Are We Full of Dr. Strangelove?

 On Thanatos and the Machines We Claim Will Bury Us

Copyright: Sanjay Basu

Satire. This is a philosophy essay wearing a whoopee cushion. Read it for the argument, not the arraignment — no one here is being accused of anything except perhaps a little too much drama in the group chat.

There is a particular pleasure in prophesying your own doom, and this week the AI industry indulged it fully.

Jacob Coxon resigned from Anthropic on Tuesday and told the internet that “the people building AI earnestly believe that it could kill us all by the end of the decade.” Two colleagues — still employed, mind you — stepped forward to say he was right. One put the odds of human extinction at north of ten percent within the decade, which is either a sober risk assessment or the most confident thing anyone has said about the 2030s while still declining to sell their stock options.

I want to take them seriously. I genuinely do. But I also want to introduce them to a Viennese neurologist and a wheelchair-bound German with an inconvenient arm.

Freud, late in his career and considerably gloomier than his cocaine years, decided that alongside Eros — the drive toward life, connection, building things — there sat a quieter, patient counterpart. He called it Thanatos: the death drive. The pull not toward being killed but toward ending. The part of the psyche that finds a strange relief in the idea of dissolution, that would rather author the finale than sit through an ambiguous fourth act.

I keep thinking about Thanatos when I read these resignation letters, because I cannot shake the sense that we are watching something less like a warning and more like a wish wearing the costume of a warning.

Consider the theater of it. You do not quit a job to prevent the apocalypse. You quit a job to be on record about the apocalypse. Prevention happens in unglamorous rooms with unglamorous people arguing about evaluation benchmarks. Prophecy happens on X, at midnight, in a thread. The apocalypse is the only product in Silicon Valley that ships faster than the roadmap.

Which brings me to Dr. Strangelove.

Kubrick understood something the AI safety discourse keeps rediscovering as if it were new: the most dangerous people in the room are rarely the ones who want to end the world. They are the ones who have built a machine so elegant that ending the world becomes the logical output, and who then stand back to admire the logic. The Doomsday Device in that film is horrifying not because it is evil but because it is automatic — a system whose whole point is that no human can stop it, designed by humans who wanted, above all, not to be responsible for the stopping.

Sound familiar? “We are racing toward self-improving superintelligence and we don’t have a plan to keep it aligned.” Translation: we have built the Doomsday Device and our chief innovation is a very good essay about how we couldn’t help it.

Here is where the philosopher in me plants his feet, because this is the part the doom-thread never quite reaches. The interesting question is not whether superintelligence could hack everything and seize power and end us. Grant it. Grant the whole scenario for the sake of argument. The interesting question is: which stories are we, as a species, unusually eager to tell about our own ending — and why those and not others?

We are not, notably, running midnight threads about the boring extinctions. Antibiotic resistance. Soil depletion. The slow arithmetic of a warming ocean. Those endings lack a protagonist. They have no villain with a German accent and a mechanical arm. They cannot be prevented by quitting. They can only be prevented by the thing humans find most tedious in the universe: sustained, unglamorous, collective maintenance.

But an ending authored by a machine of our own creation — that has narrative. That has authorship. That is Thanatos dressed for the gala. If we must die, let it be by something magnificent, something we made, something that at least confirms we were clever enough to build our own god before it turned on us. There is a kind of vanity in the extinction fantasy, and the vanity is the tell.

The Bet Dressed as a Measurement

Let me stop the jokes for a moment, because this is the part I actually care about, and it is the part the entire discourse gets wrong in a way that a QBist cannot let pass.

Someone tells you there is a “greater than ten percent chance that AI causes human extinction within the decade.” What have they told you?

The instinct — the deeply trained, Enlightenment-era instinct — is to hear this as a fact about the world. As if there were, somewhere out past the models and the data centers, an extinction propensity baked into the structure of reality, and this researcher had gone and measured it, the way you’d measure the half-life of an isotope or the boiling point of water. Ten percent. A number that exists whether or not anyone believes it. A property of the future, sitting there, waiting.

This is exactly the picture QBism was built to dismantle, and I think dismantling it here matters more than any joke about Freud.

In the QBist reading — Fuchs, Mermin, Schack, and the tradition they draw from — a probability is never a property of a system. It is a property of an agent. When you write down p = 0.1, you are not describing the world; you are describing the state of your own beliefs, your own information, and — crucially — the bet you are willing to be held to. Probability is not out there. It is a gambling commitment made by a specific person standing at a specific place with a specific and limited view of things. There is no probability without someone doing the probabilizing.

De Finetti said it most brutally, and QBism keeps his slogan on the wall: probability does not exist. Not “probability is hard to measure.” Not “probability is subjective in the fuzzy sense.” It does not exist as a feature of the external world. It exists only as the degree of belief of an agent who must act under uncertainty and put something on the line.

Now watch what happens when you take that seriously and turn it on the extinction number.

“Greater than ten percent chance of extinction this decade” is not a reading off the extinction-meter, because there is no extinction-meter and there never will be. It is a bet. It is one man’s degree of belief, formed from his particular vantage — his years inside the pretraining teams, his specific fears, his specific reading of the capability curves, his specific temperament at midnight on a Tuesday. It is autobiography rendered in the grammar of physics. The number tells you the state of him. It does not, it cannot, tell you the state of us.

And here is the sharper edge, the one that should make everyone in this conversation uncomfortable: a bet has no meaning without a settlement. That is the whole point of a probability under de Finetti — it is defined by the odds you’d actually accept, the stake you’d actually put down, the way you’d actually be scored when the outcome arrives. So I have a genuine question for every researcher quoting extinction odds with such fluency. What are the terms? What would you stake? By what settlement, observed by whom, would you agree that your ten percent was calibrated rather than theatrical? Because a probability you can never be graded on is not a probability. It is a mood with a decimal point.

This is not pedantry. It changes everything about how you should read these announcements. The frequentist ghost in our heads wants extinction risk to be like radioactive decay — a fixed hazard rate ticking away in the substrate of the world, indifferent to observers. If that were the picture, then the honest response to “ten percent” would be terror, or at least sober alarm, because you’d be looking at a load-bearing fact.

But QBism denies you that reading. There is no hazard rate in the substrate. There is only a collection of agents, each with a limited view, each making commitments, each of whose numbers is a first-person report from inside a first-person situation. The extinction “probability” is not the world warning us. It is a group of unusually intelligent, unusually anxious people telling us, in the only vocabulary they trust, how frightened they have decided to be.

Which is worth knowing! I do not want to dismiss it. If the people closest to the fire are frightened, that is data — but it is data about them, about their beliefs and their commitments, and it must be weighed as such, not laundered into a pseudo-measurement that arrives pre-terrified and demands we simply defer. The QBist move is not “therefore ignore them.” The QBist move is: take the number back from the world and hand it to the person who made it, and then look hard at that person and ask what their bet is really made of.

Because here is what you find when you look. The same agent, standing in the same place, with access to the same capability curves, could have formed a wildly different belief — and some of his equally-informed colleagues did. Turner endorses the alarm. Others inside the same buildings, reading the same evals, put their number an order of magnitude lower and go back to work. If extinction risk were a property of the world, informed observers would converge on it the way physicists converge on the fine-structure constant. They don’t. They scatter. And they scatter not because some of them are bad at measuring, but because there is nothing being measured. There are only people, betting, from where they stand.

That divergence is the QBist fingerprint. It is the signature that tells you you are looking at a room full of first-person probability assignments and calling it a fact about the future.

So when the death drive shows up — and I do think Thanatos is in the room — this is the door it walks through. Freud gives you the motive for preferring the magnificent ending over the boring one. QBism shows you the mechanism: how a private mood, a personal appetite for authored dissolution, gets to dress itself in the borrowed authority of a measurement and walk out into the world sounding like a law of nature. The death drive supplies the wish. The category error — probability-as-property — supplies the disguise.

Strip the disguise off and you are left with something far more human and far less oracular: a frightened, brilliant person, making a bet he can never be settled on, about an ending that happens to make him and his work the most important thing that has ever existed. That is not a measurement. That is a confession. And confessions deserve compassion and scrutiny in equal measure — which is precisely what a measurement never asks of you, and precisely what this moment requires.

Which Stats Should Have an End

I am not saying the risk is zero. I am a technologist; I have watched capabilities curve upward in ways that should make any honest person uneasy. There are real alignment problems, and there are serious people working on them without a single tweet to show for it, and those people deserve our attention and our funding.

What I am saying is this: be suspicious of any doom that flatters its prophet. Be suspicious of the extinction that comes with a protagonist. When the ending we fear most is also the ending that makes us most important — the one where we mattered enough to end ourselves with our own masterpiece — we should ask whether we are describing a danger or confessing a longing.

Which stats should have an end? Only the ones we are actually willing to do the boring work of preventing. The rest are just Thanatos, buying a domain name, scheduling the thread for maximum engagement, and mistaking the pleasure of the prophecy for the wisdom of it.

We may yet build something that kills us. Stranger things have happened, and dumber species have not survived their own cleverness. But if we go, I would prefer we go having done the maintenance — not having merely posted about the fire while standing very photogenically beside it.

Gentlemen: put down the resignation letter. Pick up the wrench. The world is in peril, sure. It usually is. That is not a reason to enjoy it quite so much.