06 April 2020
Do I Look Like a Criminal?
One of the busier intersections in the study of algorithmic bias has been in automatic judgments along race and gender, but decisions about how to frame and present information to people making decisions are immensely important, and from what I understand it seems that the evidence from studies isn’t consistent about whether something like a photo of the person involved in a given context helps or harms them. I’ve also been thinking about what the human component of AI-training systems might look like (for people evaluating appeals, or for people doing labeling tasks, for instance), and this paper seemed to approach that topic pretty directly. In this paper - “Do I Look Like a Criminal? Examining how Race Presentation Impacts Human Judgement of Recidivism” by Keri Mallari, Kori Inkpen, Paul Johns, Sarah Tan, Divya Ramesh, and Ece Kamar - Mallari et al. replicate an important study (and I think come up with problematizing findings, but offer some context to help it make sense), and go further by thinking about what this might mean when we design systems.