I study social inequality where institutions and individuals come into contact. In doing so, I employ computational methods to sample from interactions and physical space, and use these traces in behavioral studies. For example, my collaborators and I have used natural language processing to measure police respect from body camera footage: how these aspects of interactions differ across race, and whether these conversations can be changed. In other work, we use street-level imagery as stimuli to compare neighborhood-level indicators with participants’ perceptions of neighborhoods -and the people who live in them. By employing computational tools, we can better link stratification in our social world to biases in individuals’ minds.
Please describe one or two of your most interesting projects.
My research has two arms. In the first, my collaborators and I use body camera footage as data to observe police officer interactions. Specifically, these recordings capture one of the most important tools at an officers’ disposal: their words. In both behavioral experiments and NLP analyses from transcripts, we observe disparities in officer respect, with officers communicating more respectfully towards White drivers. These interpersonal cues, in turn, shape how individuals view institutions such as police departments. We also use this footage to assess trainings to change police interactions, giving us a powerful tool to capture change as well as trends in policing.
In another line of work, I use Google Street View images to study the psychology of race and space: how the built environment informs racial biases, and vice versa. We have built a tool to sample SV images at any scale the researcher chooses, and have used it to relate neighborhood disadvantage to lay perceptions of neighborhoods. For example, sampling neighborhoods in Chicago that are higher or lower in police-community trust, we find that participants endorse more aggressive policing in the communities most in need of trustworthy guardians.
What makes you excited about your data science and AI research?
I am increasingly interested in AI as a solution to a thorny predicament in police training: how to build officers’ communication skills in a controlled but responsive setting. The majority of police trainings are of the “choose your own adventure” variety: there are a fixed set of forks in a police simulator, operated by a controller. Most of these forks lead to the use of force. The alternative is using human actors, a more fluid approach, but also one that is difficulty to scale. LLMs offer a third option: responsive simulated interactions between officer and citizen. This is exciting not just for the research questions controlled “virtual stops” can answer, but for their potential use in police training. We can take what we know about the language of stops, and have officers practice these common but consequential conversations.