Standard captcha are either broken or are an accessibility nuisance or both. In its most noble form - reCaptcha - it is used brilliantly to digitize books and provide people over at India with a living:
http://recaptcha.net
http://blogs.zdnet.com/security/?p=1835
http://www.w3.org/TR/turingtest/
Smart engineers all over the world are busy trying to invent the most inaccessible new generation captcha possible, which typically involves analyzing or playing around with images.
However, it came to my mind recently that there might be another way, accessible and more faithful to the original Turing test (en.wikipedia.org/wiki/Turing_test). Can a computer administer an effective textual reverse Turing test to block registration spam? Can Eliza, the friendly nonsensical therapist, the famous first Turing test runner up, do that?
Imagine you try to register a new account at web20.com and suddenly Eliza pops up and asks you to elaborate on why you chose to submit the spammy looking email address john23kjdsf784@mail.ru
Seriously, spam filters such as Akismet (for blogs) or Gmail have become so efficient that they have practically eliminated comment and email spam. By requiring a registrant to submit enough textual information it may be possible to apply these filters to registration spam.
In addition, such a conversation, while being entertaining enough for the casual registrant, can be too time consuming for professional-human-captcha-solvers, relieving the potential problem of relay attacks.
Finally, if a new developer vs spammer war ensues around chat captcha we will soon see the first computer to pass the Turing test