Predictive Policing

Summary

The prevention of criminal activity has changed dramatically over the past two decades, largely due to the increased reliance on systems that provide crime data analysis. Created specifically for police, judicial sentencing, and prison applications, these systems conduct both predictive and retrospective analysis to aid decision making within the criminal justice system. Furthermore, these software platforms typically combine spatial informatics packages and advanced statistical features behind user-friendly interfaces. Recent studies have demonstrated problems with both the flawed logic within these systems’ algorithms and the inherent biases in the underlying data.

Module Learning Outcomes

  • Students will be able to articulate how the goals of predictive policing algorithms can differ from the impact of these algorithms on communities of color.

  • Students will be able to identify and describe the sources of potential bias in predictive algorithms.

  • Students will be able to demonstrate an understanding of the ways to check for and mitigate algorithmic bias when designing and testing predictive algorithms.

Activity 1

Understanding the Problem Through Narratives

Student Learning Objectives:

  • Students will articulate initial opinions of predictive policing algorithms.

  • Students will discuss potential sources of algorithmic bias based on initial knowledge of predictive crime algorithms.

Activity 2

Predictive Policing Algorithms use reported crime data from local police data. In the City of Oakland, CA, data is crime/arrest is classified based on geographic ‘police beat’ zones that align with one or more postal zip codes. In this activity, we want you to dig a little deeper into the type of annual crime data that would drive a predictive policing algorithm created by software companies like Lexis Nexis, PredPol, COMPAS that sell systems to police departments with the goal of efficiently identifying high risk areas in order to predict and prevent future crimes.

Student Learning outcomes:

  • Students will analyze open source policing data and SES data using measures of central tendency.

  • Students will be able to articulate thoughts on the data analysis activity and discuss the readings with other classmates in a small group setting.

  • Students will be able to differentiate areas likely to be identified as high risk and low risk by predictive policing algorithms


Narratives to Read/Listen for Activity 2

Two students from the group will work with the following readings and a podcast to characterize why the use of these predictive law enforcement systems are problematic.

Articles Folder Link (pdfs)

Last couple minutes is a good introduction to how we can think about predictive crime algorithms

Activity 3

Student Learning Objectives:

  • Students will be able to synthesize different perspectives based on their analyses and narratives discussions and demonstrate the ability to incorporate new narratives, including raising questions about the role of technologists in creating these types of predictive technologies.

  • Students will be able to articulate what they have learned about who these technologies were designed for, who they benefit, and how these technologies impact individuals and communities differently.

Instructor Notes for Activity 3

Activity Wrap-Up and Discussion: The main point of this last activity is to help students better understand the problems with the underlying assumptions about the data (i.e., that unless every single crime is reported, and unless police pursue all types of crimes committed by all people equally, it’s impossible to have a reinforcement learning system that predicts crime accurately). Instead, these predictive systems are just creating a self-fulfilling prophecy, where police find crimes in the same places they’ve been told to look for them, rather than everywhere.


While class is coming into classroom and getting ready show selected Minority Report clips (from Kinolab links) without saying much.

Clip 1: Vote Yes

Clip 2: Predetermination Happens All the Time

Clip 3: Minority Reports


Question for small groups: So would it be a good idea to implement system predicting academic integrity violations (e.g., cheating) using past data about students involved and their characteristics to reduce cheating at your college/university?

Question for small groups: How about using data on past assignment grades and student demographic factors in CS courses to predict the success of students who are allowed to take CS courses?

Whole class discuss prompts:

Why is it important for computer science students to have a deeper knowledge of this topic, as it is the software engineers, UX/UI designers, information systems engineers who are responsible for building the algorithms and applications based on existing (and perhaps flawed) statistical models?

What are some of the problematic assumptions linked to systems that identify people who are more likely to be involved in a crime with 'hotlists', which are designed to include victims and suspects involved in previous crimes, as well as their friends, and families?