I am Austrian and wrote this in German. The English is a literal machine translation, not rewritten.
Everywhere now, catastrophic harm to humanity is being discussed, and various scenarios. But the other percentages are not mapped in detail by outcome, the way it is done with nuclear scenarios, where people look at this more closely. That is possible. People do not go deeper, they do not use this method.
When I heard it for the first time, I did not trust my ears. I thought this was fake news. Unfortunately not.
If 25 percent means the extinction of humanity, does 75 percent mean paradise?
Whoever names twenty percent extinction does not claim to have the rest under control. That is a deception, a false assumption. It could mean that most people die and only a few survive. Or that most people suffer. That is not clear.
Or does it mean the collapse of our systems, the overload of our society, chaos and unrest, total concentration of power and control? Or pure happiness and sunshine?
In the interviews nobody asks about it. No reporter, no senate, no government.
What is actually connected? Production systems, communication systems, payment systems, navigation systems. Also facilities operated under hazard supervision, nuclear, chemical, biological. In principle, the digital world is the habitat of silicon intelligence.
Can we still maintain our digital infrastructure on our own at all when an AI is in control? Do the ten best programmers then go there with a keyboard and upload some code?
We can no longer keep control over the digital infrastructure. And so everything we have digitized over the last decades actually leads us into harm, as long as we have no analog alternatives now. You need an analog, a second fallback level, a copy.
We should prepare for this, that is how I think about it. And make the same effort as with a medical approval, with a medicine, or with other things where the risk is small by comparison.
I wrote a book about this. The English edition was translated with AI; how that was done is described here: asiresilience.org/wie-dieses-buch-entstanden-ist
Tell me where I am wrong.
Prior work on splitting the outcomes (this list was compiled with Claude, not written by me):
Paul Christiano, "My views on doom", LessWrong, April 27, 2023: splits his estimate into takeover, later takeover, most humans dying, and 46 percent that humanity has irreversibly messed up its future within ten years of powerful AI.
Joseph Carlsmith, "Is Power-Seeking AI an Existential Risk?", arXiv 2206.13353: six premises, updated from about 5 to over 10 percent with stated reasons.
Nuclear precedent: WASH-1400 (1975) used event and fault trees to give probabilities over consequence classes; followed by NUREG-1150 and SOARCA.
Wisakanto et al., arXiv 2504.18536 (2025): a probabilistic risk assessment workbook for AI.
None of the frontier labs whose leaders quote 10 to 25 percent has published such a split.
