Multiple Choice

For tasks like book recommendations, ad selection, or stock prediction, why is human-level performance a weaker baseline?