> > > I like the ones with crooked letters that the robot can't read. They are widely used, so it's just a matter of time before they figure out a way around them.
> > > Howard
> >
> > Actually, I dislike CAPTCHAs, personally, because it always takes three attempts to prove I'm human. But if the effort of solving them also served a useful purpose, then it would be a good thing on balance.
>
> Ok, I have a question about this scheme here. Say you try to register on some site somewhere. You're presented with a captcha image that is actually two words of a book that has been scanned, and OCR can't catch these particular words, so by providing the ASCII translation as part of your registration process, you're helping to make this book available online in ASCII form.
>
> How, then, does the captcha image also work as validation? It can only work as validation if the registration engine on the server already knows the answer. If it knows the answer, it doesn't need you to tell it what the answer is. If it doesn't, then, yes, you're helping out with the translation, but you're not validated as a human being -- which means that hacking scripts could flood the registration engine with invalid ASCII translations of the captcha image.
>
> So I don't understand how this works.

I think I understand. It's explained in the blog post linked here. The reCAPTCHA system gathers replies to the same CAPTCHA from multiple users, and if they all agree on a single interpretation, that interpretation is accepted as valid. If the users are humans, they are very nearly certain to come to one single interpretation, but if they are computers, there can be minor differences.
This still means that when presented an entirely new CAPTCHA, a human won't get accepted as one even if he/she answers the CAPTCHA correctly, because the reCAPTCHA system doesn't yet know he/she did so.