Paper short abstract:

Drawing on observations from an ongoing project, and comments from members of this consortium, this presentation reflects on issues of democratic accountability and the complexities of integrating different work logics and practices into the development of predictive policing tools.

Paper long abstract:

This presentation contributes to the long tradition of predictive policing studies by focusing on the early stages of developing AI tools, and in particular on computer science engineers and programmers. It introduces an ongoing R&D project and explores the complexities of integrating different work logics and practices into the development of predictive policing tools. Not only does this project involve the development of algorithms at different levels of maturity, but it also brings together engineers and developers, members of different police departments, legal practitioners, and social sciences and ethics scholars with their associated practices and worldviews. Moreover, as a publicly funded development, it carries with it the additional expectation of societal contribution, and transparency and explainability of its results.

References to predictive policing are abundant, as the vast array of literature shows (Egbert & Leese, 2021; Kaufmann, Egbert & Lesse, 2019), but there is less emphasis on the socio-technical and procedural aspects of the development of these systems (Fest et al., 2023). Although studies such as those by Lally (2022) or Widder & Nafus (2023) emphasise how the abstract knowledge practices typical of the software industry contribute to the compartmentalised and isolated developments that AI systems entail, how accountability is, or can be, enacted in the development of AIs is a missing element. Drawing on observations from an ongoing project, and comments from members of this consortium, this paper reflects on how the need for democratic accountability can be embedded in the project and its members’ actions and work.