Broadcasting police communications can pose privacy risks, especially for Black men

Police radio transmissions contain personally identifiable information and may pose a privacy risk to the public, especially black men, according to a new study by researchers at Penn State University and the University of Chicago. be.

“This study provides insight into police activity as events unfold,” said Shomil Wilson, associate professor of information science and technology at Penn State University and co-author of the study. . “We found that because black suspects are disproportionately involved in police radio communications, these communications pose a proportionately higher privacy risk for black people.”

The researchers analyzed human-transcribed and annotated police broadcast communications sent in three Chicago dispatch zones, areas used to coordinate police operations, for a total of 24 hours a day. I investigated. According to U.S. Census data, one zone was majority non-Hispanic white, one was majority Hispanic, and the other was majority non-Hispanic black. The researchers found that men were mentioned nine times more often than women in police broadcasts, and black men were mentioned the most of all groups, even in predominantly white areas. did.

The researchers presented their findings at the 27th Society for Computing Machinery Conference on Computer-Supported Collaboration and Social Computing, held in Costa Rica from November 9-13. The team received recognition from the conference’s awards committee for diversity, equity and inclusion.

“Typical police radio transmissions are short and serve a coordinating purpose, such as, ‘Car 54, where are you?'” said the University of Chicago research assistant professor and co-author of the study. said Chris Graziul, one of the researchers. “These transmissions seek to communicate what is happening and who is involved. In the process, sensitive information is often disclosed.”

The researchers obtained 9,115 transmissions, which they called “utterances,” that occurred when police and dispatch agencies communicated over radio broadcasts. They manually transcribed the submissions and randomly selected 2,000 utterances across the three zones for further analysis. They developed a qualitative annotation scheme to label text. They classified the annotated data into six categories. The scope of this ranges from event information such as “house alarm intrusion” and “traffic stop,” to procedural transmissions such as “Car No. 54,” to casual transmissions such as “Good morning, squad.” It will span. ”

Researchers found that event utterances contained the most references (approximately 60%) to gender, race/ethnicity, age, and protected health information that could be used to identify an individual. I discovered it. Almost 68% of utterances that included sociodemographic indicators used sexist terms for men, and about 69% of those utterances referred to black people, the researchers said.

“Our findings contribute to more evidence on racial disparities in policing. What’s new here is the data source: radio communications,” Graziul said. “Despite its heavy use in police systems around the world, few have investigated what this means of communication can tell us about how policing actually works.

“The disproportionate mention of Black people reflects a new way of looking at how police officers’ attention is unevenly distributed across racial/ethnic groups, and identifying this disparity is , helps us understand the challenges to the ethical use of this data source for research, such as preventing the leakage of sensitive personal information.”This information impacts Black communities more than other communities. Possibly. ”

After studying the utterances, the researchers tested a widely used artificial intelligence tool, large-scale language models (LLMs), to confirm its ability to find personal information in the transcripts. Despite the unique nature of the language involved in police broadcast communications, LLM detected personally identifiable information with high accuracy, highlighting the risk of privacy vulnerabilities. Researchers say identity thieves and other criminals could use AI technology to quickly find and exploit personal information contained in police radio activity records.

“This study reveals an alarming trend in racial inequality when it comes to the exposure of classified information during police radio transmissions,” said the doctoral student in informatics in the School of Information Science and Engineering. said Pranav Narayanan Venkit, a graduate student and lead author of the paper. paper. “This study will help researchers and developers think more about the interactions between LLMs and different parts of society (police communities, ethnic minorities, and various other groups) to identify bias and protect personal information.” It may be helpful.”

Miranda Goodman, who graduated from Penn State this summer with a bachelor’s degree, and Samantha Kenney, a senior at Penn State, also contributed to the study.