Should we go further in the automation of the hunt for tax evasion?

“These technologies must first be regulated by laws”

David Hadwickresearcher at the Digitax center, at the University of Antwerp (Belgium), and at the FWO (Belgian equivalent of the CNRS)

“France is one of the most advanced countries in the use of artificial intelligence (AI) against tax evasion. Other leading EU member states include Belgium and the Netherlands, where AI was first used to facilitate customs checks at the ports of Antwerp and Rotterdam. Germany introduced AI through a tax administration modernization law passed in 2017.

Should we go further in the use of AI? I am nuanced. On the one hand, some models do indeed improve the efficiency of tax administration. But we must remain extremely vigilant because some devices use sensitive data and the way the algorithms are designed may contain biases, even discrimination.

This is what happened in a scandal revealed in the Netherlands in 2021. Several thousand parents had been unfairly accused of fraud in connection with their childcare allowances. They had to repay all the aid – up to several tens of thousands of euros –, causing human tragedies (suicides, etc.) which still have serious consequences.

It was discovered that the Dutch tax authorities were using a tool to target potential fraudsters, based on criteria such as holding dual nationality, which amounts to discrimination. The danger is all the greater as the AI ​​tools are massive, errors therefore have multiplied effects.

These technologies must absolutely be placed under the control of humans, and framed by laws. However, there is no European text and of the twenty or so Member States that use AI, only four have a law, sometimes very fragmented as in France. This issue of legality is fundamental and must be tackled head-on. »

“Questionable algorithms for democracy and privacy”

Bastien Le Querrecof the association La Quadrature du Net, which defends and promotes rights and freedoms on the Internet

“It is not desirable to go further in the use of artificial intelligence (AI) to fight against tax evasion. First, the algorithms used are for the most part secret, and even when certain elements are known, the whole remains opaque. The tax administration communicates little, whereas over the years it has ceased to collect increasingly broad data: those held by public administrations, first, then by focusing on private platforms (such as Leboncoin or eBay) and now by using social networks.

Some justify the lack of transparency by claiming that any advertising on these means would allow fraudsters to circumvent them. But the objective of the algorithms is to consolidate control, and therefore to track down small taxpayers. Serious crime – which really has the capacity to circumvent the system – is not at all targeted by these tools.

The second objection to the deployment of AI comes from the need to respect privacy. Indeed, the tax authorities go further and further in the collection of data, without any real limit at this stage, insofar as most of these measures are taken by regulatory means, far from Parliament.

As a result, these algorithms are democratically very questionable: without transparency, without knowing exactly how they work, it is impossible to establish a public debate on their existence, their uses, and on the societal consequences generated. Certainly, if there were transparency, the opposition of public opinion would be such that the only choice would be to stop using it. »

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *