Web measurement for fairness and transparency.docx

Let’s consider some examples. The “filter bubble” arises when algorithmic systems, such as Google search or the Facebook news feed, decide what information to show a user based on her past pattern of searches and clicks. The worry is that users will be fed reinforcing viewpoints and eventually be isolated in their own bubble. At the level of demographics, the seemingly fair principle of treating “similar” users similarly can lead to a deepening of existing disparities. Online ads have been shown to display racial bias, and online prices and deals have been shown to vary based on users’ personal attributes.