Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Responsible AI in criminal justice: LLMs in policing and risks to case progression

Muffy Calder, Marion Oswald, Elizabeth McClory-Tiarks, Michele Sevegnani, Evdoxia Taka

Abstract

There is growing interest in the use of Large Language Models (LLMs) in policing, but there are potential risks. We have developed a practical approach to identifying risks, grounded in the policing and legal system of England and Wales. We identify 15 policing tasks that could be implemented using LLMs and 17 risks from their use, then illustrate with over 40 examples of impact on case progression. As good practice is agreed, many risks could be reduced. But this requires effort: we need to address these risks in a timely manner and define system wide impacts and benefits.

Responsible AI in criminal justice: LLMs in policing and risks to case progression

Abstract

There is growing interest in the use of Large Language Models (LLMs) in policing, but there are potential risks. We have developed a practical approach to identifying risks, grounded in the policing and legal system of England and Wales. We identify 15 policing tasks that could be implemented using LLMs and 17 risks from their use, then illustrate with over 40 examples of impact on case progression. As good practice is agreed, many risks could be reduced. But this requires effort: we need to address these risks in a timely manner and define system wide impacts and benefits.
Paper Structure (37 sections, 1 figure)

This paper contains 37 sections, 1 figure.

Table of Contents

  1. Introduction
  2. Research contribution
  3. Organisation of paper
  4. Background and related work
  5. From language models (LMs) to large language models (LLMs)
  6. LLMs beyond text generation
  7. Related work
  8. Criminal justice context
  9. Policing tasks that could involve LLM-based tools
  10. Community policing and offender management
  11. Intelligence
  12. Investigation
  13. Charging decision
  14. Risks
  15. Outputs
  16. ...and 22 more sections

Figures (1)

  • Figure 1: Four categories of risk of deploying a LLM-based tool: inputs, outputs, evaluation, and systems engineering.