As agencies explore AI for complex calls, questions about fallibility, limits and public trust are moving to the centre of ...