are distinctive because they are divisible only by themselves and one. Prime numbers thus have only two divisors.
To his friend’s delight, his program tried looking at the opposite of prime numbers: numbers that are especially rich in divisors. According to Davis, “My friend was intrigued because he hadn’t anticipated this outcome. He then did a survey of research in this area and found only one other example of someone looking at what were called maximally divisible numbers: the Indian mathematician Srinivasa Ramanujan, a self-taught genius who died in 1920. This was clearly a creative act performed by the program.”
Of course, an application or a machine still relies on a human to place it in a certain path and give it instructions to guide its way. “Otherwise it will get lost in the possibilities,” says Davis. “We can’t tell it what it will find at the end of the path, but we can give it criteria that help it distinguish between promising paths and unpromising paths. Because of a computer’s analytical capabilities, however, it’s not unusual for it to surprise us.” — Jason M. Rubin
Thanks to 16-year-old Armando Montalvo from Nuevo León, Mexico, for this question.