Image from “The Library of Missing Datasets” by Mimi Onuoha (2016)
As algorithmic decisions and likelihood predictions reach ever more deeply, and with increasing consequence, into our lives, there is an increasing mandate that they be “fair”. This program comprises a short course on the theory of algorithmic fairness taught by Dwork and Rothblum, as well as research talks by leading researchers in some application areas.
After an investigation of an array of “first wave” fairness definitions and their behaviors under composition, the course will highlight a class of desiderata that aim to bridge the gap between statistical and individual fairness notions and examine the meaning of likelihood predictions through the lens of complexity theory. Attention will be paid to open fairness questions surrounding the choice of data by which individuals are represented to the algorithm, and the proxies used for outcomes, the fairness desiderata, and the pressing problem of moving from algorithms that reproduce the world as it is to algorithms that lead us to a more ideal world. The course will end with a deeper look into 1-2 application areas.
This program is followed by a workshop Who Counts?, Sex and Gender Bias in Data, organized by Ruha Benjamin (Princeton University), Cynthia Dwork, and Patricia Williams (Northeastern University).
Noa Dagan
(Harvard Medical School)
Cynthia Dwork
(Harvard University)
Guy Rothblum
(Apple Inc.)
Copyright. All Rights Reserved.