Re:Algorithmic injustice: a relational ethics approach
by simulo on Friday March 12, 2021 @09:36AM
(#26436233)
> it's the only kind of approach that
> generalizes across different cultures.
I agree that the "rational" approache’s mobility is very interesting and powerful.
The general point from science and technology studies would be that, to move research results around and get this agreement, there needs to be a vast infrastructure that standardizes what rationality depends on – and even then, scientific replication, e.g., can be quite difficult (Collins story of the attempt to replicate building an early TEA-Laser is an interesting example).
The point specific to the paper is that determining what you assume to be bias and what you assume to be unbiased is already a political and ethical decision that can not be 'rationally' decided from within the research.
But I think she's neglecting the important reason why what she calls a "rational" approach is so popular: it's the only kind of approach that generalizes across different cultures. When you identify an object detection model that tends to perform poorly on people with dark skin, researchers in Dublin, Tokyo, Beijing, or Lagos can understand and agree with what you've found. If you want to find ways to reduce this bias, they'll all be able to help you look, and you can help them with their challenges in turn.
When you move to her "relational" approach, and start talking about "personhood, data, justice, and everything in between", those researchers are going to have wildly divergent and often irreconcilable views. Even two researchers in the same city might struggle to work together on these terms, if they come from different subcultures with different views about how the world should be. The possibility of cross-cultural collaboration is a big thing to give up, and to most people (including myself) it's really a core requirement for any scientific research practice.