Towards using transparency mechanisms to build better mental models