Turning noise into signal: the legibility work of welfare recipients in digital systems

3 min read

Welfare systems are being reshaped by AI across the globe,. New technologies are being introduced with the professed aim of making welfare systems more efficient. But this transformation can also result in unfair practices. In recent years Amnesty International and Lighthouse Reports have highlighted instances of algorithmic discrimination in welfare systems across Europe, which have had sometimes devastating consequences for people who do not fit the norms embedded in such AI systems. These consequences include falling deeper into debt, losing their homes or suffering from health issues. 

I have spent the past two years working as a legal advisor at a local welfare foundation in Rotterdam. The consequences just referred to were all too evident. In my current research, I focus on the gap between welfare recipients’ everyday realities and the normative categories embedded in digital welfare systems, but also how welfare recipients may make themselves legible in a system that has limited room for nuance and flexibility. I argue that we should regard the adjustments people make when interacting with digital welfare systems not as instances of friction or »noise«, but as signals that may be used to develop strategic design interventions with a view to building welfare systems that work for everyone. 

Where the frontline and the frontend of the welfare state meet

I visited libraries and neighborhood centers in Rotterdam at which advisors and volunteers offer social support, legal advice and digital help, all free of charge. I spoke to many advisors and welfare recipients and listened to their stories about how they try to cope with the digital welfare state. 

During interactions in which welfare recipients and advisors were completing the relevant applications, I began to notice a number of subtle developments. There seemed to be a kind of negotiation going on between advisors and welfare recipients with regard to what information was important or relevant and what information was not. More importantly, through this process of negotiation, welfare recipients were trying to describe their circumstances on paper so that it would fit into the rigid parameters of digital forms.

Two particular stories from migrants stuck with me; Amir’s experiences with the Dutch Tax Authority and Indra’s experiences with the municipality of Rotterdam. 

Doubting people’s deservingness

The first time I met Indra, she had come to the library to apply for a local tax waiver available to citizens on low incomes. Applications for this waiver must be made online. They include questions about household members, basic expenses and the applicant’s current income and assets.

Although Indra was eligible for the waiver, she was worried that the municipality might misconstrue the amount of money she had in her bank account. She had not yet paid the rent that month, which was due in a week, making her balance at that point look a lot higher than it usually is. As a result she chose not to submit the application – which was otherwise complete – at that time, and instead to do so the following week. In the meantime, she paid her rent, thereby lowering her bank balance and, in her view, removing any doubt that she »deserved« the waiver.

This simple but strategic action enabled Indra to tailor her application to match what she expected the system »wanted«. She left out information she feared might jeopardise her application. In order to avoid even the perception of ineligibility, she responded proactively to how the municipality might process her information.

To apply or not to apply?

Indra’s decision to excise certain information from her application was strategic. In other cases, information may have to be left out of applications for the purpose of accessing welfare. This was the case, for example, with Amir and his family, who had fallen on hard times because the breadwinner, Amir, needed emergency surgery and shortly afterwards became unemployed. The point of interest with regard to the subject of this article is that although he was technically eligible for a wide range of financial benefits to support the family, he was required to provide his wife’s citizen service number so the authorities could determine their eligibility as a couple. However, his wife was still awaiting an immigration decision and was yet to be allocated a citizen service number. Furthermore, it seemed that the relevant decision was still some months away. The authorities thus advised him to apply for the benefits as a single parent and supply his wife’s information later on. In effect, this entailed avoiding any mention of his wife in his application, an omission that might be tantamount to fraud. Understandably, this ploy left them feeling extremely anxious. But what other choice did they have? 

These stories should most emphatically not be construed as attempts to manipulate or misrepresent reality. Rather, Indra and Amir acted in an effort to make themselves »legible« and perceived as »deserving« in a system not designed to accommodate subtlety. Their stories may be said to illuminate the everyday interpretive labour sometimes involved in accessing welfare in digital welfare systems.

Such labour also strongly resembles what researchers call »data cleaning«. Data cleaning describes the removal of potential »noisy data« – inconsistencies or errors – from datasets in order to render them legible by the relevant system. In a sense like researchers and data scientists, welfare recipients thus also clean their data, albeit with different goals. They make deliberate choices about what data to include and what to omit in order to access welfare.

From noisy data to a signal for help

All digital systems require standardisation. But serious issues may arise when such standardisation masks signals for help. We must thus reframe the »noisy data« characteristic of welfare application processes – whether they be anxieties and fears, or data for which no input option exists – not as errors to be corrected but as signals that digital welfare systems as currently designed are unable to take account of the realities of welfare recipients’ everyday lives. We need to reimagine welfare processes so that they are able to do justice to non-normative experiences by offering custom, indeed »human« solutions when implementing eligibility screening. Only by actually heeding the living signals of welfare recipients – turning what systems dismiss as noise into knowledge – can we rebuild welfare infrastructures that are constructive, inclusive and genuinely supportive of the people who rely on them.

The Inclusive AI Lab at Utrecht University is dedicated to incubating leaders and helping to build inclusive, responsible and ethical AI data, tools, services, policies and platforms, with a special focus on the Global South.

Rana Kuseyri is a PhD candidate at Utrecht University and a Responsible AI researcher at the Inclusive AI Lab. Rana’s expertise focuses on digitisation, data justice and the welfare state. She has over four years’ experience working with leading civil society organisations such as Je Goed Recht, Systemic Justice and PILP (formerly the Public Interest Litigation Project).