AI vs. Human Transcription: Evaluating Accuracy and Meaning in Police Body-Worn Camera Footage

• In Event: Video Data Analysis for Policing Research: Innovations and Next Steps

Abstract

Body-worn cameras (BWC) play a crucial role in modern policing by providing an objective view of interactions between police officers and civilians. Law enforcement agencies across the country utilize BWC footage to review officer behavior and resolve disputes. While video recordings offer detailed accounts of these encounters, transcriptions facilitate quicker analysis of large datasets, allowing for timely insights in extensive research and policy evaluations. Despite these advantages, concerns remain about the accuracy and interpretative reliability of commonly used AI-driven transcription tools in processing vast amounts of BWC footage. This study examines two sets of transcripts—one created by AI and another transcribed by human coders—to assess meaningful differences in content, accuracy, and interpretative nuance. A key focus is whether AI captures the same meanings behind spoken words as human coders and whether reliance on AI-generated transcripts alters the researcher’s interpretation of incidents. Additionally, we explore the limitations of AI transcription, particularly in handling punctuation and tone—areas where human transcription may excel. By identifying discrepancies and potential misinterpretations introduced by AI, this research contributes to the broader discussion on the reliability of automated tools in qualitative analysis and the implications of their use in policing research.

Authors

• Savannah A. Reid, Northeastern University

• Stephen Abeyta, New York University

• Eric L. Piza, Northeastern University

• Nathan Connealy, University of Tampa

• Victoria A. Sytsma, Queen's University