Prediction Revisited

7Toward Complexity

Our relevance-based approach to prediction is intended to overcome the failure of linear regression analysis to address asymmetric relationships between inputs and predictions. But you may question the need for yet another apparatus to deal with complexities such as asymmetry, given the rapidly expanding suite of machine learning algorithms. In this chapter we describe how partial sample regression is related to machine learning and why we should view these approaches as complementary rather than mutually exclusive. We also explain why partial sample regression might often be the preferred approach.

Toward Complexity Conceptually

The idea that machines could match, or exceed, the power of human decision making has captivated people for centuries. Often the test of a machine's ability to match human decision making has focused on popular games like chess. One notorious and amusing example dates to 1770. A Hungarian inventor known as Wolfgang von Kempelen unveiled an elaborate cabinet full of gears topped by a life-sized replica of a man dressed in traditional Ottoman clothing with a turban and a smoking pipe. His invention, known as the Turk, was able to grasp and move chess pieces and play effectively enough to win most matches. The spectacle mystified crowds across the world. The contraption publicly defeated Napoleon Bonaparte and Benjamin Franklin, among others. But the machine's talent was a clever illusion. A skilled chess player would sit inside a ...