yeah it's hard to claim that it assesses "the truth". I am doing a research project in this exact field and we believe this to be a semi-hard AI problem with very little chance of being approximable.