Durham Constabulary experiments with predictive policing 

Examples

In 2012, Durham Constabulary, in partnership with computer science academics at Cambridge University, began developing the Harm Assessment Risk Tool (HART), an artificial intelligence system designed to predict whether suspects are at low, moderate, or high risk of committing further crimes in the next two years. The tool is used to decide whether to recommend referral to the rehabilitation programme Checkpoint, which aims to reduce reoffending by helping remediate the individual's problems, where possible. The algorithm, one of the first to be used by police forces in the UK, analyses 34 different categories of data, including age, gender, and offending history. Among these is postcode information, some of which is being removed; this type of information can act as a proxy for socio-demographic characteristics and therefore reinforce existing biases in the criminal justice system. 


A September 2017 paper by a group including both academics (Marion Oswald, University of  Winchester; Jamie Grace, Sheffield Hallam University; and Geoffrey Barnes, University of Cambridge) and a member of the police force (Sheena Urwin) is the first to examine the system. Using it as a case study, the paper proposes approaches to assessing and managing the use of such algorithms to ensure both more efficient use of police resources and more consistent, evidence-based decision-making. The paper also argues that there are some decisions that should be removed altogether from the influence of algorithmic decision-making because their impact on society and individuals is too great. In a talk at the Royal Society in October 2017, Marion Oswald presented this research in detail (audio is available).


http://www.wired.co.uk/article/police-ai-uk-durham-hart-checkpoint-algorithm-edit
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3029345
http://downloads.royalsociety.org/events/2017/10/algorithms-society/oswald.mp3
Writer: Matt Burgess, Marion Oswald
Publication: Wired, SSRN, Royal Society
 

Related learning resources