Gladsaxe model

The Gladsaxe model was a decision tree that Gladsaxe Municipality, a Copenhagen suburb in Denmark, built in house. A decision tree reaches an estimate through a series of questions about a case. This one was to estimate, for every child aged 0 to 6, the probability that the child was living in vulnerability.

Why Gladsaxe built it

Gladsaxe is a municipality of about 70,000 people on the edge of Copenhagen. In 2016 its family department found that children in distress were typically first identified around age 8 to 12. A case review showed different municipal departments had each noted earlier warning signs. Those signs were never connected into one picture.

The project's formal name was Dataunderstoettet Tidlig Opsporing af udsatte boern, or DTO. That means, roughly, data-supported early detection of vulnerable children. The case file calls it an early-detection project.

How it was meant to work

The model was built around about 44 risk indicators, drawn from about nine municipal data sources. It was to score every child aged 0 to 6, not only families already receiving help.

When a child's estimated probability rose above a set level, the model was to flag the case. A specialist adviser was to make a first assessment. The adviser could reject the case and delete it, or ask the family's consent for a full assessment. Without consent, the case was to be deleted.

The municipality meant the model only to identify cases. It was not to make casework decisions, in whole or in part.

What stopped it

The project first sought legal cover as a free-municipality project, which can allow exemptions from current law. In December 2017 the Ministry of Economy and Interior refused the exemption. The municipality then hoped for a legal basis in the new Data Protection Act. The Act, passed in May 2018, did not provide one.

The project became tangled in national debates over the government's "ghetto" package and over GDPR, the European Union's data protection rules. The package set special measures for neighbourhoods classed as "ghettos". It cited Gladsaxe's application as inspiration. In March 2018 a media storm compared it to surveillance states, among them East Germany.

In December 2018 four laptops were stolen from the city hall. One held a spreadsheet with about 20,000 citizens' personal identification numbers. The case file calls the theft unrelated to the project, but the media linked the two.

On 14 December 2018, in response, Liberal Alliance, a party in the governing coalition, withdrew its support. The project was put on hold in January 2019 and ended in its development phase that year. It never ran on live decisions.

What the case study found

A 2022 case study by Kenneth Kristensen found a "decoupled" chain of governance. Councillors had approved the project's purpose. They were not close enough to its method to explain or defend it when the media storm hit.

The case study found the municipality had the staff and skills to run the project. Professional organisations were brought in late, and citizens not at all.

What each side said

The municipality said the model was neither a points system nor mass surveillance, but a targeted use of data. The Danish Association of Social Workers feared the model would flag many families with no real cause for concern. It warned that would spend resources on the wrong families and worry many families without reason.

The case file's reading

Most cases in this collection turn on what a deployed system did. The case file reads Gladsaxe as a case about what a system was never permitted to do. Its decisive controls all sat ahead of the model itself: legal authorization, deployment sign-off, the scope of its data, and data-security capacity.

The municipality had a part-time data-protection officer, against an estimated need of ten to seventeen. The case file concludes that legitimacy, legal basis, and security capacity are controls in their own right. A failure of any one of them can stop a tool long before its accuracy is tested.

The double ethics

The case file calls Gladsaxe its clearest case of "double ethics". The case file's concern about profiling every young child has to be weighed against a documented cost. Warning signs left unconnected had kept vulnerable children unseen until age eight or later.

Gladsaxe's municipal director at the time used the phrase in a public debate in 2019. The case study reports the director's point: there are ethical questions in using the municipality's data about citizens, and also in not using it.

Where the facts come from

The facts here come from Offentlig AI, a university-run catalogue of AI in Danish public services, and Kristensen's 2022 case study. Other sources are a 2023 article by Helene Friis Ratner and Kasper Elmholdt, a 2021 Uppsala University law article by Katarina Fast Lappalainen, and Danish news reporting.

What the available tools can and cannot address

A failure pathway is a link between two parts of the network, where a mistake made by one part can be passed on to the other. A pathway is closed when mistakes stop passing along it. The work along it can go on.

This case has a budget of 12 units. Explore (No Targets) sets no targets. There, any one tool except Require sign-off keeps mistakes contained: corrected rather than building on one another.

Under Service Targets Only, the targets can be met. That level also asks for the model to be helping the advisers' work. Vet connections alone meets them, for 3 units. Within the budget, 40 different sets of tools meet them, and every one includes Vet connections.

Under Service and Safety Targets, the targets can be met. That level adds closing every failure pathway, and it asks that the service not be strained. The cheapest combination costs 7 of the 12 units: Vet connections with Understand the system, which costs 4 units at this level.

Four sets of tools meet these targets, and each includes both of those tools. The other three add Require sign-off, Upgrade model, or both.

Under All Governance Targets, the targets are not fully addressable with the available tools. That level asks for the model to be clearly helping the work. No combination that keeps mistakes contained and closes every failure pathway does that. The best leaves the model helping, but not clearly.

More money does not change that. Every tool at its strongest setting at once costs 31 units, far over the budget. It closes every failure pathway, but the model then helps the work far less. It meets the targets at none of the three levels that set them.