The government is increasingly using algorithms in the application of laws and regulations. It is the responsibility of Parliament to ensure that this is done in a responsible manner. However, various reports suggest that parliamentarians lack sufficient knowledge of AI to adequately assess the use of such systems. PhD researcher Juliette Ermers from the AI Lab for Public Services is investigating how Parliament can gain control over the algorithmic application.
Two days a week, PhD researcher Juliette Ermers is in The Hague. There, she works as a staff member for the Digitalisation Committee of the Senate, supporting senators when legislative proposals concerning digitalisation are on the agenda. "I need to ensure that they are well-prepared when handling legislative proposals in the field of digitalisation and AI," says Ermers. "Additionally, we are working on increasing their knowledge of digital matters in various ways, such as organising expert meetings and field visits."
According to Ermers, who studied in Tilburg, this upskilling of Senate members is incredibly important. "Agencies like the Tax Office and the Employee Insurance Agency (UWV) work daily with complex algorithmic systems, but due to the inherent opacity of these systems, it is difficult for members of Parliament to understand how they work technically. Moreover, developments are happening at lightning speed. This makes it challenging to meaningfully assess the work of these agencies, which are ultimately under ministerial responsibility."
From night-watchman state to welfare state
This difficulty arises in part because Senate members perform their parliamentary work part-time, making it more challenging to keep up with all digital developments. Additionally, over recent years, power has increasingly shifted from Parliament to the executive branch-the implementers of laws and regulations. "After World War II, the Netherlands transformed from a night-watchman state into a welfare state," Ermers explains. "The government took on more responsibilities-too many to execute effectively on its own. Implementation became slow and expensive, so many public tasks were privatised or delegated to independent administrative bodies." Ermers adds that room was created for the executive branch to set its own rules to address the challenges of the welfare state. "This so-called ’administrative state’ is characterised by a shift in power from the legislature to the executive."
Partly as a result of austerity measures since 2007, many of these tasks have also been digitalised and standardised. Ermers says, "Gradually, the executive branch decided to perform more public tasks using automated systems and algorithms. This marked the rise of the so-called ’algorithmic administrative state’, where power shifted even further away from the legislative branch, now towards the programmers and system developers."
It is difficult for members of Parliament to understand how these AI systems work technically
Juliette Ermers from the AI Lab for Public Services
This shift has its downsides, says Ermers. "A programmer translates laws and regulations into computer code. However, the law is not always easy to translate into code, especially by a programmer who has little legal expertise." As a result, the application of the law in computer code may deviate from the intentions of the legislator.
The next step is that power may shift to the systems themselves, Ermers warns. "Self-learning systems can further develop decision-making rules based on the law by using large amounts of data and predictions. In that case, the system might make decisions that no longer directly follow from a democratically established law, but are based on predictions or estimates made by the system. This raises fundamental questions about the democratic legitimacy of the decision. Additionally, due to the technical complexity, it becomes even harder for Parliament to maintain control over this algorithmic decision-making."
Strengthen its oversight
’Gaining control’ is the key phrase Ermers focuses on during the other three days of her work week. Besides her job as a staff member, she is conducting PhD research at Utrecht University into how Parliament can strengthen its oversight of the algorithmic implementation of laws.
"There are formal and informal mechanisms and information relationships to ensure that there is a system of checks and balances, whereby the government must inform Parliament about the functioning of executive agencies to ensure parliamentary and democratic oversight. However, research shows that Parliament often struggles to find the right information and to identify those responsible for choices made and any shortcomings."
A programmer translates laws and regulations into computer code. However, the law is not always easy to translate into code
Juliette Ermers from the AI Lab for Public Services
Through case studies and interviews with two executive agencies, as well as with experts and members of Parliament, Ermers aims to map out successful practices and challenges in the information relationships and accountability mechanisms, so it becomes clear which powers and tools Parliament can use to gain more control over the algorithmic application and what is practically needed to achieve this.
The Tax Office is one of the executive agencies Ermers plans to visit. "In the Dutch childcare benefits scandal, the government failed to adequately inform Parliament about what was happening. At the same time, Parliament failed to deploy the right tools to monitor and detect the issues within the executive agencies. My goal is to explore possible improvements to these mechanisms and relationships, so we can hopefully prevent a ’Childcare Benefits Scandal 2.0’."
Cultural shift
Ermers hopes her research will contribute to increased expertise on this issue among Senate members. "I notice that some of them have limited knowledge in the field of digitalisation. But if you are assessing laws for feasibility and you lack sufficient insight into how an algorithmic system works and what impact it might have on citizens, how can you make an informed decision about a legislative proposal or monitor the algorithmic implementation of legislation?"
She hopes that her dual role as both a staff member and researcher will help her recommendations gain traction. "Several reports have already been published stating that the government’s information management is not in order and that Parliament lacks sufficient control over digitalisation. So, there is still much to be done. With this dual PhD position at the AI Lab for Public Services, I have one foot in the university and the other in the Senate. I would love to contribute from within to strengthening parliamentary oversight of algorithmic application so that citizens do not end up caught in the system."
Utrecht AI Labs
At the Utrecht AI Labs, Utrecht University brings together science and practice by collaborating closely with businesses, the public sector, and other partners. Researchers in the Labs work on responsible applications of AI while simultaneously training the AI talent of the future.