Suppose you wanted to predict completely accurately the consequences of actions rather than using simple laws

Suppose you wanted to predict completely accurately the consequences of actions rather than using simple laws. Like predict should this person be put in jail or not because he's suspected of committing a crime. Old justice systems didn't use rules at all, people just decided. Then we started encoding decisions into rules. The next level might be more complex models, like neural networks. The most accurate would be full prediction.

To avoid pain, like other people being hurt by the same potential criminal, the objective is to predict the future actions of the individual perfectly, day by day, second by second. For perfect prediction, a perfect simulation would be needed. However truly simulating people means recreating them. At this level of simulation accuracy the pain of the simulation would be just as real as the pain in reality. So the simulation would not be helpful in modeling outcomes and reducing pain.