As Americans enter the midterms not entirely knowing what these giant ugly data centers are, or what they do, or that they’re using them right now, with likewise no real sense of what artificial intelligence even really is, or how it works, or if it might improve their lives, the small community of technologists, businessmen, and researchers most proximate to this technology has frustratingly mainstreamed an alarming question: is this new “intelligence,” which we are actively building ourselves, going to kill every single person on the planet — imminently?

If we keep building artificial intelligence, are you, and everyone you know and love, going to die?

Artificial intelligence poses several real and obvious risks, from the integrity of our internet, and by extension really almost everything that upholds our modern way of life, to the future of war. But I think our risk of accidentally building something roughly as powerful as God, which is, without exaggeration, how most major safetyists conceive of this threat, and of that God killing every human alive as it rearranges all matter in the galaxy in such a way as it finds pleasing, is negligible.

This may seem to you an odd thing to argue. It may even seem unfair. Surely, I must be cherry-picking the most extreme safetyist views to make them seem insane. But while I disagree with them to some extent, I don’t think AI safetyists are insane. And this is not a fringe idea.

The motivating safetyist fear, which has now made its way to your evening news, is the runaway artificial superintelligence (ASI) taking over a company, or companies, then infrastructure throughout the material world, where it will rapidly conquer our planet and kill everyone. Yes, they will talk about a terrorist or whatever learning how to build a bioweapon. Yes, they will contribute to the latest “chatbots made me do it” crime or suicide hysteria. And yes, they will fret, as do I, about the future of the internet. But their central concern is we are accidentally building a malevolent God. The writing on this subject is extensive. Most of it is blogs.

Henceforth, we will refer to this field of thought as the Doom.

Before I get into my analysis, I think it’s probably helpful to share a reading list, so you have some sense of where I’m coming from.

I first became interested in existential risk over a decade ago, when a fascination with the concept of simulated realities led me to Professor Nick Bostrom, who not only popularized that strange thought space, but another: how likely are we all to die, and at the hands of machine superintelligence specifically?

While I was already long surrounded, then, by enthusiastic optimists excitedly proselytizing in the space of artificial intelligence, or building in it, Bostrom’s book Superintelligence (2014) was my first introduction to how this might turn out very badly, and how we might prevent the very badly thing from happening. I had read a bit of Kurzweil already, specifically and obviously The Singularity is Near (2006), which was required reading at the time (at least among interesting people), and his far less popular but very interesting How to Create A Mind (2012), which tried to get at the question of what intelligence and thinking really are, and how to build something that “thinks.”

When I returned to the topic more recently, the Doom had totally taken hold of a community formerly shaped of Singularity enthusiasts, now entirely run by a small, priestly class of “rationalists,” most of whom believed we were imminently (a year or two out — before the next presidential election) in danger of de facto losing sovereignty to a superintelligence, which was going to kill everyone (just a handful of years after that).

What follows is by no means an exhaustive list of writing on the subject. Most thinkers in the space of AI safety have been at this for close to a decade, and especially these last five years. You could read lengthy posts on the subject for as much time, and still not read everything. But the seminal works are probably: first, If Anyone Builds it, Everyone Dies, a book written by Eliezer Yudkowsky and Nate Soares (2025), and second, AI 2027, a very long essay filled with hard predictions written by Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland, and Romeo Dean (2025).

The former work is written by rationalists. In the words of Eliezer, one of the key figures in this community, these are people dedicated to the employment of reason in “systematically improving their beliefs,” and then “systematically achieving their values.” But you maybe know them as the OG freaks. The biggest, nastiest brains around. The doomsday sex cult people (not all of them, and only allegedly). People really seem to hate them, but they are, frankly, really interesting. And to his credit, Eliezer understood AI would matter very early. Earlier than almost any other safetyist. Earlier than almost every person building AI today.

The latter work, AI 2027, was written by effective altruists, or writers very adjacent to effective altruism. The well-meaning charity enjoyers who transitioned to full-time Doom slaying. The wonks. The power seekers (for good reasons, they promise!). Some of whom, I’m sure, are very fine people.

A handful of other immediately relevant readings: “Superintelligence FAQ,” by Scott Alexander (2016), “AI Could Defeat All Of Us Combined,” Holden Karnovsky (2022), “Without specific countermeasures, the easiest path to transformative AI likely leads to AI takeover,” by Ajeya Cotra (2022), “The Hugging Face Attack Surprised Me,” Ajeya Cotra (2026), and “We Must Pace the Frontier,” Dario Amodei (2026). I also think the debate, “Yudkowsky vs Hanson — Singularity Debate” (2017), is worth watching. Both men do a great job making their respective cases.

The selection above highlights most of the major Doom prophets, as well as two of the most prominent “safety experts,” both explaining the recent Hugging Face incident as well as their own thinking generally, and of course the CEO of Anthropic, Dario, as he called for a dramatic slowdown on advanced AI development.

Now, what do these people believe?

Most adherents of the Doom arrive at our extinction from the same handful of assumptions, themselves each constructed of assumptions: