Once bitten, Netherlands wants to move early on algorithm supervision

[Coffeemill/Shutterstock]

The Dutch government is set to be an early implementer of several crucial provisions of an upcoming EU legislation to regulate Artificial Intelligence (AI) as part of an ambitious working agenda for digitalisation.

In the past years, the Netherlands was the setting of a large-scale scandal because of a flawed algorithm that led the tax authorities to falsely accuse thousands of fraud related to a childcare benefit scheme.

The Hague is now trying to spearhead the human rights vetting of algorithms and the creation of a national database to keep track of the AI systems used by public bodies, anticipating two aspects of the AI Act, flagship legislation to regulate AI applications based on their potential for harm.

“The AI Act should lead to explainable AI systems which respect human rights and lead to meaningful transparency that generates trust, which is, of course, a vital building block of a flourishing AI market and also important in a democracy,” Alexandra van Huffelen, the Dutch minister for digitalisation, told EURACTIV.

The working agenda adopted by the government also includes new initiatives to foster digital skills and a digital wallet.

Public register & supervision

A new supervisory body, starting on 1 January 2023, will enforce a set of new rules that essentially anticipate the EU’s upcoming rules regarding transparency obligations, compliance marking and the right to complain.

Similarly, next month, the Dutch government will introduce a mandatory algorithm register for AI used by public bodies, from the central administration to the local authorities. The idea is that citizens cannot enjoy the right to redress unless they know that an AI is being used.

In the original version of the AI Act, only providers of high-risk AI systems were obliged to register on an EU-wide database. However, following pressure from countries like the Netherlands, the obligation was also extended to public authorities who use an AI system.

One of the blueprints for the algorithm transparency register is the one from the City of Amsterdam, which was set up voluntarily by three different groups: the general population, civil servants, and technical experts.

“We need to work together with the companies. Otherwise, we cannot make our algorithms transparent. IT departments are definitively interested in opening up and having a dialogue,” a representative of the Municipality of Amsterdam told an event in the European Parliament last week.

Impact assessment

Moreover, public authorities will have to assess the algorithms’ impact on human rights before they are used.

For this purpose, the Dutch government developed an AI impact assessment for evaluating the human rights risks of the system, passing which the system will receive a conformity assessment mark.

“Evaluating AI is not just looking at the technology, it’s looking at the context where that technology is implemented. We don’t want to stare blindly at a black box. The way the black box is used is important to consider,” Mirko Schafer, an associate professor at the University of Utrecht and co-developer of the impact assessment, told the same event.

Schafer explained that the assessment is based on the context and added value of using the algorithms, how the models have been trained, the risk mitigation measures in place, how often the AI needs to be reviewed, and the potential implications on human rights.

The register is made public for oversight bodies to judge the deliberation process, which civil society organisations and journalists might also vet. The register will become mandatory for private companies once the AI Act is adopted.

For van Huffelen, the Dutch impact assessment does not go beyond the EU Council’s position on the AI Act, which in her review, is sufficiently robust. “The Netherlands is committed to ensuring this will not be watered down during future negotiations,” she added.

Last-minute changes to EU Council’s AI Act text ahead of general approach

The position of the EU Council on the flagship legislation to regulate Artificial Intelligence was shared on Friday (8 November) with some final last-minute adjustments made by the Czech Presidency.

Digital skills

The Dutch authorities estimate that 20% of the population in the Netherlands lacks basic skills. This phenomenon is not limited to the uneducated but also those that need to be retrained to meet the digital proficiency needed for today’s professions.

The agenda also includes new digital skills curricula to teach coding from primary and secondary schools, which new training offers will accompany at the regional level with low thresholds intended to retrain or requalify school drop-outs with advanced digital skills.

The Hague will also introduce digital information points in public libraries where citizens can get help to access essential public services online. Also in this regard, the Dutch government is working on a digital wallet that anticipates upcoming rules at the EU level.

The wallet will include not only the national ID but all public documentation such as driving licenses, health insurance and educational diplomas. It will follow the data minimisation principle, so for instance, if a night shop uses it to verify someone’s age, they will not be able to access other information such as a personal address.

A beta version of the wallet is expected in the coming weeks, with the official launch expected for 2023.

Czech Presidency pushes forward on digital identity following political discussions

The Czech Presidency of the EU Council presented a new compromise last week based on a debate at the ambassadors’ level on addressing the most sensitive questions blocking the European digital identity.

[Edited by Zoran Radosavljevic]

Read more with Euractiv

Subscribe to our newsletters

Subscribe