"Doomer" Is Not
an Argument

I have been called a doomer more times than I can count. So has nearly everyone who works on AI safety in public. The label arrives before the argument does, and that is the point of it. Before I have said a word about probabilities or timelines or what "control" would even mean, I have already been sorted into a bin — the frightened people, the ones who want to slow things down because they cannot handle the future. Sorted, and therefore dismissed.

It is worth looking closely at how that trick works, because it is the most effective move the accelerationist side has, and it has almost nothing to do with whether they are right.

Where the word came from

Around 2022 and 2023, the AI safety debate stopped being an in-house conversation among researchers and spilled onto social media, where arguments get teams and teams get names. The accelerationist camp — people who believe AI should be built as fast as possible — took the suffix "e/acc" as a badge and wore it in their usernames. They needed a name for the other side too. "Doomer" was it, usually paired with "decel," short for deceleration, the sin of wanting to go slower.

The mood hardened that October, when Marc Andreessen published "The Techno-Optimist Manifesto." It reads less like an argument than a creed, and it includes a list of what it calls the enemy — among the entries, "existential risk" and "the precautionary principle," the plain idea that you might want to understand a powerful technology before you deploy it at scale. Once concern itself is filed under "the enemy," the people who hold it need only a shorter label for daily use. Doomer did the rest.

I want to be fair to the movement it came from, because there is a real argument underneath, and we have taken it seriously at length elsewhere. Some accelerationists make substantive claims: that progress has been the great humanitarian force, that heavy regulation carries its own dangers, that you learn a technology by building it. Those are worth engaging. "Doomer" is not one of them. It is what gets used instead.

What the label actually does

A doomer is not a claim. It is a character. That is the whole function of the word: it turns an argument into a personality type, and personality types do not have to be answered, only diagnosed.

Look at what the safety side is actually saying. Roughly this: the chance of catastrophe from a superintelligent system is high enough — not certain, high enough — that we should build the ability to control these systems before we build the systems. You can push on every part of that. You can argue the probability is lower than we think, that the timeline is longer, that "control" is the wrong frame, that governance will do more harm than the risk it addresses. All of that is a real conversation, and some of it is a conversation worth losing.

"Doomer" skips it. The word says: you are the kind of person who is scared of tomorrow. Your position is a mood, not a thesis, and moods do not need rebutting. It is the oldest maneuver in rhetoric, the argument against the person rather than the point, wearing new clothes. The elegance of it is that it never has to touch the actual disagreement. It relocates the whole fight from "is this true" to "what kind of person believes it," where the accelerationist has already decided the answer.

The word turns an argument into a personality type — and personality types do not have to be answered, only diagnosed.

Why it lands

The label works because it is built out of things people already want to believe about themselves.

Start with the obvious: no one wants to be the pessimist. Optimism reads as confidence and drive; caution reads as timidity and fear. Frame the choice as builder versus doomer and most people already know which chair they would rather sit in. The framing has done its work before a single fact reaches the table.

It draws on something deeper, too — a genuine and largely honorable story America tells about itself. Progress is good. The people who feared the railroad, the printing press, the electric light were wrong, and history remembers them as fools. The accelerationist takes that story off the shelf whole and casts the safety advocate as the newest name on the list of frightened people who will be embarrassed by the future. It is a powerful story precisely because it is often true. The move is to assume, without argument, that it is true again this time — that superintelligence is the printing press and not something new. Whether it is the same story is the entire question, and the label exists so the question never gets asked.

Who the word gets aimed at

Line up the people it is thrown at and the caricature comes apart in your hands.

Geoffrey Hinton left Google in 2023 so he could talk about the risk without a corporate minder; he is as close to a founding father as the field has. Yoshua Bengio, among the most-cited researchers alive, arrived at the same worry from his own work. In May 2023, hundreds of scientists and executives — including the leaders of the very labs sprinting to build the technology — put their names to a single sentence placing extinction risk from AI in the same category as pandemics and nuclear war. These are not people afraid of technology. They built this one. Whatever you want to call their position, "scared of the future" is not a description of it, it is a way of not listening to it.

The word also flattens a wide spectrum down to a dot. There is enormous distance between someone who thinks catastrophe is nearly certain and someone who puts it at a few percent and wants insurance anyway — the distance between a prophet and an actuary. Both get filed as doomers. The label cannot see the difference because it was never built to; its job is to make a technical, varied, often quite moderate set of views look like one wild-eyed cult. We wrote a whole piece on the real range of expert estimates for exactly this reason. The range is the thing the label is designed to erase.

The tell

Here is what I keep returning to. A movement confident in its answer to the safety question would not need a nickname for the people asking it.

If catastrophic risk really is near zero, that is a claim you can defend in the open. Show the reasoning. Explain why the researchers who built the systems are wrong about the systems they built. Make the case that alignment turns out to be easy, or that greater intelligence reliably brings greater care, or that the race between labs somehow corrects itself before anyone cuts a fatal corner. Those are hard arguments, but they exist, and when accelerationists actually make them we can have the debate — the one we tried to have in full.

The word "doomer" is what fills the space where that argument would otherwise be. You reach for a label when you would rather not address the point. Naming your opponent is cheaper than refuting him, and from a distance it looks enough like a win that the crowd often cannot tell which one it just watched.

How to answer it

Refuse the frame. The instant the conversation becomes builder-versus-doomer, the substance has already left the room. Drag it back to the object level with one question: what probability are you putting on catastrophic risk, and why that number? Almost no one who swings the word as a weapon can answer, because the word is there so they never had to. If you want to see how such a number is actually built rather than asserted, we made a P(doom) calculator that walks through the pieces.

Then pull the label off the claim. "Doomer" staples a caricature — a gloom-addicted pessimist who hates progress — onto a claim that is usually far narrower and far duller: that this particular technology carries a particular risk that warrants particular care. Peel them apart. You can love technology, want the abundance the accelerationists describe, believe the future can be extraordinary, and still think we ought to be able to steer a superintelligence before we switch one on. Those are not opposing positions. The case for the good future runs straight through getting this part right, not around it.

I don't think most people who use the word are arguing in bad faith. The framing is genuinely seductive, and confidence is easy to mistake for correctness, especially when it comes with a good story attached. But when the strongest move a movement can make against its critics is to hand them an unflattering name, that tells you something about the ground it is standing on. Our case does not run on fear and it does not run on hating progress. It runs on one sturdy, unglamorous piece of prudence: for a mistake you cannot take back, you build the brakes first. Call that doom if you want. It has no bearing on whether it is right.