ΑΙhub.org
 

Błażej Kuźniacki on why we need transparency around AI in tax


by
02 March 2023



share this:
Money in a jar next to three piles of coins

Over the course of a number of years, thousands of parents were falsely accused of fraud by the Dutch tax authorities due to discriminative algorithms. The consequences for families were devastating. But, the fact that the scandal was eventually brought to light might prove the Netherlands is ahead of other countries, says Assistant Professor Błażej Kuźniacki. He urges for more transparency about the use of artificial intelligence (AI) in tax related tasks.

The childcare benefit scandal led to allowances being taken away, debt, broken marriages and children being removed from their homes. Do we really need AI in tax?

AI cannot be ignored. It’s of great importance when it comes to tax. Humans are not capable of going through a massive amount of data as fast and accurately as algorithms. And since tax authorities have access to big data, it would be a waste not to use AI. You can train and improve algorithms using this great quantity of data. The point is to use AI in a right way, in particular to not harm taxpayer’s rights.

How do you then prevent AI from making discriminatory decisions?

We need to understand why AI makes certain decisions. You can’t say: “I impose tax on you because AI suggested it”. In the end there must be a human with the authority to make a decision and an understanding of the inner logic of AI. We’ve seen in the childcare benefit scandal that it goes wrong when the process is too automated and secret. AI was allegedly able to use information that has no legal importance in decision making, such as sex, religion, ethnicity, and address. That can lead to discriminatory treatment. Tax authorities must be able to explain their decisions, otherwise they can’t justify them effectively. Trust cannot be fully or even mainly converted from humans to machines (e.g. algorithms).

Do we still rely too much on AI in tax?

The problem is that many decisions and strategies are still hidden, including the use of AI. There are more and more requirements for taxpayers to be transparent. By contrast, tax authorities seem to go the opposite direction due to the increasing use of non-explainable AI systems. That is frightening. AI itself has become so complex that it is hard for humans to fully understand and explain the decisions made by machine learning (ML) algorithms. And on top of that there is tax secrecy that prevents transparency, and sometimes also trade secrecy.

Is the lack of transparency what caused the Dutch childcare benefit scandal?

That was part of it. The Dutch legislation itself doesn’t allow the AI automated decision making to be checked. And there wasn’t enough room for interaction with humans. The procedures were too automatized and secretive. One of the big mistakes in this case was even after it was clear something went wrong, the authorities did not try to help immediately. But this scandal doesn’t mean the Netherlands is one of the worst. It might be the opposite. It could be much worse in other counties. The fact that this scandal came to light a few years ago says that society was able to go through several layers that prevented transparency. It was still found out something was wrong. People eventually went to court over it and effectively defended their fundamental right to respect for private life.

What kind of future do you see for AI in tax?

We need more transparency upfront. Tax secrecy can be reduced by parliament. That is a matter of changing the rules. But understanding the systems of AI will be more difficult. There is no law that requires you to use only explainable AI. Moreover, there are laws preventing you to explain AI because of tax secrecy. We should impose minimal legal requirements for the use of AI. This will force companies and governments to think about the explainability of AI systems they develop, deploy and use because otherwise they will face legal compliance problems. The higher the risks, the higher the explainability requirements should be. We should avoid being passive until another disaster happens.




University of Amsterdam




            AIhub is supported by:



Related posts :



Deploying agentic AI: what worked, what broke, and what we learned

  15 Sep 2025
AI scientist and researcher Francis Osei investigates what happens when Agentic AI systems are used in real projects, where trust and reproducibility are not optional.

Memory traces in reinforcement learning

  12 Sep 2025
Onno writes about work presented at ICML 2025, introducing an alternative memory framework.

Apertus: a fully open, transparent, multilingual language model

  11 Sep 2025
EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) released Apertus today, Switzerland’s first large-scale, open, multilingual language model.

Interview with Yezi Liu: Trustworthy and efficient machine learning

  10 Sep 2025
Read the latest interview in our series featuring the AAAI/SIGAI Doctoral Consortium participants.

Advanced AI models are not always better than simple ones

  09 Sep 2025
Researchers have developed Systema, a new tool to evaluate how well AI models work when predicting the effects of genetic perturbations.

The Machine Ethics podcast: Autonomy AI with Adir Ben-Yehuda

This episode Adir and Ben chat about AI automation for frontend web development, where human-machine interface could be going, allowing an LLM to optimism itself, job displacement, vibe coding and more.

Using generative AI, researchers design compounds that can kill drug-resistant bacteria

  05 Sep 2025
The team used two different AI approaches to design novel antibiotics, including one that showed promise against MRSA.

#IJCAI2025 distinguished paper: Combining MORL with restraining bolts to learn normative behaviour

and   04 Sep 2025
The authors introduce a framework for guiding reinforcement learning agents to comply with social, legal, and ethical norms.



 

AIhub is supported by:






 












©2025.05 - Association for the Understanding of Artificial Intelligence