AI audits for assessing design logics and building ethical systems: the case of predictive policing algorithms

Abstract

Organisations, governments, institutions and others across several jurisdictions are using AI systems for a constellation of high-stakes decisions that pose implications for human rights and civil liberties. But a fast-growing multidisciplinary scholarship on AI bias is currently documenting problems such as the discriminatory labelling and surveillance of historically marginalised subgroups. One of the ways in which AI systems generate such downstream outcomes is through their inputs. This paper focuses on a specific input dynamic which is the theoretical foundation that informs the design, operation, and outputs of such systems. The paper uses the set of technologies known as predictive policing algorithms as a case example to illustrate how theoretical assumptions can pose adverse social consequences and should therefore be systematically evaluated during audits if the objective is to detect unknown risks, avoid AI harms, and build ethical systems. In its analysis of these issues, the paper adds a new dimension to the literature on AI ethics and audits by investigating algorithmic impact in the context of underpinning theory. In doing so, the paper provides insights that can usefully inform auditing policy and practice instituted by relevant stakeholders including the developers, vendors, and procurers of AI systems as well as independent auditors.

A Data-driven analysis of the interplay between Criminological theory and predictive policing algorithms

“…Indeed, several studies have found this to be the case and have shown that close proximity and interactions between residents and the police can artificially inflate crime rates as the police are likely to encounter and observe more crimes than in locations that are not as heavily policed [19,35].…”

“…A social implication is that locations populated by groups that are historically vulnerable to racially-biased overpolicing can consequently become exposed to even more over-policing and disproportionately high rates of criminalisation. Racially biased policing can compound the problem since it generates crime data that prompts algorithmic preidctive models to designate their geographical regions as 'high crime' and in need of enhanced policing activity [14,16,35]. Developers of such predictive algorithms and their proponents refute this [6,24], and argue that police dispatch to a predicted high crime location does not artificially inflate crime rates and trigger a positive feedback loop.…”

“…Although a burgeoning corpus of very insightful theoretical literature on the design and application of predictive policing algorithms now exists internationally [1,7,18,22], there is a dearth of empirical analysis and consequently limited insight on the impact of their underpinning theory or design rationality. By design rationality, we mean the theoretical model/s that inform data choices and configurations, algorithmic modelling, as well as predictions [35].…”

Predictive Algorithms in Justice Systems and the Limits of Tech-Reformism

“…It is however, worth noting that the aim of debiasing is to attain data neutrality and, as such, it is techreformist given the utopic assumption that technical components of an algorithm, in this case, underpinning data, can be neutral and unaffected by data collection and processing choices and general interpretations (see also Ugwudike 2021). It ignores structural conditions such as the digital capital with which the choices, preferences, views and values of those involved in creating algorithms and/or debiasing data permeate datasets and introduce often hidden biases, even when the data appear ostensibly neutral.…”

Design of Distributed Database System Based on Improved DES Algorithm

“…Therefore, in order to prevent some important contents from being damaged and causing unnecessary loss or repeated use, it is necessary to save these key parts for viewing, calling and retrieving the application requirements of other relevant functional modules in the future query. At the same time, It ensures that users can safely access and process the required information [13][14]. Figure 1 Figure1.Distributed data base…”