Crime-Stopping AI

Share this graphic on your site!

The Hidden Social Impact Of Crime-Stopping AI

How Can AI Stops Crimes Before They Happen?

Predictive Policing
Based on “broken windows” policing which targets patrols to “hotspots” with high rates of minor crime like vandalism and theft

  • Crime Is Like A Virus: Disease and crime tend to cluster in geographic areas and spread through social networks
  • Crime Creates Crime: Minor crimes like vandalism and theft lead to more and more serious crimes

The Tech: Crime prediction software adapts existing AI models using historical crime data, including:

  • Epidemiological models used to predict the spread of disease
  • Geological models used to predict an earthquake’s aftershocks

Example: PredPol

  • Developed by LAPD & UCLA, Launched in 2008
  • Forecasts hotspots for minor crimes based on recent police reports — Targeting patrols down to a 500 ft area

Mass Surveillance
By instantly cross-referencing data from 911 calls, CCTV footage, and criminal records — AI lets police can act more quickly to stop crime

The Tech: Systems rely on machine vision to identify people and vehicles

  • Facial Recognition: Scans faces on CCTV cameras and photos to identify people — usually based on DMV records
  • License Plate Readers: Identify vehicles and track their location from camera to camera
  • ShotSpotter: Listens for gunshots or check for AR-15 Rifles and automatically alerts law enforcement with a location

Example: Domain Awareness System

  • Developed NYPD & Microsoft, Launched in 2012
  • Instantly displays information from various sources to enable faster reactions — more than 3000 cameras, 911 calls, plate readers, and more

Autonomous AI
AI-powered autonomous vehicles and robots can take over the routine tasks of law enforcement, helping reduce the growing shortfall of officers

The Tech:

Knightscope K5

  • Deployed in Huntington Park, CA in 2019
  • An autonomous robot —weighing in at 5’ 2” and 500 lbs— patrols the streets
  • Notifies police of the location of “blacklisted” individuals — using license plates, mobile device identifiers, and facial recognition

Ford’s Self-Driving Police Car

  • Patent application filed in 2018
  • Designed to detect traffic violations, pursue the perpetrator until they pull over, identify the vehicle and driver, and issue a warning or ticket

“There’s a massive opportunity for using big data to have positive social impact . . . But at the same time, we need to be aware of its limitations and be honest in terms of its performance.” — Nuria Oliver, Chief Data Scientist at Data-Pop Alliance

The Spread Of Crime-Stopping AI

Who Is Already Using Crime-Predicting AI?*

City Police Departments

  • At least 5 major cities use real-time facial recognition software
  • Over 50 police departments across the U.S.A. use PredPol
  • More are rumored to be using the tech without public disclosure
  • It’s not just for governments — UC Berkeley’s Campus Police use predictive software as well
    State & Federal Agencies
  • Immigration & Customs Enforcement, National Security Agency, Department of Defense, and more
  • The Northern California Regional Intelligence Center (NCRIC) grants access to predictive data to 300 cities in the region
  • By using an intermediary, cities avoid publicly disclosure

Around The World

  • China, Denmark, Germany, India, the Netherlands, and the United Kingdom have implemented predictive tools
  • Japan is implementing a new program in anticipation of the 2020 Olympics in Tokyo

Does Predictive Policing Work?

Los Angeles

  • In its first 21 months, PredPol resulted in 2X the arrests as professional criminal analysts — Suggesting to a greater reduction in crime
  • But a few years later, property times were back up


  • From 2015-2017, Chicago saw a surge in gun violence, bringing media attention to the city’s use of predictive policing
  • In a 2017 investigation, the New York Times found that
    • ¼ to ⅓ of the highest risk individuals were involved in a violent incident in 2016
    • Less than 20% of the city’s total gun violence involved high risk individuals

New York City

  • First instituted predictive policing in 1994
  • By 2003, murders the lowest since 1964
  • BUT, the same decrease in violent crime was seen across the country — including regions were predictive models weren’t in use
  • In 2014, the Domain Awareness System identified 103 gang members through Facebook and prison phone calls — resulting in the largest gang takedown in NYC’s history


  • Violent crime is falling, but the public hasn’t noticed
  • From 1993-2018, the majority of Americans believed that crime increased nearly every year
  • Over the same period,
    • Violent crime fell by 51-71%
    • Property crime fell by 54-69%

Roughly ½ of Americans are listed in facial databases accessible to the FBI

The Problem: AI’s Data Feedback Loop

AI’s Flaws Marginalize Minorities

  • African Americans are more likely to be included in facial recognition databases due to over-policing of black communities
  • Facial recognition algorithms* are more likely to misidentify Asians, African Americans, and Native Americans
  • In a test of Amazon’s facial recognition software, 28 members of Congress were falsely identified as criminals
    • 20% of members of congress are people of color
    • 39% of false matches were people of color

When AI Is Trained On Historical Crime Data

  • Existing bias becomes a core component in its predictive algorithms
  • Predictions may become self-fulling, as police ramp up enforcement in communities the software labels as “bad”
  • Results and enforcement ignore crimes that go unreported
  • According to the Bureau of Justice Statistics, only 43% of violent crime and 34% of property crime was reported to police in 2018
    • Why? People are less likely to report crimes they think will go unsolved

“There’s a real danger, with any kind of data-driven policing, to forget that there are human beings on both sides of the equation.” — Andrew Ferguson, Law Professor at the University of the District of Columbia and author of The Rise Of Big Data Policing

AI is here to stay in law enforcement – but at what cost?

Crime Stopping AI