This paper was a ground-breaking paper that discusses fairness as a fundamental concept among individuals, as opposed to requiring being classified among an already identified salient group. Read sections 1-3 for the main ideas (don’t worry about the details of the proofs).
Remark: This paper considers classifiers that potentially return distributions of outcomes (as opposed to outcomes itself). As such, the metric on the target space is a distance between distributions. However, the metric that’s primarily used is the total variation distance, which takes the point-wise difference of outcomes; this is exactly the case covered in class.
Carefully read a response to the notion of Individual Fairness criteria:
The following blog post explains some warnings about matching processes. You may find it useful to read the previous posts as well (intro to confounding bias; intro to propensity score matching). We will go over these background concepts in class.
Why Propensity Scores Should Not Be Used for Matching by Gary King and Richard Nielsen
From the What’s Fair about Individual Fairness paper, summarize the three main concerns about individual fairness. Do you agree with these concerns? Give your informed opinion on the relative merits of individual vs group notions of fairness.