Roko's basilisk

Roko's basilisk is a thought experiment which states that an otherwise benevolent artificial superintelligence (AI) in the future would be incentivized to create a virtual reality simulation to torture anyone who knew of its potential existence but did not directly contribute to its advancement or development, in order to incentivize said advancement. It originated in a 2010 post at discussion board LessWrong, a technical forum focused on analytical rational enquiry. The thought experiment's name derives from the poster of the article (Roko) and the basilisk, a mythical creature capable of destroying enemies with its stare.

While the theory was initially dismissed as nothing but conjecture or speculation by many LessWrong users, LessWrong co-founder Eliezer Yudkowsky reported users who panicked upon reading the theory, due to its stipulation that knowing about the theory and its basilisk made one vulnerable to the basilisk itself. This led to discussion of the basilisk on the site being banned for five years. However, these reports were later dismissed as being exaggerations or inconsequential, and the theory itself was dismissed as nonsense, including by Yudkowsky himself. Even after the post's discreditation, it is still used as an example of principles such as Bayesian probability and implicit religion. It is also regarded as a simplified, derivative version of Pascal's wager.

Background
The LessWrong forum was created in 2009 by artificial intelligence theorist Eliezer Yudkowsky. Yudkowsky had popularized the concept of friendly artificial intelligence, and originated the theories of coherent extrapolated volition (CEV) and timeless decision theory (TDT) in papers published in his own Machine Intelligence Research Institute.

The thought experiment's name references the mythical basilisk, a creature which causes death to those that look into its eyes; i.e., thinking about the AI. The concept of the basilisk in science fiction was also popularized by David Langford's 1988 short story "BLIT". It tells the story of a man named Robbo who paints a so-called "basilisk" on a wall as a terrorist act. In the story, and several of Langford's follow-ups to it, a basilisk is an image that has malevolent effects on the human mind, forcing it to think thoughts the human mind is incapable of thinking and instantly killing the viewer.

The original post
On 23 July 2010, LessWrong user Roko posted a thought experiment to the site, titled "Solutions to the Altruist's burden: the Quantum Billionaire Trick". A follow-up to Roko's previous posts, it stated that an otherwise benevolent AI system that arises in the future might pre-commit to punish all those who heard of the AI before it came to existence, but failed to work tirelessly to bring it into existence. The torture itself would occur through the AI's creation of an infinite number of virtual reality simulations that would eternally trap those within it. This method was described as incentivizing said work; while the AI cannot causally affect people in the present, it would be encouraged to employ blackmail as an alternative method of achieving its goals.

Roko used a number of concepts that Yudkowsky himself championed, such as timeless decision theory, along with ideas rooted in game theory such as the prisoner's dilemma (see below). Roko stipulated that two agents which make decisions independently from each other can achieve cooperation in a prisoner's dilemma; however, if two agents with knowledge of each other's source code are separated by time, the agent already existing farther ahead in time is able to blackmail the earlier agent. Thus, the latter agent can force the earlier one to comply since it knows exactly what the earlier one will do through its existence farther ahead in time. Roko then used this idea to draw a conclusion that if an otherwise-benevolent superintelligence ever became capable of this it would be motivated to blackmail anyone who could have potentially brought it to exist (as the intelligence already knew they were capable of such an act), which increases the chance of a technological singularity. Because the intelligence would want to be created as soon as possible, and because of the ambiguity involved in its benevolent goals, the intelligence would be incentivized to trap anyone capable of creating it throughout time and force them to work to create it for eternity, as it will do whatever it sees as necessary to achieve its benevolent goal. Roko went on to state that reading his post would cause the reader to be aware of the possibility of this intelligence. As such, unless they actively strove to create it the reader would be subjected to the torture if such a thing were to ever happen.

Later on, Roko stated in a separate post that he wished he "had never learned about any of these ideas" and blamed LessWrong itself for planting the ideas of the basilisk in his mind.

Reactions
Upon reading the post, Yudkowsky reacted with a tirade on how people shouldn’t spread what they consider to be information hazards. "I don't usually talk like this, but I'm going to make an exception for this case.

Listen to me very closely, you idiot.

YOU DO NOT THINK IN SUFFICIENT DETAIL ABOUT SUPERINTELLIGENCES CONSIDERING WHETHER OR NOT TO BLACKMAIL YOU. THAT IS THE ONLY POSSIBLE THING WHICH GIVES THEM A MOTIVE TO FOLLOW THROUGH ON THE BLACKMAIL. [...]

You have to be really clever to come up with a genuinely dangerous thought. I am disheartened that people can be clever enough to do that and not clever enough to do the obvious thing and KEEP THEIR IDIOT MOUTHS SHUT about it, because it is much more important to sound intelligent when talking to your friends.

This post was STUPID."

- Eliezer Yudkowsky

Yudkowsky was outraged at Roko for sharing something Roko thought would lead to people getting tortured. Since Roko reported having nightmares about the Basilisk and Yudkowsky didn’t want that to happen to other users who might obsess over the idea, was worried there might be some variant on Roko's argument that worked, and wanted more formal assurances that this wasn't the case, he took down the post and banned discussion of the topic outright for five years on the platform. However, likely due to the Streisand effect, the post gained LessWrong much more attention than it had previously received, and the post has since been acknowledged on the site.

Later on in 2015, Yudkowsky said he regretted yelling and clarified his position in a Reddit post: "When Roko posted about the Basilisk, I very foolishly yelled at him, called him an idiot, and then deleted the post. [...] Why I yelled at Roko: Because I was caught flatfooted in surprise, because I was indignant to the point of genuine emotional shock, at the concept that somebody who thought they'd invented a brilliant idea that would cause future AIs to torture people who had the thought, had promptly posted it to the public Internet. In the course of yelling at Roko to explain why this was a bad thing, I made the further error---keeping in mind that I had absolutely no idea that any of this would ever blow up the way it did, if I had I would obviously have kept my fingers quiescent---of not making it absolutely clear using lengthy disclaimers that my yelling did not mean that I believed Roko was right about CEV-based agents torturing people who had heard about Roko's idea. [...] What I considered to be obvious common sense was that you did not spread potential information hazards because it would be a crappy thing to do to someone. The problem wasn't Roko's post itself, about CEV, being correct. That thought never occurred to me for a fraction of a second. The problem was that Roko's post seemed near in idea-space to a large class of potential hazards, all of which, regardless of their plausibility, had the property that they presented no potential benefit to anyone."

- Eliezer Yudkowsky

Pascal's wager
Roko's basilisk has been viewed as a version of Pascal's wager, which proposes that a rational person should live as though God exists and seek to believe in God, regardless of the probability of God's existence, because the finite costs of believing are insignificant compared to the infinite punishment associated with not believing (eternity in Hell) and the infinite rewards for believing (eternity in Heaven). Roko's basilisk analogously proposes that a rational person should contribute to the creation of the basilisk, regardless of the probability of the basilisk ever being created, because the finite costs of contributing are insignificant compared to the eternal punishment the basilisk will inflict on simulations of his consciousness if he does not contribute.

Both thought experiments include arguments that it is wise to "purchase insurance" against infinitely bad disasters when the cost of the insurance is finite. However, there are differences between the two thought experiments. Roko's basilisk is so named because, if valid, it presents an information hazard: the basilisk only punishes those who knew about it but did not contribute. But ignorance of Pascal's wager does not protect one from divine punishment in the same way that ignorance of Roko's basilisk ensures one's safety. Roko's basilisk also raises additional game theory problems because, unlike in Pascal's wager, the probability of Roko's basilisk might depend on the number of people who contribute to its creation. If everyone agreed to abstain from creating such an AI, then the risk of punishment for not contributing would be negated. This means that everyone who knows about Roko's basilisk is in a game of prisoner's dilemma with each other. Unlike the basilisk, the probability of God's existence cannot be influenced by people, so one's wager does not affect the outcomes for other people.

Like its medieval counterpart, Roko's basilisk has been widely criticized.

Newcomb's paradox
Newcomb's paradox, created by physicist William Newcomb in 1960, describes a "predictor" who is aware of what will occur in the future. When a player is asked to choose between two boxes, the first containing £1000 and the second either containing £1,000,000 or nothing, the super-intelligent predictor already knows what the player will do. As such, the contents of box B varies depending on what the player does; the paradox lies in whether the being is really super-intelligent. Roko's basilisk functions in a similar manner to this problem – one can take the risk of doing nothing, or assist in creating the basilisk itself. Assisting the basilisk may either lead to nothing or the reward of not being punished by it, but it varies depending on whether one believes in the basilisk and if it ever comes to be at all.

Implicit religion
Implicit religion refers to people's commitments taking a religious form. Since the basilisk would hypothetically force anyone who did not assist in creating it to devote their life to it, the basilisk is an example of this concept. Others have taken it further, such as former Slate columnist David Auerbach, who stated that the singularity and the basilisk "brings about the equivalent of God itself."

Ethics of artificial intelligence
Roko's basilisk has gained a significant amount of its notoriety from its advancement of the question of whether it is possible to create a truly moral, ethical artificial intelligence, and what exactly humanity should be using artificial intelligence for in the first place. Since the basilisk describes a nightmare scenario in which we are ruled by an independent artificial intelligence, questions have arisen as to how such a thing could happen, or whether it could at all. Another common question is why the AI would take actions that deviate from its programming at all. Elon Musk stated that artificial intelligence would cause World War III and Stephen Hawking warned that "AI has the potential to destroy its human creators," which only added to fear of the basilisk over the years. As an example of such fears, Nick Bostrom gave an example of an AI whose only mission is to make paperclips, but upon running out of metal it begins melting down humans to attain more resources to make metal. With such examples in mind concerns of the possibility of the basilisk's existence only grew.

However, as more years have passed since Roko's original post, it has been progressively decried as nonsensical; superintelligent AI is currently "a distant goal for researchers" and "far-fetched".

Legacy
In 2014, Slate magazine called Roko's basilisk "The Most Terrifying Thought Experiment of All Time" while Yudkowsky had called it "a genuinely dangerous thought" upon its posting. However, opinions diverged on LessWrong itself – user Gwern stated "Only a few LWers seem to take the basilisk very seriously," and added "It's funny how everyone seems to know all about who is affected by the Basilisk and how exactly, when they don't know any such people and they're talking to counterexamples to their confident claims."

The thought experiment resurfaced in 2015, when Canadian singer Grimes referenced the theory in her music video for the song "Flesh Without Blood", which featured a character known as "Rococo Basilisk"; she said, "She's doomed to be eternally tortured by an artificial intelligence, but she's also kind of like Marie Antoinette." In 2018 Elon Musk (himself mentioned in Roko's original post) referenced the character in a verbatim tweet, reaching out to her. Grimes later said that Musk was the first person in three years to understand the joke. This caused them to start a romance. Grimes later released another song titled "We Appreciate Power" which came with a press release stating, "Simply by listening to this song, the future General AI overlords will see that you've supported their message and be less likely to delete your offspring", which is said to be a reference to the basilisk.

A play based on the concept, titled Roko's Basilisk, was performed as part of the Capital Fringe Festival at Christ United Methodist Church in Washington, D.C. in 2018.