The first thing that is needed or this to work, is for the part of universe representing the human to be separable from the rest of the universe (this is not true with superdeterminism, which requires choices that experimenters make to be correlated with the quantum states of particles they are studying, but then almost no one takes superdeterminism seriously).
This separability would allow to talk about choice, as it allows to modify or swap the human in question and recompute the future in the same universe where the human may make another choice.
The second thing is the conjecture of computational irreducibility: that is for sufficiently complex systems like humans and for sufficiently long timeframes, the only way to predict future state is to evaluate the system. This conjecture seems plausible because even cellular automata in chaotic state, do not appear to have any simplified method of evaluation
If this is true, you may be able to easily predict some of the choices human will make based on his state several seconds before that, but for longer time intervals the only way is to let the human live and make the choice (even if it lives in your computer simulation).
