New bill would allow authorities to use AI

The government has introduced a new bill that would give authorities such as municipalities and state agencies a clear legal basis, meaning an explicit right in the law, to use artificial intelligence (AI) when they process citizens' personal data. It is the government's second attempt. The first bill lapsed when an election was called, and since then many AI projects in the public sector have stalled because the authorities have not had a clear legal basis to act on.

New right to use AI

With the new bill, all state and municipal authorities, as well as entities such as ATP and Danmarks Nationalbank, get the right to use AI, including as decision support, where an AI system suggests a solution that a human can use to make a decision. This applies only when a human still makes the final decision, though. AI may thus not decide cases entirely on its own. For sensitive data, for example about health, extra requirements apply, and the authorities must still follow the general rules on data protection and public administration, including the EU's data protection rules (GDPR) and the EU's new AI Act, which sets common rules for the use of artificial intelligence.

Criticism of broad access

Several sources believe the bill has improved on the original draft. But the concerns haven't gone away. Johan Busse, chair of the Danish Data Ethics Council, a council that advises on ethical issues in the use of data, points out that the biggest problem is how broad the bill is. The council doesn't consider it responsible that both large agencies with plenty of legal and technical expertise and small municipalities with fewer resources get the same broad access to use AI, without clear requirements at the same time that the systems be just as accurate and fair as the ones they replace.

Among other things, the council proposes common minimum requirements for how many errors AI systems may make, and for how to avoid skew in the systems' assessments, known as bias. The council also wants it established that the employees who use AI must genuinely be able to assess the cases themselves and not just rubber-stamp what the machine suggests. In addition, the council wants authorities to make an ethical assessment of the consequences before they adopt new AI systems. Busse compares it to the rules for medicine and food: there must be rules so you can trust what you use.

Unclear interplay with other rules

A lawyer also points out that it is still unclear how the new bill fits with the rules that already apply to AI in specific areas, for example in employment services or elder care. According to the lawyer, this interplay between the different sets of rules is not resolved in the new bill.

For ordinary Danes, the case means that the public sector will probably find it easier to use AI when handling cases concerning citizens. But it is still unclear how to ensure it is done properly and can be checked. The bill will now be further debated in the Danish Parliament.