The ReCaptcha guys need to invent an algorithm that gives an estimate of the likelihood that a human could actually decipher a word, and then only present those above a certain threshold. That’s a different problem to the one of having a machine actually decipher the text, and I reckon it’s probably an easier one. The ReCaptcha above is just plain silly.
Hi,
I’m the chief engineer on reCAPTCHA. You’re right that this type of issue occurs sometimes. In this case, the blob on the left is part of a book that OCR (understandably!) couldn’t read. If the user just enters “female” they will pass the CAPTCHA (or they can just press refresh to get a new one).
We actually do apply a number of filters to prevent “bad” stuff from being displayed, though these are harder to get right than you might think — think about the cost of false positives.