1. A giant asteroid is hurtling toward the planet, and will arrive very soon. If we mess up and fail to deflect the asteroid, then we all die. This is presumably a classic one-shot scenario, and perhaps few people disagree with that assessment, but I’m not sure.
Couldn’t one have multiple parallel projects to deflect it, thereby giving multiple shots? The difference with AI safety being that if there are multiple parallel projects to build a safe ASI, the fastest one is the one that determines the outcome.
Couldn’t one have multiple parallel projects to deflect it, thereby giving multiple shots? The difference with AI safety being that if there are multiple parallel projects to build a safe ASI, the fastest one is the one that determines the outcome.