Machines That Learn; April 1995; Scientific American Magazine; by Abu-Mostafa; 6 Page(s)
Why is an elephant big, dark and strangely shaped?" The question goes. "Because if it was small, white and round, it would be an aspirin." This answer may ring funny to human ears, but it could well prove informative to a computer trying to identify such objects as elephants or aspirin. Knowledge we commonly take for granted is not available to machines unless carefully spelled out. For machines, learning is not at all simple.
Despite the challenges, machine learning is one of the fastest-growing technologies today. The past few years have witnessed an explosion of applications, ranging from automated reading of handwritten zip codes at the post office to predicting seat demand in the airline industry. Indeed, the last time you received a credit card from a bank, chances are it was approved by a machine that learned on its own how to evaluate credit risk. And the future of machine learning is on the rise.