Imagine a police department that can anticipate where crime will strike tomorrow, next week, or next month—and position officers there before it happens. This isn’t science fiction. It’s the promise of predictive policing, a data-driven approach that uses algorithms to forecast crime patterns and inform deployment decisions. The promise is compelling: smarter policing, fewer victims, and more efficient use of limited resources. The reality is far more complicated.
How Predictive Policing Works
At its core, predictive policing analyzes large volumes of data—such as historical crime reports, calls for service, arrest records, and sometimes contextual information like time of day or location—to identify patterns. These patterns are then used to generate forecasts. Some systems predict places (for example, identifying “hot spots” where crime is likely to concentrate), while others focus on people (such as individuals at higher risk of being involved in violence). The output becomes a kind of probabilistic map that guides patrol routes, resource allocation, and investigative priorities.
U.S. Practices
Across the United States, predictive policing has most commonly taken the form of place-based forecasting. Large cities such as Los Angeles, Chicago, and New York experimented with tools that use historical crime data to identify high-risk locations or time windows for offenses like burglary or violent crime.
The appeal is straightforward. With shrinking budgets and rising expectations, police departments are under pressure to do more with less. Predictive analytics promise to shift policing from reactive to proactive, concentrating attention where the data suggest it’s needed most. In theory, this means fewer crimes, faster response times, and communities made safer through strategic intervention.
But theory and practice often diverge.
The Bias Problem: Garbage In, Bias Out
Civil rights organizations and researchers have raised a fundamental concern: predictive policing systems are only as good as the data they’re trained on. And in the U.S., those data carry the weight of history.
Scholars have highlighted how predictive algorithms trained or used with historical data could reinforce deep-rooted implicit biases (Arcas, 2017; Lum & Isaac, 2016) and structural inequity (Ferguson, 2014). Richardson et al. (2019) warn that predictive policing systems often rely on data generated during periods of “dirty” policing—marked by racial bias, unfair treatment, and even illegal law enforcement practices.
Historical crime data do not reflect where crime happens—they reflect where police have been and whom they’ve arrested. Decades of over-policing in communities of color mean those neighborhoods generate disproportionately more reports, stops, and arrests. When algorithms learn from these data, they don’t predict crime; they predict policing patterns.
Conclusion
Predictive policing can be a useful instrument—but without strong institutional safeguards, it risks becoming a digital Pandora’s box, potentially scaling existing inequalities while imposing significant costs on public trust. We have opened this box, and it is up to us to ensure it is utilized responsibly.
References
Arcas, B. A. y. (2017, May 20). Physiognomy’s New Clothes. Medium. https://medium.com/@blaisea/physiognomys-new-clothes-f2d4b59fdd6a
Ferguson, A. G. (2014). Big Data and Predictive Reasonable Suspicion. University of Pennsylvania Law Review, 163(2), 327–410.
Lum, K., & Isaac, W. (2016). To Predict and Serve? Significance, 13(5), 14–19. https://doi.org/10.1111/j.1740-9713.2016.00960.x
Richardson, R., Schultz, J. M., & Crawford, K. (2019). Dirty data, bad predictions: How civil rights violations impact police data, predictive policing systems, and justice. NYUL Rev. Online, 94, 15.