Roko’s Basilisk

Source: https://www.lesswrong.com/tag/rokos-basilisk Author: Rob Bensinger, community Date: 2025-09-19

Summary

Overview of Roko’s Basilisk — a thought experiment suggesting that a future AI might punish people who knew about it but didn’t help bring it into existence. Controversial thought experiment in AI safety circles.

Key Claims

  • The basilisk: a hypothetical future AI that, to incentivize cooperation, threatens to simulate and torture people who knew about AI risks but didn’t help prevent bad AI or accelerate good AI.
  • Why it’s notable: it caused significant controversy on LessWrong — the original post was deleted because thinking about it was claimed to risk harm (acausal blackmail).
  • The philosophical issues: rests on controversial decision theories (timeless decision theory, causal vs. evidential decision theory).
  • Why most people dismiss it: requires many implausible assumptions; most credible AI safety researchers don’t think it’s a real concern.
  • Why it’s mentioned here: it illustrates galaxy-brained reasoning — a chain of logic that seems internally consistent but leads to absurd conclusions.

Concepts

  • AI Alignment — illustrates edge cases in alignment reasoning