A major problem with captchas is that OCR is good enough to fool it some of the time. Will machine learning do the same to Mechanical Turk?