Large language models can argue in convincing ways about politics, but humans dislike AI authors: implications for governance