Police Scotland practitioners advocate for systemic overhaul in data science integration

The world is seeing an increasing recognition of the potential for artificial intelligence and data science to be utilised in historically ‘human’ decision-making.

At the same time, rising prison and probation populations, alongside budgetary cuts, mean some criminal justice professionals and law enforcement agencies are now turning to data science in an attempt to maintain service quality with fewer resources.

DPhil candidate Helen Kosc recently collaborated with scholars from machine learning and data science backgrounds to study the merging of these two trends.

Along with Dr Miri Zilka (University of Cambridge), Caitlin Kearney (Technical University Munich) and Jiri Hron (Google DeepMind), Helen investigated how strategic decisions are made about where, when and how to employ data science in policing.

The paper was presented at the Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency in June 2024.

With increasing demands and strained resources, there is a growing reliance on data science and AI to solve problems of justice. So far, this technology has been utilised in tools which aim to forecast when and where crime will occur; predict the risk of certain individuals re-offending; as well as automate facial recognition.

However, to date there has been very little research into the potential benefits or risks of such applications. In fact, past deployments of this technology have been criticised due to their low accuracy and bias, by researchers and journalists alike.

Alongside these growing concerns, Helen spoke to 40 practitioners from Police Scotland in a series of in-person and online workshops, investigating what they believe to be crucial to successfully incorporating data science into their ways of working.

Helen (centre) at Police Scotland Headquarters in Glasgow

 
The participants generally agreed that the utilisation of data science should be increased within the organisation. However, the current uses of such technology required a major overhaul.

Bucking the external trend, the practitioners distanced themselves from tools like facial recognition and risk assessment.

Instead of focusing on individual use-cases, their primary concerns for the future centred around:

(i) systemic issues around data collection and use, with current tools perceived to be unreliable, untrustworthy and lacking nuance.

(ii) goal misalignment between leadership and operational levels - in the past, an over-reliance on metrics led to the undermining of officers' perceived autonomy, as well as damaging public trust.

(iii) the fear that datafication may undervalue important aspects of policing, for example, by decreasing preventative work and increasing administrative burden. 

(iv) appropriate ways of interaction between data science teams and operational officers, including ensuring consistent officer consultation from the outset.

Participants noted that a successful incorporation of data science within their organisation should include simple and efficient tools, where any limitations were well-documented and circulated. The need for training before the use of any such technology was emphasised.

Alongside the insights particular to Police Scotland, this work reaffirms how participatory approaches can go beyond the technical, and uncover structural and political barriers to success.