Predictive policing

Predictive policing refers to the use of specialized artificial intelligence (AI) or analytical software to predict the likely occurrence of crimes. In some instances, predictive policing is used to predict how likely it is that a previous offender will reoffend. The software then recommends that the police heavily monitor that individual. In other instances, predictive policing software monitors specific areas, then recommends that police increase patrols in regions where higher instances of crimes are likely to occur. In order to accomplish these goals, predictive policing software uses complex algorithms to sort through massive amounts of data and establish patterns. It then applies those patterns to events as they unfold, using them to make predictions.

Some people have criticized predictive policing software, arguing that it could become biased. They worry that such a bias could cause innocent people to be arrested. Others criticize the software on its effectiveness record, arguing that, when tested, most software has not outperformed human intuition.

Background

Artificial intelligence (AI) refers to using computer systems and algorithms to simulate the intelligence and decision-making capabilities found in biological organisms. In the early twenty-first century, advances in machine learning—in particular, deep learning and large language models—enabled AI systems to generate text, images, code, and other output with increasing sophistication. Though contemporary AI still did not possess human-level understanding or consciousness, its capabilities had expanded greatly by the mid-2020s, and researchers continued to develop ever-more complex AI systems.

AI systems are trained to emulate specific processes more commonly carried out by biological organisms. Most prominent among these processes is learning, meaning the process of acquiring and utilizing new information. Additionally, AI commonly emulates reasoning, or using rules to reach conclusions, and self-correction, or assessing data and fixing inaccuracies.

Some introductory frameworks divided AI into four types: reactive machines, limited memory, theory of mind, and self-awareness. The first type, reactive machines, can identify objects and make predictions. It is able to analyze and make decisions to achieve a goal. However, it has no memory and cannot effectively use its past experiences to improve its decision-making process. The second type of AI, limited memory, allows the computer to temporarily store decisions in order to react to them. For example, if a self-driving car observes another vehicle engaging its turn signal, it will retain that information and monitor for the other vehicle to change lanes. However, limited-memory AI systems do not store data long-term.

The third type of AI, theory of mind, is a program that understands that other beings have beliefs and intentions that impact their decisions. It then accommodates for such variation in its decision-making process. Researchers and scientists are still working on developing such a system. The fourth and final type of AI, self-awareness, involves a machine that is conscious and aware of its own existence. Scientists are also still working to develop this type of system, and some are concerned about the ethics of working toward its creation.

Overview

Predictive policing refers to the use of complex algorithms and AI tools to predict where crimes will occur. This allows police to better utilize their resources, focusing on problem areas and preventing crime instead of reacting to crimes as they happen. Such software is controversial, and it is not in use in every police department. Despite this, many police officers believe that such techniques make them more effective at reducing crime rates.

Different types of predictive policing software produce different types of data for police departments. Some types of predictive policing software, such as Operation LASER, track the information of any registered offender within two years of the present date. It then tracks their various offenses, assigning severity values to each interaction with police. For example, the program would note that a person has been on parole, used to belong to a gang, and has been recently stopped by police. Each of those instances would increase the person's score. If a score hits a certain point value, the program recommends that police forces closely monitor the person by placing them on the Chronic Offender Bulletin. The predictive policing software has independently assessed that such an individual is likely to reoffend in the future. Some models of this type of software independently analyze arrest records to form patterns, giving them even more information to determine the probability that an individual might reoffend in the future.

Other types of predictive policing software monitor locations instead of people by analyzing incidences of petty crimes and violence in an area, then checking them against established patterns to warn police officers when crime in an area passes a preset threshold. Police officers can then devote extra resources to patrolling the area, reducing the crime rate.

Most predictive policing software uses some of the six established methodologies to make predictions. Regression methods map statistical correlations between variables that might influence crime rates. Hot-spot analysis predicts areas with a high likelihood of crime based on the police department's crime data. Data-mining techniques analyze large quantities of data to establish patterns, then use those patterns to predict future criminal incidents. These techniques commonly utilize complex algorithms to sort through and classify data. Risk-terrain analysis accounts for the various sociological factors that correlate with higher crime in an area. Near-repeat methods track data for specific crimes, then use that data to predict when and where those crimes might occur in the future.

Some critics of predictive policing worry that the software might become biased. They worry that if such an occurrence were to happen and police or judges put too much stake in the conclusions of the software, it might result in innocent people being convicted. Additionally, current generations of predictive policing software have been criticized for ineffectiveness. In some studies, predictive policing software performed no better at predicting crimes than did human reasoning. In the UK, Amnesty International publicly called for a ban on predictive policing in 2025, citing that the practice is racist and dangerous. Meanwhile, some organizations, such as the US Department of Justice, have emphasized the need for ethical governance and safeguards around the use of AI in law enforcement to counteract such critiques.

Bibliography

"Aided by Palantir, the LAPD Uses Predictive Policing to Monitor Specific People and Neighborhoods," The Intercept, 11 May 2018, theintercept.com/2018/05/11/predictive-policing-surveillance-los-angeles/. Accessed 5 Sept. 2019.

Dodd, Vikram. "UK Use of Predictive Policing Is Racist and Should Be Banned, Says Amnesty." The Guardian, 19 Feb. 2025, www.theguardian.com/uk-news/2025/feb/19/uk-use-of-predictive-policing-is-racist-and-should-be-banned-says-amnesty. Accessed 10 Mar. 2026.

George, Rohan. "Predictive Policing: What Is It, How It Works, and Its Legal Implications," The Centre for Internet and Society, 24 Nov. 2015, cis-india.org/internet-governance/blog/predictive-policing-what-is-it-how-it-works-and-it-legal-implications. Accessed 10 Mar. 2026.

Hickman, Blair. "What You Need to Know About Predictive Policing," The Marshall Project, 22 Feb. 2016, www.themarshallproject.org/2016/02/21/what-you-need-to-know-about-predictive-policing. Accessed 10 Mar. 2026.

Lapowsky, Issie. "How the LAPD Uses Data to Predict Crime," Wired, 22 May 2018, www.wired.com/story/los-angeles-police-department-predictive-policing/. Accessed 10 Mar. 2026.

Lau, Tim. "Predictive Policing Explained." Brennan Center for Justice, 1 Apr. 2020, www.brennancenter.org/our-work/research-reports/predictive-policing-explained. Accessed 5 Feb. 2025.

Pearsall, Beth. "Predictive Policing: The Future of Law Enforcement?" National Institute of Justice, 22 Jun. 2010, nij.ojp.gov/topics/articles/predictive-policing-future-law-enforcement. Accessed 10 Mar. 2026.

Rieland, Randy. "Artificial Intelligence Is Now Used to Predict Crime. But Is It Biased?" Smithsonian Magazine, 5 Mar. 2018, www.smithsonianmag.com/innovation/artificial-intelligence-is-now-used-predict-crime-is-it-biased-180968337/. Accessed 10 Mar. 2026.

"Safety and Justice Program," RAND, 2013, www.rand.org/content/dam/rand/pubs/research_reports/RR200/RR233/RAND_RR233.sum.pdf. Accessed 5 Sept. 2019.

Siegel, Eric. "How to Fight Bias with Predictive Policing," Scientific American, 19 Feb. 2018, blogs.scientificamerican.com/voices/how-to-fight-bias-with-predictive-policing/. Accessed 10 Mar. 2026.

US Department of Justice. Artificial Intelligence and Criminal Justice Final Report. 3 Dec. 2024, www.justice.gov/olp/media/1381796/dl. Accessed 10 Mar. 2026.