Künstliche Intelligenz und der neue Faschismus

written on October 17, 2026, read March 2026

Reclam publishes a small but, in my opinion, very fine series of little books. The series is called “Was bedeutet das alles?” (“What does it all mean?”). The books are roughly the size of the Reclam booklets I knew from school. Only they aren’t yellow, but colourful. “What does it all mean?” is indeed a question I ask myself quite often.

By now I have bought several volumes from the series. One of them is “Künstliche Intelligenz und der neue Faschismus” (“AI and the New Fascism”). The author states clearly that AI is not a fascist technology, but he draws a very clean line to how this technology can be used and how it fits into certain ideologies. So any naivety we allow ourselves in our thinking (“it’s just a tool”) when we encounter this technology (as with others) needs to be put to a serious test.

In my own thinking, however, one particular sub-problem got firmly “stuck”: the discussion of how probabilistic decision-making (and AI is, after all, a process based on probabilities) relates to the fundamental principles of the rule of law and to our framework of norms.

The author describes this as the contrast between the individual case decision and the case wager. This is nothing specific to AI; it applies to all technologies in which decisions are not made on the basis of a careful examination of the individual situation, but on the basis of statistical processes and comparability with other cases, producing a decision that is right most of the time, or almost always, but not necessarily always, especially because it is subject to the optimisation dilemma of such processes: optimising at the same time for being right in every case and for not overlooking any special cases. The author illustrates this with the example of job applications. Filtering out only suitable candidates from the applications while at the same time finding qualified applicants who don’t meet the criteria but fit just as well or better – ensuring both at once is difficult, if not impossible.

That may be acceptable for business processes. But we hold the application of laws in particular to higher standards, which – as I’d like to point out – is already reflected in the fact that legal expenses insurance has more or less become the done thing today, to minimise the cost risk should you want to have decisions reviewed in court.¹

And we haven’t even got to the challenge that we have the right to have a decision explained to us – which can prove difficult when using a technology in which this “why?” is still an active research problem.²

I want to point out that I am explicitly not taking an anti-AI position here. A process made up of individual case decisions isn’t free of errors either. People make mistakes, and automated systems that support them can be flawed, too.

What if, for example, a process based on automated case wagers makes fewer mistakes than a process made up of individual case decisions made by people? What if, due to social developments such as the double demographic problem (more demand because of the change, and at the same time fewer staff because of the very same development), we won’t even have enough people in the future to make individual case decisions in a reasonable time? And how might case wagers and individual case decisions be combined?

Could one possible way be not to let AI decide individual cases itself, but to use it to develop deterministic systems that support people in making individual case decisions?³ The decisions would then still not necessarily be right or fair. But the decision path could be traced unambiguously and, where necessary, corrected or extended. Moreover, LLM-based AI would be used where one of its great strengths lies: in analysing linguistic constructs – such as laws – and in producing new linguistic constructs, in this case program code.

This is a discussion that needs to be had. Just formulating the problem of case wager versus individual case decision so clearly and placing it in a social context made this book worth reading for me, regardless of the rest of the text. I consider this one chapter alone to be an important contribution to the debate, one that must be taken into account in all the discussion about efficiency in government processes.⁴ That a book gives its readers food for thought is perhaps the highest praise one can give it. And for me, this book delivered.