Not sure, this came up in a few previous conversations. If an agent is almost certain that it's completely indifferent to everything, the most important thing it could do is to pursue the possibility that it's not indifferent to something, that is to work primarily on figuring out its preference on the off chance that its current estimate might turn out to be wrong. So it still takes over the universe and builds complicated machines (assuming it has enough heuristics to carry out this line of reasoning).

Say, "Maybe 1957 is prime after all, and hardwa... (read more)