“Parliament must gain control over the use of algorithms”

Our government is increasingly using algorithms in the implementation of laws and regulations. Parliament is charged with monitoring whether this is done responsibly. But several reports show that members of parliament lack the understanding of AI to adequately review the use of such systems. PhD student Juliette Ermers of the AI Lab for Public Services is researching how parliament can get a handle on algorithmic implementations.

Two days a week, PhD student Juliette Ermers is based in The Hague, where she works as a staff assistant for the Senate Digitalisation Committee. This means she supports senators when legislative proposals dealing with digitalisation are tabled. “It’s my job to ensure they are well prepared when dealing with proposed legislation in the field of digitisation and AI,’ Ermers said. ‘We also try to increase their general knowledge on digital issues in many ways, for example by organising meetings with experts and field trips.”

Ermers, who got her Master’s at Tilburg University, explains that keeping senators up to speed is incredibly important. “Executive bodies, e.g. the Belastingdienst and UWV, work with complex algorithmic systems on a daily basis, but it is difficult for members of parliament to keep up with all these developments.  Understanding how such systems work technically is quite complicated. This makes meaningful scrutiny of the work carried out by executive bodies, which in principle falls under the ministerial remit, a constant challenge.”

From night watch state to welfare state

This is partly due to the fact that members of the Senate serve part-time, which complicates keeping up with all the developments going on with regard to digital technology. Moreover, in recent years, power has increasingly shifted from parliament to the executive - the implementers of laws and regulations. “After World War II, the Netherlands changed from a night watch state to a welfare state,” Ermers explains. “The state was given more and more tasks; too many to carry out properly on its own. Performing all these tasks was time-consuming and expensive. That is why numerous public tasks were privatised or outsourced to independent administrative bodies.” In addition, according to Ermers, the government was given the scope to set its own rules so that it could meet the challenges of the welfare state. “This so-called administrative state is characterised by a power shift from legislature to administration.”

Partly as a result of budget cuts from 2007 onwards, many of these tasks have in turn been digitised and standardised. Ermers: “The authorities gradually decided to use automated systems and algorithms to carry out more and more public tasks. This process marked the rise of the so-called ‘algorithmic’ administrative state, where power shifted even further away from the lawmakers, i.e. to programmers and systems developers.”

Understanding how systems work technically is complicated

There is a downside to that, says Ermers. “A programmer translates legislation and regulations into computer code. But the law is not always readily converted into code, especially by a programmer, who will have virtually no legal expertise. Consequently, the result may be that the effect of a law in computer code ends up deviating from the legislator's actual intent.

As a next step, the power is likely to shift to the systems themselves, says Ermers. “Self-teaching systems can evolve and develop the – originally law-based – rules of decision, based on large amounts of data and prediction. Subsequently, such a system can suddenly be seen to make decisions that do not in fact follow directly from a democratically enacted law, but which are based on the system's own predictions or assessments. This raises fundamental questions as to the democratic legitimacy of these decisions. Moreover, the technical complexities further hinder parliament’s ability to grasp such algorithmic decision-making.”

Taking back the reins

Consequently, ‘Gaining a firm grasp’ is the key phrase Ermers engages with for the remaining three days of her working week. For besides her work as a staffer, she is conducting research as a PhD student at Utrecht University, looking into ways for parliament to strengthen its monitoring role with regard to algorithmic operations.

“Formal and informal mechanisms and avenues of information are in place to ensure a system of checks and balances, whereby the government must report to parliament on the functioning of executive bodies for the purpose of parliamentary, democratic scrutiny. But research has shown that parliament frequently ends up on laborious quests for the right information, for those responsible for choices made, and for any shortcomings.”    

The law is not always readily converted into code, especially by a programmer with virtually no legal expertise

In order to clarify which powers and tools parliament can use to get a better grip on algorithmic implementation and what is needed to achieve this in practice, Ermers aims to identify best practice and obstacles to information relations and accountability mechanisms, using case studies and interviews at two executive bodies and with experts and members of parliament.

One of the executive bodies Ermers is keen to visit is the Belastingdienst, the Dutch tax authority, known for the notorious Toeslagenaffaire which saw innocent people targeted and penalised for allegedly fraudulent use of a tax credit scheme. ‘’With the Toeslagenaffaire, government did not adequately inform parliament about what was going on. At the same time, parliament itself failed to deploy instruments at their disposal to effectively identify problems with implementation. As such, my objective is to explore what improvements to these mechanisms and relationships are possible, so that hopefully we can prevent anything like the Toeslagenaffaire from happening again.”

Changing cultures

With her research, Ermers wants to contribute to increased expertise on this topic among members of the Senate. “Currently I observe limited knowledge in the field of digitisation. However, when testing laws for enforceability, how can anyone make an informed decision on a bill or monitor the algorithmic implementation of such legislation, without sufficient insight into how those algorithmic systems work and what the impact on the public could be?”

She hopes that her combined position as both staffer and researcher can contribute to making her recommendations stick. “Several reports have concluded that the government's information management is not up to scratch and that parliament has an insufficient grip on digitisation. So there is still a lot of ground to cover. This dual PhD track in the AI Lab for Public Services allows me to have one foot in the university and the other in the Senate. I would love to be able to contribute to strengthening parliamentary control over algorithmic implementation from the inside, so that members of the public are not caught in the fray.”