Maybe we could use deep RNN to produce novel but likely soundbytes that politicians are likely to say to be included in copyright audio tracks, and then hit the politicians' videos themselves.