GroenLinks wants government agencies that work with algorithms to be required to sign up for the algorithm registry. There should also be a human rights test for algorithms. Secretary of State for Digitalization Alexandra van Huffelen thinks it is a good idea, but says a legal obligation cannot take effect until 2024 at the earliest.
That was one of the wishes expressed Wednesday in the Standing Committee on Digital Affairs' debate on artificial intelligence, RTL News writes (1).
An algorithm registry has long been talked about, and in December of last year the first version of it was a reality. The algorithm registry, as the name suggests, is a database that lists which companies and agencies use an algorithm. It provides clarity about the type of algorithm an organization uses, as well as what data an algorithm collects and processes.
By providing openness and transparency, the government wants the Dutch to gain confidence in artificial intelligence and algorithms. Also, experts can contribute to the operation of algorithms. In this way, the government wants to prevent a new Benefits Affair or Fraud Detection Facility (FSV).
While the algorithm register is online, its development is far from complete. One of the wishes of the Lower House is that government organizations working with an algorithm be required to sign up for the algorithm register. State Secretary Van Huffelen agrees, but emphasizes that it will take at least until 2024 before a legal obligation can be realized.
"We want government parties to ensure that they register as many algorithms as possible, and especially high-risk algorithms, such as those that process personal data or are used for risk assessments," Van Huffelen told members of the standing committee on Digital Affairs on Wednesday.
Kauthar Bouchallikht (GroenLinks) is concerned about an exception the cabinet is making for organizations that use algorithms for law enforcement. According to her, this amounts to a curtailment of civil rights.
State Secretary Van Huffelen tries to reassure the MP by saying that the Autoriteit Persoonsgegevens has insight into how algorithms used for enforcement and defense work. "It's undesirable to make them public, because you don't want people to abuse them. But we also want those algorithms to meet all the requirements," the admiral said.
Another topic Bouchallikht touched on was a human rights test on algorithms. She believes this should become "mandatory, binding and transparent" to rule out exclusion and discrimination. Van Huffelen promised to push for a "robust test.
"Developers of high-risk AI systems will be required to carry out a human rights test with respect to fundamental rights, safety and health," the state secretary said. This mandatory test is not expected to occur for another two years, when the European AI regulation comes into force.
Van Huffelen promised late last year to further develop the algorithm register next year. In the current version, data was entered manually. In the future, data should be automatically retrieved from decentralized algorithm registers. Government organizations that do not have their own algorithm register should be able to post information about their algorithms directly to the central register.
Earlier this month, the Autoriteit Persoonsgegevens announced that from now on it will also oversee the use of algorithms. The regulator plays a coordinating role in supervision and seeks cooperation with stakeholders and experts. "Other bodies also have a lot of experience with the supervision of algorithms, just like the AP. Together we can pool a lot of knowledge and insight. Thanks in part to everyone's input, thinking along and counter-thinking, the foundation has now been laid for this broad, cross-domain algorithm supervision," said AP board chairman Aleid Wolfsen.
https://www.rtlnieuws.nl/tech/artikel/5361653/algoritmes-overheid-ai-regels-tweede-kamer
