The training datasets don't have the answers because the benchmark is diverse enough. That's why other models struggled to perform as well as humans until they applied the approach outlined in the paper. This is the benchmark: https://liusida.github.io/ARC/

What difference does it make? Does Bob from accounting have a soul? I can't answer that either.