Local authorities are increasingly using algorithms to try to identify potential lawbreakers such as criminal gangs and benefit fraudsters, according to a survey by NOS.

At least 25 municipalities use systems that create ‘risk profiles’ for offences that impact on council funding, as well as identifying people at risk of running up high debts.

The tax office was criticised for using dual nationality as an indicator of risk in families suspected of fraud in the childcare benefits scandal, which led to the resignation of the cabinet last month. A series of investigations found that many people were wrongly targeted and blacklisted because they made small administrative errors.

Last year a court ordered the government to stop using its algorithm-based fraud detection system SyRI, which used risk profiling to draw up lists of people suspected of housing or social security fraud.

Judges ruled that the system, created by social affairs ministry officials in 2014, breached European human rights and privacy laws because it labelled people as potential fraudsters without any basis in evidence.

Anton Ekker, one of the lawyers who challenged the government’s use of SyRI in court, said the proliferation of algorithms was cause for concern. ‘The cabinet has pledged that there will not be a repeat of the childcare benefits scandal, but there is no clear picture of what all the municipalities are doing,’ he told NOS.

Simple algorithms are used extensively across government – for example, to calculate how much a driver should be fined for exceeding the speed limit. But more sophisticated systems carry the risk of discrimination if they focus on particular social groups or neighbourhoods.

Holiday lets

Amsterdam uses algorithms to try to detect illegal holiday letting as it seeks to curb Airbnb-style rentals in residential areas, as well as people at risk of falling into severe debt.

Brielle en Nissewaard, near Rotterdam, uses ‘smart modelling’ to identify potential benefit fraud by highlighting distinctive patterns of behaviour. Breda, in Brabant, evaluates the effect of proposed construction or demolition projects on neighbourhoods with the help of machine learning.

Leudal en Maastricht, in Limburg, assesses construction and real estate paperwork to try to spot patterns indicating the presence of criminal gangs.

Ekker told NOS the use of artificial intelligence risked leading to further situations such as the benefits scandal. ‘I’m not sure if local authorities have enough in-house expertise to do this in the right way,’ he said. ‘They are often highly dependent on external IT suppliers.

‘If it threatens to go wrong again, we cannot rule out going back to court.’