How Runaway Algorithms Brought Down the Dutch Government

(Click on the “CC” button for subtitles if the Dutch accents get in the way.)

By the Center for Human Rights and Global Justice, New York University School of Law.

This concerns shockingly discriminatory use of self-learning black-box algorithms, designed to discriminate on the basis of factors such as socioeconomic status and nationality in the Dutch tax authorities’ (and other organizations’) scrutiny for fraud. And that’s far from the whole story.

It’s also true that regulation in the Netherlands is often way too rigid. It makes us cloggies abroad often look really strange (OCD-like) because we are so used to having to be super precise and document everything. This can for example lead to people having to pay back 90,000 euro because they made a 100-euro typo in a form or perhaps because they forgot to sign the form or because paid the childcare center one day late or the amount they paid the childcare facility was 10 euro less than it should be.

However, this scandal was about people being labeled fraudsters and being treated as fraudsters mostly on the basis of nationality without them being fraudsters at all and without them having an idea what error they were supposed to have made and beingg unable to obtain any information. It’s destroyed many lives. Some people lost their homes, others even their children (taken into care) or their marriage.

The first signs of the scandal emerged in 2014. It would take until 2018 before some people started waking up and started looking into it. (Do we owe this delay to the phenomenon of conspiracy theorists?)

It’s 2022 now. It’s doubtful that all the people who were harmed in this have already received compensation, so I understand. (I know that several individual court cases are ongoing or will be started soon.)

A problem with the complete opacity of and the secrecy surrounding these black box algorithms is that you wouldn’t even know it if hackers interfered and, for example, inserted their own parameters into the model. They could perhaps, also feed it highly biased learning data, but indeed, the innate biases in the model would also steer it toward concluding that poor people and foreigners are more likely to commit fraud.

How does this work? If you are looking for fallen leaves in parks, but only or predominantly look in two specific parks, you might erroneously conclude that those two parks have high numbers of fallen leaves and that there are no leaves in other parks. As you haven’t looked at all (or well) in those other parks, you found no or much fewer fallen leaves there.

Btw, the term “white-washing” is Dunglish. It’s called “money laundering” in English.

Similarly, “Minister President” is “Prime Minister” (Rutte). Turns out that he (Mark Rutte) was found guilty (by the court) of encouraging discrimination in his role in 2003, when he was not Prime Minister but ran the Dutch version of the English DWP. He had asked for people of Somali descent to be traced and scrutinized for fraud. Rutte did not get it at all at the time and expressed surprise. Apparently, he laughed and commented that the law should be changed, then. The reason for this was that some Somalis had committed fraud. That’s like saying that all Dutch people are blond and blue-eyed because some are. It’s nonsense.

(This latter court case ran its course after I had moved to England. I was not aware of it.)

Another worrisome development in the Netherlands is that the banks want to start sharing their customer data. That does not feel right. What’s next? Insurance companies? Car dealers? Broadband and cable service companies? Just in case you consume “too much data”, perhaps so that they can refuse you as a customer?

The poverty levels in the Netherlands are still nowhere near what you see under the UK’s often so mean-spirited governments. Another difference is that Dutch food banks usually provide weekly food packages so that people are actually supported.

Feel free to share your opinion below, please.

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.