Artificial Police Agents: Looking Awry

The document reports on a seminar titled 'Artificial Police Agents: Looking Awry', featuring Prof. Dr. Mireille Hildebrandt. The discussion focuses on the risks of 'predictive forensics' and machine learning in law enforcement, specifically how the choice of training sets and the inherent bias of algorithms can lead to fragile predictions and the erosion of the presumption of innocence. Hildebrandt argues for 'legal protection by design', where fundamental rights are integrated into the technical architecture of surveillance tools, and proposes the establishment of a watchdog to verify the design and operation of such systems.

• Machine learning in policing relies on the detection of bias within datasets to create predictive patterns; however, these predictions are fragile because the 'target function' (the real statistical relationship between input and output) is unknown since future data cannot be accessed.
• The selection of training data is a political and methodological choice that determines the output; using 'low hanging fruit' (easily available but irrelevant data) or excessively large datasets can lead to incorrect results or the detection of spurious correlations.
• Predictive technologies in law enforcement may undermine the presumption of innocence by enabling the screening and monitoring of potential suspects before any criminal offense has occurred.
• Hildebrandt proposes two specific legal safeguards: the imposition of rules regarding the design and default settings of data-driven systems to protect individual rights, and the creation of a watchdog to verify these systems, especially during secret investigations.
• The concept of 'legal protection by design' suggests that system architectures should be built based on democratic legislation, enable resistance, and allow for contestation in court to protect fundamental rights.
• There is a risk that replacing human police discretion with rigid software systems leads to a 'legalistic' system that ignores the need for interpretation and the growth of professional expertise.

APA

Bergersen, S. (2016). Artificial Police Agents: Looking Awry. Peace Research Institute Oslo. https://cdn.cloud.prio.org/files/9b812e0f-9b37-4406-9a9a-ea47359c734f/Artificial Police Agents LASIE Seminar Report 2.pdf?inline=true

Chicago

Bergersen, Stine. Artificial Police Agents: Looking Awry. Peace Research Institute Oslo, 2016. https://cdn.cloud.prio.org/files/9b812e0f-9b37-4406-9a9a-ea47359c734f/Artificial Police Agents LASIE Seminar Report 2.pdf?inline=true.

Wikipedia

{{cite report |last1=Bergersen |first1=Stine |title=Artificial Police Agents: Looking Awry |publisher=Peace Research Institute Oslo |date=18 November 2016 |url=https://cdn.cloud.prio.org/files/9b812e0f-9b37-4406-9a9a-ea47359c734f/Artificial Police Agents LASIE Seminar Report 2.pdf?inline=true |access-date=23 October 2026 |via=Climate Insights Directory}}

BibTeX

@techreport{bergersen2016artificial, author = {Bergersen, Stine}, title = {{Artificial Police Agents: Looking Awry}}, institution = {Peace Research Institute Oslo}, year = {2016}, month = nov, url = {https://cdn.cloud.prio.org/files/9b812e0f-9b37-4406-9a9a-ea47359c734f/Artificial Police Agents LASIE Seminar Report 2.pdf?inline=true}, urldate = {2026-10-23}, note = {Indexed by Climate Insights Directory} }