Speaking of Tarski's theorem: http://johncarlosbaez.wordpress.com/2013/03/31/probability-theory-and-the-undefinability-of-truth/ … http://intelligence.org/wp-content/uploads/2013/03/Christiano-et-al-Naturalistic-reflection-early-draft.pdf … Probabilistic logic as a way around Tarski? Also, note coauthor.
@GabrielDuquette lots of machine learning papers—author thought algo was learning to find faces but actually it just found pink pixels, e.g.