Machine Learning and the Police: Asking the Right Questions

Abstract

How can we secure an accessible and open democratic debate about police use of predictive analytics when the technology itself is a specialized area of expertise? Police utilize technologies of prediction and automation where the underlying technology is often a machine learning (ML) model. The article argues that important issues concerning ML decision models can be unveiled without detailed knowledge about the learning algorithm, empowering non-ML experts and stakeholders in debates over if, and how to, include them, for example, in the form of predictive policing. Non-ML experts can, and should, review ML models. We provide a ‘toolbox’ of questions about three elements of a decision model that can be fruitfully scrutinized by non-ML experts: the learning data, the learning goal, and constructivism. Showing this room for fruitful criticism can empower non-ML experts and improve democratic accountability when using ML models in policing.

Conceptualizing an AI-based Police Robot for Preventing Online Child Sexual Exploitation and Abuse:

“…Two aspects are particularly important concerning the training data: First, the quality and amount of data influence the learning model (Vestby & Vestby, 2019). A model developed on data sets generated by real-life activity (e.g., real chat logs between offenders and child victims), will be different from a model developed on manufactured data sets mimicking real-life activity (e.g., the data in the study by Meyer (2015), see Section 2.3).…”

“…Domain expertise is important when developing a tool to ensure that relevant features of the domain in which it shall be used are taken into account. PrevBOT should therefore be developed on basis of robust knowledge and data reflecting the particular characteristics of online CSEA (Kaufmann, 2017;Vestby & Vestby, 2019). Considering that the Authorship Analysis researchers' primary field of expertise is computer science/informatics, it might be a weakness that their assessments relating to CSEA are based upon research from other areas, such as cognitive and social psychology, or criminology.…”

AI, Trustworthiness, and the Digital Dirty Harry Problem

“…At first, the police used AI for knowledge management purposes (Alzou'bi et al, 2014;Vestby & Vestby, 2019). Today, AI is crucial to many police forces in performing order maintenance, criminal investigations and crime prevention through systems like PredPol, Valcri or Palantir.…”

Trust for sale

“…The participatory design community has suggested a number of methods for including societal values in AI infrastructures (Loi, Lodato, Wolf, Arar, & Blomberg, 2018). Norwegian scholars have proposed methods for including non-experts in technical discussions about the design of AI for policing (Vestby & Vestby, 2019). These approaches incorporate subjective, human engagement in AI processes, and represent the best mechanisms for ensuring that AI systems manifest the values of trust, transparency and openness espoused by the Nordic Strategies for AI (Robinson, 2020).…”