Discerning Scripted Narratives in Use of Force Reports: Using AI to Read Historical Data

This paper discusses the process and preliminary findings of a scholarly partnership with Dr. Ashique KhudaBukhsh, a software engineer, and his graduate students who designed AI software to scrape and analyze large amounts of historical text in order to predict the political persuasion of the texts' author. Using Dr. KhudaBukhsh’s software, more than 6,000 Subject Resistance Reports (SSRs) spanning five years, were analyzed to discern how officers write narratives about their use of force encounters with civilians in Rochester, NY­–a city with a lengthy history of both police violence and movements for police accountability. The purpose of the research is to better understand how officers present themselves in these reports and detect any patterns or “tells” that might give away the demographics of the subject in the encounter. For instance, I ask are there scripted ways of describing use of force encounters with civilians of different and intersecting demographic identities (e.g. race, gender, ability, age, location, medical issues, mental health issues). Further, have these scripts changed over time? I also determine if the type and severity of force causes officers to write their reports in specific ways. SRRs have been used by the department since the 1970s and are a part of the documentation presented at arraignments where judges determine whether or not defendants should be released, remanded, or be allowed to post bail. Such analysis might provide insight into how historical shifts in policing and public pressure shape the truth and veracity of officer statements.