May 16
CS Tea: Leah Ajmani presents "Ethics in High Stakes Predictive Settings"
Leah Ajmani is a PhD candidate in Computer Science at the GroupLens Lab at The University of Minnesota, and she is advised by Stevie Chancellor. Her research interest lies in using philosophical approaches to preempt ethical problems with machine learning technologies in high-stakes settings (e.g., mental health prediction). Prior to UMN, she received a BA in Philosophy, a BA in Computer Science, and an MEng in Computer Science from Cornell University. Her work has been published at CSCW and FAccT, premier venues in human-computer interaction and machine-learning ethics. Her work has been supported by the CS&E PhD Research Fellowship from UMN.
Abstract: In this talk, I will describe how data decisions rely on unchecked assumptions about “right” and “wrong.” If continuously left ungoverned, these assumptions produce material harm in high-stakes settings such as mental health support and criminal justice settings. My research investigates two pillars of progressing the ethics of AI/ML applications: disclosure and interrogation. These two pillars work together to help us operationalize our own ethical intuitions on user-generated data. Using the case study of predicting mental health status from social media data, I will show that ethics disclosures create implicit “standard practices” for computer scientists. This implicitness allows problematic data practices to go unchecked. Then, I will demonstrate how to use a popular philosophy method—thought experiments—to interrogate these implicit stances and make them explicit. I will conclude by discussing how we can build on our own ethical intuitions to create theories that have evaluative power, as well as robust and rigorous theories of how we ought to treat data used in predictive systems.
Join us for treats, beverages and community!
from Computer Science