Examples of unfairness in Artificial Intelligence abound in the popular press, despite relentless growth in the power of machine learning. The emerging field of algorithmic fairness explores such questions as:
What does it mean for an algorithm to be fair, and how can we translate this into conditions that can be mathematically specified and checked?
What can cause algorithms to be unfair in the first place, and how can this be prevented?
When, and for what kinds of fairness, is there a tradeoff between fairness and accuracy? How do the equations change in long-lived systems?
We will survey the landscape of algorithmic fairness, highlighting some recent work in fair prediction that improves accuracy and has surprising applications to settings in which fairness itself is not a concern.
Copyright. All Rights Reserved.