A computer programmed to deal with sarcasm would contrast this statement with (local) popular opinion and deduce from the large contrast between the certitude of the statement and the popular opinion that this statement is either sarcasm or obliviousness. I think exactly as humans do. It's not that sarcasm is hard to detect for computers, it's just that it's hard for computers to collect enough contextual information to judge the validity of any statement.