The LAPD announced on Tuesday that researchers will utilize artificial intelligence to analyze the tone and word choice utilized by officers during traffic stops. This study aims to investigate if police language can sometimes unnecessarily escalate public encounters. Conducted by researchers from USC and other institutions, the findings will contribute to the training of officers in effectively managing interactions with the public and promoting accountability. In a meeting with the Board of Police Commissioners, Cmdr. Marla R. Ciuffetelli of the Office of Constitutional Policing & Policy stated that while machine learning is still in its early stages, it will undoubtedly become a significant aspect of officer training in the future. Over a span of three years, researchers will review body camera footage from approximately 1, 000 traffic stops. They will then establish criteria for what constitutes an appropriate interaction based on feedback from the public and officers, as well as a review of the department's policies. Benjamin A. T. Graham, an associate professor of international relations at USC and one of the study's authors, explained that these criteria will be inputted into a machine learning program. The program will "learn" how to independently review videos and flag instances where officers cross the line. Graham acknowledged that certain standards are subjective, and different researchers may have varying observations of a single interaction, such as whether the officer introduced themselves. During the analysis, researchers will consider factors like the location of the stop, the driver's race, and the officer's rank, age, and experience. Care will be taken to anonymize both officers and subjects. The study also involves researchers from Georgetown, UC Riverside, and Texas universities. Commissioner William Briggs, a prominent supporter of the study, expressed hope in its potential but raised concerns about establishing "safety valves" for increasingly powerful AI systems like ChatGPT. The need for regulation of AI technology has been highlighted, with President Biden acknowledging the risks it presents, including the spread of disinformation and job losses.

A group of scientists and executives has also warned about the potential dangers of AI to humanity, although regulations have been slow to develop. Unlike other major police agencies, such as the NYPD, the LAPD currently lacks a dedicated unit for auditing the vast amount of body camera footage collected from police encounters each month. The department primarily reviews footage in cases involving force or when a personnel complaint is filed. Officer rudeness is a common public complaint. While LAPD policy prohibits aggressive or profane language and does not endorse so-called "tactical language" to gain compliance, its use is sometimes tolerated, and disciplinary measures are rarely imposed. At the Academy, LAPD police cadets are trained on the impact of their words and demeanor during encounters, with de-escalation emphasized. The USC study follows similar efforts by researchers at Stanford and the University of Michigan. The Michigan study played audio recordings from routine traffic stops to participants who were asked to assess officer interactions with motorists without knowing their race. The report revealed that officers communicated in a more respectful and friendly manner with white men, while displaying less positive attitudes towards Black men. The study also found that the beginning of the conversation, specifically within the officer's first 40 words, significantly influenced the trajectory of the encounter and whether it resulted in a search or arrest. The LAPD's Inspector General's office is conducting its own study on officer language use. During the presentation, department officials did not specify if the study's findings would only be utilized for training purposes or if any misconduct discovered could lead to disciplinary actions, a point that the police union is likely to oppose. Cmdr. Ciuffetelli stated that the study would likely proceed in four phases and its findings would eventually inform the department's training models. Graham estimated that data gathering would take approximately six months, with preliminary results expected after the first year of the study. In response to a question from Commission Vice President Rasha Shields, Graham confirmed that a "corollary analysis" would be conducted on the actions of the individuals being stopped and how their behavior might also influence the outcome of the encounter, noting that most of the questions revolve around officers' speech.