Gov’t use of AI can go wrong in many ways, experts say, including lack of transparency on how tools are trained.