"Hey, man," says the officer sauntering up to your car. The nonchalant greeting might seem insignificant - but it's not. If you're white, that police officer is statistically more likely to lead with "Hello, sir."
Jennifer Eberhardt, a social psychologist at Stanford University, heads a team of computational linguists, engineers and computer scientists, which is developing speech-recognition and transcript-analysis software for policing. Using machine intelligence, the system scans transcripts from body-camera footage to recognise patterns of racial disparity.
With African Americans: man, warrant, your ID, listen, ain't, gonna, plate, passenger, sticker, paperwork, have insurance, driving without, probation, suspended, tow, you got, hold, hey
With white people: sir, registration, your headlights, hello, actually, oh, OK, try, help, sign, red light, citation, contest, the fine, I just, the reason, questions about, safely, can go
Eberhardt refined her computational tools during a two-year study of the Oakland Police Department, scrutinising more than 36,000 statements from traffic stops. The resulting data found that officers were more likely to ask questions of black drivers, less likely to state a reason for pulling them over and more likely to use less respectful and abrupt language. In a data set representing an equal number of encounters with black people and white people, the model could predict the driver's race 68 per cent of the time, based solely on the officer's language.
"We're looking at police body-cam footage because we want to improve, not just tense encounters, but everyday interactions," says Eberhardt. Such analysis on a wider scale "could be industry-changing", Eberhardt says, giving police supervisors the means to identify patterns of slanted speech and conflict escalation - before the interactions make the headlines. Video analysis of police body language is likely to be the next step.
This story was originally published in the US edition of WIRED magazine
This article was originally published by WIRED UK
