This document discusses various types of biases that can occur in search engines and recommender systems. It begins by defining statistical, cultural, and cognitive biases. It then explains how user data collected by these systems is biased based on the choices the systems present, and how the systems can learn to reinforce their own biases. The document discusses how personalization can lead to filter bubbles and sub-optimal solutions. It also notes that improvements in one system may degrade another system using a different optimization function. The document emphasizes that biases can occur at many levels, from the data collected to algorithms, rankings, interactions and beyond. It concludes by noting that while biases are problematic, the web itself functions thanks to certain biases like caching and popularity effects.