A machine that could answer arbitrary binary questions about the future might lead to paradoxes. An alternative formulation: upon being asked a question, the machine simulates two versions of the universe, one in which it answered 'true' and one in which it answered 'false'. It chooses one of these universes at random to base its answer on.

This results in no paradox: the answer can never be dependent on itself, because it has been sampled from a fair distribution of outcomes.

A surprising thing that pops out of this thought experiment: the more you try to create a universe in which the subject of the question is in a specific state, the more the answer you originally got from the machine becomes more likely to have been random in the first place. The machine remains more accurate for questions in which acting upon the answer does not encourage a change in the answer (i.e: the machine is very accurate when answering questions about things you couldn't possibly affect).