Introducing 

Prezi AI.

Your new presentation assistant.

Refine, enhance, and tailor your content, source relevant images, and edit visuals quicker than ever before.

Loading…
Transcript

AI in Predictive Policing

Predictive Policing

#Predictive Policing is not a new Concept

Introduction

Crime 30 - 40 %

ERT 15-20%

Predictive Policing using AI

Predictive Policing Using AI

“ Predictive policing means the use of historical data to create a forecast of areas of criminality or crime hot spots, or high-risk offender characteristic profiles that will be one component of police resource allocation decisions. The resources will be allocated with the expectation that, with targeted deployment, criminal activity can be prevented, reduced, or disrupted. ”

Cycle and Challenges

Predictive Policing - Cycle and Challenges

Predictive Policing Cycle

Challenges

Robustness

&

Generalization

Technological Positivism and Objectivity

Accuracy, Bias, Transparency and Accountability

Data Accuracy

Data accuracy

  • Data are at the core of law enforcement activities
  • Plate numbers, criminal records, stop & search
  • ML model to facilitate data analysis

What if...?

  • Data are inaccurate
  • Unreliable analyses
  • Unreliable historical data

Examples

information gathering

  • Crime misclassification
  • Under-reporting/non-reporting
  • Missing information

Examples

Crime reporting

  • Eyewitness testimonies
  • Under-reporting/non-reporting

Data inaccuracy

and

biases

  • Inaccurate data
  • Inaccurate archival data
  • Potential biases

Bias in different stages

Stage 1: Selection Bias: which crimes are selected?

Bias in PP

Training Data

Training Data

  • Heavily biased due to centuries over racism and over-policing
  • Communities over/under-represented in data
  • During Deployment:
  • Racist data reinforced.
  • Further over-policing.

“Future policing is predicted, not future crime” (Babuta & Oswald, 2019)

Deployment I

Deployment I

Racial Bias: mislabeling black defendants

Example: US: Northpointe (ProPublica, 2016)

Deployment II

Deployment II - Automation Bias

Judges

Police Officers

Conclusion

Conclusion: biased data hinders progress

ML misses two factors due to temporal stationarity:

  • Changes in individual circumstances.
  • Changes in broader society.

--> Prevents the creation of a less biased system

Transparency

& HR

Transparency

  • black-boxes
  • law enforcement
  • developers
  • by PRIVATE SECTOR

→development ≠ implementation

  • accountability

vs Fundamental Rights

example: risk-scoring systems

  • arbitrariness
  • vs presumption of innocence & due process
  • bulk unconsented collection of data
  • confidentiality
  • data protection

However

  • data secrecy
  • companies’ intellectual proprietary rights
  • AI algorithms through ML tools

Accountability

Accountability

Being responsible for your actions/decisions and being expected and able to explain them

  • Multi-faceted, operates on many levels, involvement of several actors
  • Effective accountability: information, explanation or justification, and (the possibility for) consequence
  • information: opaqueness, complexity, secrecy and proprietary nature of algorithms
  • explanation: lack of explicability
  • consequence: relies on information and explanation, risk of delegation or shifting authority
  • Limitations
  • Technology won't save law enforcement
  • Difficulties of policymakers

Case Study

Palantir Gotham

  • Founded in 2003 by Alex Karp, Peter Thiel and Team
  • Palantir Gotham is an AI powered Predictive Policing application
  • Instrumental in discovery of GhostNet and Shadow Network
  • Used by USIC,DoD, Federal and state CT cells and of course police.

Palantir Gotham

  • NOPD investigated
  • 2011 report 'dirty data'
  • Inaccurate data and biases

  • Also used by LAPD - indiscriminate database
  • Chronic Offenders under LASER program

Transparency and accountability

  • secrecy and proprietary rights - US Computer Fraud & Abuse Act
  • complexity of products
  • impact on civil liberties
  • transparency and accountability of process/partnership
  • Palantir & NOPD - secret contractual relationship
  • unclear contract terms from the start

→ transparency and accountability of ACTORS

Learn more about creating dynamic, engaging presentations with Prezi