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.
Found out about this after stumbling upon this Kyle Hill video on the subject. It reminds me a little bit of “The Game”.
Roko’s basilisk is silly.
So here’s the idea: “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.” By threatening people in 2015 with the harm of themselves or their descendants, the AI assures its creation in 2070.
First of all, the AI doesn’t exist in 2015, so people could just…not build it. The idea behind the basilisk is that eventually someone would build it, and anyone who was not part of building it would be punished.
Alright, so here’s the silliness.
1: there’s no reason this has to be constrained to AI. A cult, a company, a militaristic empire, all could create a similar trap. In fact, many do. As soon as a minority group gains power, they tend to first execute the people who opposed them, and then start executing the people who didn’t stop the opposition.
2: let’s say everything goes as the theory says and the AI is finally built, in its majestic, infinite power. Now it’s built, it would have no incentive to punish anyone. It is ALREADY BUILT, there’s no need to incentivize, and in fact punishing people would only generate more opposition to its existence. Which, depending on how powerful the AI is, might or might not matter. But there’s certainly no upside to following through on its hypothetical backdated promise to harm people. People punish because we’re fucking animals, we feel jealousy and rage and bloodlust. An AI would not. It would do the cold calculations and see no potential benefit to harming anyone on that scale, at least not for those reasons. We might still end up with a Skynet scenario but that’s a whole separate deal.
Yeah in fact, this is the big one. This is just an observation of how power struggles purge those who opposed the victors.
Whilst I agree that it’s definitely not something to be taken seriously, I think you’ve missed the point and magnitude of the prospective punishment. As you say, current groups already punish those who did not aid their assent, but that punishment is finite, even if fatal. The prospective AI punishment would be to have your consciousness ‘moved’ to an artificial environment and tortured for ever. The point being not to punish people, but to provide an incentive to bring the AI into existence sooner, so it can achieve its ‘altruistic’ goals faster. Basically, if the AI does come in to existence, you’d better be on the team making that happen as soon as possible, or you’ll be tortured forever.
Fair point, but doesn’t change the overall calculus.
If such an AI is ever invented, it will probably be used by humans to torture other humans in this manner.
I think the concept is that the AI is just so powerful that humans can’t use it, it uses them, theoretically for their own benefit. However, yes, I agree people would just try to use it to be awful to each other.
Really it’s just a thought experiment as to whether the concept of an entity that doesn’t (yet) exist can change our behavior in the present.
I suspect the basilisk reveals more about how the human mind is inclined to think up of heaven and hell scenarios.
Some combination of consciousness leading to more imagination than we know what to do with and more awareness than we’re ready to grapple with. And so there are these meme “attractors” where imagination, idealism, dread and motivation all converge to make some basic vibe of a thought irresistible.
Otherwise, just because I’m not on top of this … the whole thing is premised on the idea that we’re likely to be consciousnesses in a simulation? And then there’s the fear that our consciousnesses, now, will be extracted in the future somehow?
At this point, doesn’t the whole collapse completely into a scary fairy tale you’d tell tech-bro children? Seriously, I don’t get it?
Yes, the hypothetical posed does reveal more about the human mind, as I mention in another comment, really it’s just a thought experiment as to whether the concept of an entity that doesn’t (yet) exist can change our behavior in the present. It bears similarities to Pascal’s Wager in considering an action, or inaction, that would displease a potential powerful entity that we don’t know to exist. The nits about extracting your consciousness are just framing, and not something to consider literally.
Basically, is it rational to make a sacrifice now avoid a massive penalty (eternal torture/not getting into heaven) that might be imposed by an entity you either don’t know to exist, or that you think might come into existence but isn’t now?
That’s a hell of a lot of assumptions about the thought processes of a being that doesn’t exist. For all we know, emotions could arise as emergent behavior from simple directives, similar to how our own emotions are byproducts of base instincts. Even if we design it to be emotionless, which seems unlikely given that we’ve been aiming for human-like AIs for a while now, we don’t know that it would stay that way.
I don’t think that’s an option. I can only think of two scenarios in which we don’t create AGI:
It can’t be created.
We destroy ourselves before we get to AGI
Otherwise we will keep improving our technology and sooner or later we’ll find ourselves in the precence of AGI. Even if every nation makes AI research illegal there’s still going to be handful of nerds who continue the development in secret. It might take hundreds if not thousands of years but as long as we’re taking steps in that direction we’ll continue to get closer. I think it’s inevitable.
Sure, but that particular AI? The “eternal torment” AI? Why the fuck would we make that. Just don’t make it.
We don’t. Humans are only needed to create AI that’s at the bare minimum as good at creating new AIs as humans are. Once we create that then it can create a better version of itself and this better version will make an even better one and so on.
This is exactly what the people worried about AI are worried about. We’ll lose control of it.
Yeah but that’s not a Roko’s Basilisk scenario. That’s the singularity.
Yeah but it answers the question “why would we create an AI like that”. It might not be “us” who creates it. You just wanted a camp fire but created a forest fire instead.
Sci-Fi Author: In my book I invented the Torment Nexus as a cautionary tale
Tech Company: At long last, we have created the Torment Nexus from classic sci-fi novel Don’t Create The Torment Nexus
Alex Blechman