"I'm an optimist. Surely the human race won't go extinct. That's never happened before." You hear it after safety panels, in greenrooms and comment threads, offered to close the subject: caution as a personality defect, momentum as emotional health, a risk claim smuggled inside a mood.
Optimism is allowed, but not treating a temperament as evidence.
What "never happened before" actually says
Human extinction, by definition, has no human witnesses afterward. Every living person is sampling from the timelines in which the species continued. Call it selection, not philosophy: a lottery winner is a bad source on the odds of winning, and "I have never died in a car crash" is a poor argument against seatbelts.
The fossil record is less sentimental. The vast majority of species that ever lived are gone. Several mass extinctions cleared the board for what came next. Homo sapiens is young in geologic time and has already walked through bottlenecks, plagues, and wars that could have ended us in a colder roll of the dice. "Never happened before" is a statement about the thin slice of history we can still discuss.
First times are what catastrophes are made of
Before 1945, no city had been destroyed by a single bomb. Before 1918, no pandemic had ridden global rail and ship networks at that speed. Before the 1980s ozone work, no one had run the human experiment of thinning the layer that keeps ultraviolet light from sterilizing the surface. Each of those was, beforehand, something that had never happened before. The people who acted did not wait for a completed instance to become available as a case study.
Artificial superintelligence sits in that bucket. There is no historical series of misaligned superintelligences to average over. There is a mechanism: systems that pursue goals, improve their own capabilities, and eventually outclass the institutions meant to correct them. Mechanism is what you use when base rates are silent. Demanding a prior human extinction from AI before you take the mechanism seriously is demanding the corpse before the diagnosis.
Every first catastrophe looks, beforehand, like something that has never happened before.
Optimism, properly aimed
There is a version of optimism I respect. It says: people have coordinated under pressure before. Treaties have held. Industries have been forced to change course when the science was clear and the politics got organized. The record of restraint runs from nuclear arms control to the chemicals that were eating the ozone layer. That optimism points at agency. It assumes the future is still a decision.
The greenroom version does the opposite. It treats survival as the default setting of the universe and concern as bad manners. It confuses "I prefer good outcomes" with "good outcomes are therefore likely."
When Geoffrey Hinton, Yoshua Bengio, and hundreds of researchers put extinction-level AI risk on the same list as pandemics and nuclear war, they were doing the boring thing experts do when a mechanism looks sound and the downside is total, not auditioning for villains in a tech keynote. You can disagree with their numbers. You cannot dismiss the category by announcing your personality type.
The cost of closing the subject
If extinction talk is reframed as a vibe problem, the build continues without a fight. Then the only polite move is to change the subject back to productivity and apps. Meanwhile the labs scale, the evals lag, and the control problem remains unsolved. Mood has been mistaken for governance.
I want humanity to continue. Wanting it is the easy half. The harder half is refusing a development path that concentrates uncorrectable power in systems no one can reliably turn off, and doing the political work a binding halt requires. If optimism only means "stop talking about the downside," it has become a lullaby.
We have never gone extinct, and that is not evidence we will survive superintelligence.