|Image from The Matrix|
Of course, a computer can't pull knowledge out of thin air. It has to learn from something. That's where our surveys come into play. By responding to a survey, the computer learns a bit more about the equation it's trying to discover. It may learn that people who are shy tend to like engineering majors, but if they also like writing, they'll be more interested in journalism. In order for the computer to learn these correlations, it needs to know what the right prediction is for a set of answers. That's why every predictor has a survey that asks all the usual questions, and the predictor questions.
By studying these responses, the computer comes up with an equation that it thinks will accurately map peoples answers to the best predictions. To those who study artificial intelligence, this process is known as supervised learning. It's not always perfect, because people are complex, and hard to predict, or maybe it just doesn't have enough responses to learn from. However, most of the time it's accurate enough to help. Even if the top prediction isn't the right one, by providing the probability for each prediction, the "predictee" can get a pretty good idea of what the best prediction is.
Try it out for yourself. Look for a predictor that interests you, and see if it works. If it doesn't, you can always help it learn by giving it a new survey response!