Assessing the Risk of Algorithmic Bias

Abstract

The last chapter laid the foundation of an informed decision on whether using an algorithm is a better or worse solution to a decision-problem than alternative approaches such as human judgment, a simple criterion, or rolling a dice. That discussion concluded with the observation that empirically, in many situations, algorithms make better decisions than alternative approaches: they make less errors (especially because human decisions often are even more biased) and can be both faster and cheaper. Introducing an algorithm to make decisions may come at the price of new biases, however.