Identity & Identification in Computational Media

How might we use machine learning to both understand historical patterns of discrimination and to shape new, more just futures? How might we treat them as we do global climate change models? These models which give us the likely future, given past and present actions and mistakes, not so that we’ll accept this future, but rather seek to change it.

I’ll be discussing an ongoing research agenda aimed at characterizing the social and cultural diversity of the people represented in image sets, in the context of image retrieval systems. Specifically, I’ll be focusing on the social and ethical challenges of this project that emerge due to the necessity of algorithmically encoding/quantifying visual characteristics of people that act as signifiers for complex (and fundamentally non-visual) social categories such as gender, race, etc.

This presentation examines the ethics and challenges of measuring socially constructed identity categories. More specifically, Dr. Heldman will provide an overview of how race, gender, and other identity groups are assigned social meaning, then discuss how this complicates the process of measuring these identities in media content. She will lay out steps taken during the research process to improve validity and reliability, as well as discuss ethical issues in using identity categories in media research.

Prosocial and business imperative of leveraging/optimizing technology to mitigate bias in media in order to promote scalable and sustained progress in society.