" How Do We Define "Fair" Responsibly? "
Manuel Sabin (UC Berkeley)
Abstract:
The nascent field of Algorithmic Fairness recognizes that algorithms have the potential to perpetuate and codify systemic discrimination and attempts the noble goal of defining notions of “Fairness” that will mitigate this. The past year, however, has seen many critiques of the field’s direction and methodologies, illustrating how the field itself is in danger or perpetuating and codifying systems of discrimination.
This talk will review Fairness and Abstraction in Sociotechnical Systems, a work that outlines five sociotechnical “traps” that Algorithmic Fairness seems to routinely fall into. I will then present ongoing work of a framework for future Algorithmic Fairness research to be less susceptible to these five "traps." I will also briefly review methodologies and philosophies from other areas such as Sociology, Human-Computer Interaction, Community Organizing, Medicine, and others that have grappled with many of the same disciplinary issues facing Algorithmic Fairness as inspiration for setting new norms and notions of rigor for this emerging field.
While we use critiques of Algorithmic Fairness as guidance, the goal of this talk will not be to further critique the field but to find ways to address critiques and let us discuss ways to help the field grow into a responsible one.