Pattern Recognition: Learning from Experience
The Analytical Engine has no pretensions whatever to originate any-
thing. It can do whatever we know how to order it to perform.
—ADA LOVELACE, from her 1843 notes on the Analytical Engine
In each previous chapter, we’ve looked at an area in which the ability of computers far outstrips the ability of humans. For example, a computer can typically encrypt or decrypt a large file within a second or two, whereas it would take a human many years to perform the same computations by hand. For an even more extreme example, imagine how long it would take a human to manually compute the PageRank of billions of web pages according to the algorithm described in chapter 3. This task is so vast that, in practice, it is impossible for a human. Yet the computers at web search companies are constantly performing these computations.
In this chapter, on the other hand, we examine an area in which humans have a natural advantage: the field of pattern recognition. Pattern recognition is a subset of artificial intelligence and includes tasks such as face recognition, object recognition, speech recognition, and handwriting recognition. More specific examples would include the task of determining whether a given photograph is a picture of your sister, or determining the city and state written on a hand-addressed envelope. Thus, pattern recognition can be defined more generally as the task of getting computers to act “intelligently” based on input data that contains a lot of variability.
The word “intelligently” is in quotation marks here for good reason: the question of whether computers can ever exhibit true intelligence is highly controversial. The opening quotation of this chapter represents one of the earliest salvos in this debate: Ada Lovelace commenting, in 1843, on the design of an early mechanical computer called the Analytical Engine. Lovelace is sometimes described as the world’s first computer programmer because of her profound insights about the Analytical Engine. But in this pronouncement, she emphasizes that computers lack originality: they must slavishly follow the