A Taxonomy of Bullshit: Classifying AI Truth-Indifference in Policing Contexts

May, B, Nunan, J, Marshall, J and Palace, M ORCID: 0000-0003-3016-2118 A Taxonomy of Bullshit: Classifying AI Truth-Indifference in Policing Contexts. Policing: A Journal of Policy and Practice. (Accepted)

Text A Taxonomy of Bullshit- Classifying AI Truth-Indifference in Policing Contexts.pdf - Accepted Version Access Restricted Available under License Creative Commons Attribution. Download (365kB)

Abstract

Since 2022, Generative AI has seen rapid development across sectors, and more recently deployments in policing, including report writing, administrative automation, intelligence analysis, investigations and crime scene investigations. However, there remains a lack of understanding toward (i) how these systems systemically fail, and (ii) the threats this can create within evidentiary workflows, and criminal justice as a whole. Current discourse typically associates these failure modes under the umbrella term hallucination; however, such anthropomorphism is inappropriate and disregards the nuance of how these systems might fail on a human-AI interactionalist, diagnostic level. Using a structured approach to conceptual analyses, this article develops a more precise framing grounded in Frankfurt's concept of bullshit and synthesises existing technical, philosophical, and policing-relevant literatures into three categories of error: factuality errors, faithfulness errors, and task-specific errors. The central contribution was to map each category to the minimum review resource required to detect it: external verification, source-output comparison, and domain-expert evaluation respectively. This approach to detectability carries direct operational consequences for how policing should allocate review effort and decide which use cases are appropriate for current LLM systems. In addition, this article argues that no single approach can mitigate these issues. Specifically, factuality errors can be detectable via verification, whereas faithfulness and task-specific errors require a motivated, domain-specific expertise. In many cases, the latter two categories presuppose that policing has the sufficient resources, time and expertise required, which operational settings rarely afford. This taxonomy provides a working framework for practitioners, but is subject to refinement in future research.

Item Type:	Article
Uncontrolled Keywords:	1602 Criminology; 4402 Criminology
Subjects:	B Philosophy. Psychology. Religion > BF Psychology H Social Sciences > HV Social pathology. Social and public welfare. Criminology > HV7231 Criminal Justice Administrations
Divisions:	Psychology (from Sep 2019)
Date of acceptance:	26 August 2026
Date Deposited:	10 Sep 2026 14:56
Last Modified:	10 Sep 2026 14:56
URI:	https://researchonline.ljmu.ac.uk/id/eprint/29383

View Item