Open course ware

[MUSIC PLAYING] MIKE TEODORESCU: Hello, and welcome to this module on protected attributes and fairness through unawareness. My name is Mike Teodorescu. I'm an assistant professor of information systems at Boston College, as well as a visiting scholar at MIT D-Lab. What this module will cover will be examples of laws that codify protected attributes, as well as the base case scenario for fairness in machine learning, which is called fairness through unawareness. The use of machine learning presents both risks and opportunities. Machine learning can reduce costs by automating repetitive tasks, but could also increase biases. Certain individual attributes are commonly labeled as protected attributes, as they can be sources of social bias. These are race, religion, national origin, gender, marital status, age, and socioeconomic status. In the United States, discrimination based on these protected attributes in housing, lending, and employment is illegal. Some of the laws...

[MUSIC PLAYING] MIKE TEODORESCU: Hello, and welcome to this module on choices of fairness criteria. My name is Mike Teodorescu. I'm an assistant professor of information systems at Boston College, as was a visiting scholar at MIT D-Lab. What we'll cover in this module will be the concept of confusion matrix, as well as three popular examples of fairness criteria-- demographic parity, equalized odds, and equalized opportunity. Some countries have laws that protect specific groups of people from discrimination based on certain attributes. As we review in the previous video, these are called protected attributes. Some examples are on this slide. Regardless of the legal landscape, machine learning has the potential to produce unfair outcomes for certain groups of people. As an algorithm designer, one should make clear choices about fairness criteria. Some criteria will be reviewed in the next few slides. In the previous video, we also discussed case of fairness throug...