A problem with reCAPTCHA is if machines doing OCR (getting the answer wrong) far outnumber the humans (getting the answer right). The intended use of reCAPTCHA makes this scenario inevitable.
The "control" word for which the correct answer is known is supposed to lessen this problem, but I am skeptical it works. Say, machine OCR gets control words right only 10% of the time and always gets a certain second word wrong in a consistent manner. The number of wrong answers to the second word (filtered by correct control word) will still dwarf the correct "human" answer.
No comments :
Post a Comment