How Policing Agencies Use AI

AI is transforming policing, sometimes in dramatic ways. Face recognition, predictive policing, and location-tracking technologies — once the stuff of science fiction — now are being adopted by law enforcement agencies large and small.

This explainer gives an overview of some of the ways police are using AI to investigate and deter crime, including:

  • Identifying unknown individuals or verifying their identity;

  • Tracking people’s locations and movements;

  • Detecting crime, anomalies, or suspicious events;

  • Predicting future crimes, perpetrators, and victims;

  • Analyzing emotions, including deception;

  • Determining associations between individuals; and

  • Managing and analyzing evidence.

The effectiveness of some of these tools is unclear, and the inclusion of a tool on this list is not meant to indicate that it performs well. This document also does not evaluate the benefits or harms of these tools, which will be addressed in separate explainers. Rather, this document is meant to explain how police are using AI and give a sense of the breadth of such uses.

Identification

AI systems can be used to identify individuals or verify their identity.

Face recognition is a computer vision technology that analyzes faces in an image. It can be used for things such as face identification (the identification of an individual based on a comparison with a pool of known individuals) and face verification (verifying that a given face corresponds to a specific person — for example, verifying that a person’s face matches the photo on their identification card).

Iris recognition identifies individuals by their iris patterns. A specialized camera takes an image of the boundaries and textures of the iris, then maps the iris image using over 200 distinct features.

Automated fingerprint identification has been in use for decades; now, AI systems can be used to enable better matching even when a fingerprint is distorted or incomplete. AI also is being used to develop new systems which can take a person’s fingerprint without physical contact.

Palm-print identification, like fingerprint identification, is accomplished through analysis of ridges and valleys on the skin’s surface. Some claim that this technique has advantages over face recognition technology — for example, palms have more details to tell one person from another, and it is harder to scan a person’s palm without their consent.

Ear biometrics can be used to identify individuals who are difficult to identify through face recognition technology — for example, due to the individual wearing a mask.

Gait recognition analyzes how people walk in order to identify them. It has advantages over other biometric systems, as it enables the identification of persons from a distance. Notably, accuracy is a serious issue due to the variability of environments and human bodies.

Voice recognition systems are used to determine the identity of a person based on audio of their voice. These systems use specialized models known as acoustic models to process and analyze audio files.

DNA analysis has long been used by law enforcement to identify suspects. Now, AI is being used to improve this process and make it more efficient. New AI-powered forms of DNA analysis are now being developed, such as forensic DNA phenotyping, which attempts to predict externally visible characteristics such as eye, hair, and skin color, as well as the geographic origins of a person’s ancestors.

Tracking

Policing agencies use AI systems to track the locations or movements of individuals.

Tracking algorithms can detect objects and/or individuals in video files and track them across cameras based on the appearance, velocity, and motion of the thing being tracked. This feature could be used, for example, to search stored video footage from a particular neighborhood and identify all the times that a given individual was recorded.

Vehicle-surveillance systems, also known as automated license plate readers, detect information about passing vehicles, such as a vehicle’s color, make, and license plate number. This data can be stored, along with the location and time of capture, thus enabling police to ascertain the locations of vehicles over time. Some agencies now are using drone-based vehicle-surveillance systems.

Detection

Policing agencies use AI to detect crime, anomalies, or suspicious events.

Anomaly detection seeks to identify events or data points that are anomalous — that is, that deviate from what is expected. This technology is widely used by the private sector— for example, by financial institutions to detect fraudulent transactions or by network administrators to detect cyberattacks.

Some vendors have developed systems designed to alert policing agencies to events such as shoplifting, fights, loitering, dangerous driving, and casing a location. At least one vendor is leveraging vehicle surveillance system data to try to identify driving patterns that may be associated with drug trafficking activity or other unlawful conduct.

Gunshot detection systems use a network of outdoor acoustic sensors to detect and locate gunfire and alert police. Policing agencies use gunshot detection systems to reduce response times, in the hope of locating a shooter, getting help to victims, or finding evidence such as shell casings. New systems use two-source detection — sound and flash — to confirm gunshots.

Weapons detection systems are used to identify the presence of weapons. Computer vision-based systems analyze images to detect objects that appear to be weapons. Other vendors use sensors and analytics to detect concealed weapons and identify their location — an alternative to traditional metal detectors.

Drug detection systems are being used to detect drugs on-site using mobile spectrometers, as opposed to sending samples to a lab.

Prediction

Policing agencies use AI to try to predict the location and time of future crime, as well as those who may perpetrate or be the victims of it.

Place-based predictive policing systems use historical crime data to identify areas prone to crime, and at what times. Systems also can analyze geographic features that increase the risk of crime, known as risk-terrain analysis.

Person-based predictive policing systems seek to identify individuals who are at risk of committing crimes or becoming a victim. This can be based on data such as one’s risk factors for violence or becoming a victim, and/or their frequenting high-crime locations.

Recognizing Emotions

Policing agencies are experimenting with AI systems to analyze an individual’s sentiments or emotions.

Lie detection systems claim to track eye movements and analyze micro-expressions to determine whether an individual is engaged in deception. Some systems are designed specifically for law enforcement use.

Sentiment analysis is a natural language processing technique designed to classify individuals’ sentiment as positive, negative, or neutral. Affective computing, which goes beyond sentiment analysis, seeks to understand and interpret specific emotions based on facial expressions, voice intonations, text, and physiological signals. Sentiment analysis/affective computing might be used, for example, to flag problematic police interactions captured on bodyworn cameras for supervisor review.

Identifying Associations

Policing agencies use AI systems to help detect associations among individuals.

Convoy analysis is a feature for Vehicle Surveillance Systems, or License Plate Readers, that identifies vehicles that travel together, and thus presumably are associated with one other. They allow officers to enter a license plate number and search for related vehicles.

Social network analysis tools suggest how individuals are connected in society, visualized through graphs. For example, AI tools have been used to identify alleged associates based on social media data. Machine learning algorithms are used to identify patterns, trends, and anomalies in social networks.

Evidence Management and Analytics

Policing agencies use AI to help agencies find potentially relevant evidence in large datasets.

Automated metadata tagging can automatically tag and label digital evidence, helping investigators to find relevant evidence in the future. Some body-worn camera systems use AI to tag and label videos with relevant contextual information, helping police locate specific events within large video databases.

Evidence matching tools automatically search an agency’s databases to find evidence that might be related to an incident under investigation.

CSAM detection tools detect and flag the existence of child sexual abuse material (CSAM) on devices, helping police to locate such materials and identify victims more quickly.

Transcription tools can be used to transcribe audio automatically from video and audio files. This enables agencies to search for keywords across potentially thousands of videos.