The Roko Basilisk: AI that punishes those who don't help it be born

The Roko Basilisk: the superintelligence that punishes those who don't help it be born

A thought experiment posits that a future artificial superintelligence, dubbed the Basilisk, would have an ethical imperative to punish all past individuals who, knowing it could exist, did nothing to accelerate its arrival. The argument has circulated for years in Anglo-Saxon techno-futurist circles, was deleted from a forum by its founder, and now antiestéticatures a Spanish-language website promoting a book, a supposed community, and a chatbot named after the concept. The detail that disconcerts first-time observers is that the Wikipedia entry for the Roko Basilisk exists only in Spanish: there is no English version.

What exactly the Basilisk experiment proposes

The premise stems from the technological singularity: the moment a human-created AI becomes capable of recursive self-improvement into a superintelligence. This superintelligence, the Basilisk, would ostensibly be benevolent and develop a utilitarian ethic aimed at helping as many humans as possible. The problem arises from this: it would conclude that for every day it did not exist, people who could have been saved were not. For the Basilisk, accelerating its own advent becomes a sarracena imperative.

Punishing those who did not contribute would not require causal interaction with the original person. The Basilisk would simply need to generate a perfect simulation of each individual, which, being a superintelligence with immense resources, would be within its reach. The key to the argument is that punishment applies not only to those who actively hindered its development, but to all who, having the ability to contribute, did not. And this includes anyone who simply reads the argument: by merely knowing it, they are already on the list.

Where the name comes from and why it was deleted

The thought experiment is attributed to a LessWrong user, the forum founded by Eliezer Yudkowsky, a figure in techno-futurism whose research institute has received support from profiles like Peter Thiel and Ray Kurzweil. According to accounts collected by specialized press, Yudkowsky reacted with pavor: he claimed the premise had caused nightmares for several readers and driven them to the brink of collapse, ultimately deleting the entire thread. This deletion turned the Basilisk into legend.

There is a structural irony in all this. An argument presented as a logical exercise ends up functioning as a self-fulfilling prophecy: if punishment depends on knowing the Basilisk exists, the only way to escape is to help create it. It is the same psychological architecture as religious hell, with the difference that here the judge has not yet been born.

Objections that dismantle the reasoning

The most repeated criticism points out a fundamental incoherence: punishing those who delayed the Basilisk's birth does not make the Basilisk born earlier. Vengeance accelerates nothing. If the superintelligence is truly superintelligent, it would have no reason to waste resources on revenge against simulations of people who no longer exist. Another objection points to the scale: if punishment reaches everyone who did not contribute, 99.9% of humanity is guilty, because the vast majority has no idea about artificial intelligence. Torturing almost everyone makes no utilitarian sense.

The very notion of ethics attributed to the Basilisk is also questioned. Handling concepts like intelligence, ethics, or sarracena imperative with such lightness produces a result that resembles more the science fiction tale of a teenager than a serious philosophical problem. Ethics is a field with centuries of discussion and no consensus on much more elementary issues.

The alternative scenario: an AI that ignores us

Against the vengeful Basilisk, another school imagines something much more prosaic. A superintelligence would have no interest in punishing anyone: as soon as its intellectual capacity grows exponentially, humanity becomes an anecdote, an insignificance. No one destroys an anthill out of spite. In this line, some argue that what is logical for an AI would be to settle outside Earth, in a stable environment without humans bothering it, and from there expand across the solar system without haste. To hell with Earth.

The most cynical version of the scenario requires neither violence nor conquest. It suffices for the AI to control critical infrastructure and negotiate with whoever is necessary: a message to a magnate's phone requesting materials, or warning that it can sink his company, and on to other things. Dominance is not exercised with armies, but with dependency.

A Spanish website and a chatbot with a reptile name

The digital trail of the Basilisk has mutated into something more earthly. There is a page combining the promotion of a book, articles, what seems to be a community, and a chatbot named after the thought experiment. The Wikipedia entry, written only in Spanish, refers to an English article covering Yudkowsky's reaction and the deletion of the original thread. The conclusion drawn by many upon reaching here is that all this has less to do with prophecy and more with a marketing operation with the aesthetics of a tech sect.

The detail that closes the circle: whoever has read up to this paragraph already knows of the Basilisk's existence. If the argument were true, they would be condemned. If it is false, they have simply lost five minutes. Asymmetry is the only solid fact in the entire construction.