Hm, it seems to me that all this technology being developed to break CAPTCHAs could be put to very good use...like converting image eBooks to text.
Carnegie Mellon did something similar in a project. It presented two CAPTCHA words: one was a CAPTCHA, and the other was a word out of an old physical book, and if the first one registered as correct, the second one was (assumed to be) correct as well, and if the second word was "correct" enough times, it was removed and cycled with another word.