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”.
it has been said before and i’ll say it again: Pascal’s wager for tech bros
but not as easily dismissable
Roko’s Basilisk hinges on the concept of acausal trade. Future events can cause past events if both actors can sufficiently predict each other. The obvious problem with acausal trade is that if you’re the actor B in the future, then you can’t change what the actor A in the past did. It’s A’s prediction of B’s action that causes A’s action, not B’s action. Meaning the AI in the future gains literally nothing by exacting petty vengeance on people who didn’t support their creation.
Another thing Roko’s Basilisk hinges on is that a copy of you is also you. If you don’t believe that, then torturing a simulated copy of you doesn’t need to bother you any more than if the AI tortured a random innocent person. On a related note, the AI may not be able to create a perfect copy of you. If you die before the AI is created, and nobody scans your brain (Brain scanners currently don’t exist), then the AI will only have the surviving historical records of you to reconstruct you. It may be able to create an imitation so convincing that any historian, and even people who knew you personally will say it’s you, but it won’t be you. Some pieces of you will be forever lost.
Then a singularity type superintelligence might not be possible. The idea behind the singularity is that once we build an AI, the AI will then improve itself, and then they will be able to improve itself faster, thus leading to an exponential growth in intelligence. The problem is that it basically assumes that the marginal effort of getting more intelligent grows slower than linearly. If the marginal difficulty grows as fast as the intelligence of the AI, then the AI will become more and more intelligent, but we won’t see an exponential increase in intelligence. My guess would be that we’d see a logistical growth of intelligence. As in, the AI will first become more and more intelligent, and then the growth will slow and eventually stagnate.