Ghent University develops AI tool to guess when and where burglaries will take place

Ghent University is developing an artificial intelligence (AI) tool designed to anticipate the locations and times when burglaries might occur. 

The tool aims to help police target their patrols more effectively, but its calculation criteria and effectiveness are already causing controversy.

The project is called BIGDATPOL, which is short for Big Data Policing, and received €2 million in funding from the European Commission. 

The model draws on a significant amount of data to make its predictions, De Morgen reports. The researchers cross-reference the locations and dates of past burglaries with about 70 environmental indicators. 

The presence of a gym, a petrol station, a hospital, a nightclub or a hairdresser’s, as well as the layout of the streets, can be incorporated into the analysis.

Professor Wim Hardyns, who is leading the project, said that these factors might, according to certain criminological theories, influence the risk of burglary in a given area.

The algorithm also takes into account socio-economic data, such as population density, the age and income of residents, the number of households, the unemployment rate and the proportion of people of non-European origin - this last data point being the main source of criticism.

Researchers and anti-racist organisations are concerned about the risk of bias, noting that an algorithm can establish a statistical correlation without necessarily identifying a genuine cause-and-effect relationship. 

“If an algorithm detects a correlation between a variable and the outcome, it assigns it significant weight, but this correlation doesn’t necessarily exist in reality,” explained Eli Verwimp, a researcher in algorithms at the VUB. 

A neighbourhood under increased surveillance, for example, may appear more frequently in crime statistics, not necessarily because it experiences more offences, but because more checks are carried out there. 

Critics of BIGDATPOL say this mechanism could lead to an increased police presence in neighbourhoods that are already vulnerable or stigmatised.

Also questioned was the actual effectiveness of such artificial intelligence policing systems. Experts point out that, at this stage, there’s no solid scientific evidence to show that predictive policing algorithms actually reduce crime. 

The Netherlands abandoned a similar system after 10 years of use due to a lack of clear evidence of its added value.

The debate also comes at a time when burglary figures are falling in Belgium. Since 2020, the number has barely exceeded 40,000 per year, compared with about 56,000 in 2016.

Still in the trial phase, BIGDATPOL will need to prove its worth on two fronts: its practical usefulness to the police and its ability to prevent discrimination or the stigmatisation of certain areas.