Towards Building Safe And Robust Human Ai Systems