Bot protection - Human Computing via ability to interpret Complex Image or Question

typical example is Microsoft's Human Interactive Proof named Assira
http://research.microsoft.com/en-us/um/redmond/projects/asirra/

now don't be fooled even these can be 'broken' if the attack vector is aimed on the code
(there is limited amount of choices so even 0.1% ratio for bot-net is fine )
one of attack examples is create database of all the pictures with pre-marked what they are (thus bot has answer beforehand)
http://crypto.stanford.edu/~pgolle/papers/dogcat.html which shows up to 82% chance to break

another example is KittenAuth: http://thepcspy.com/ but it has same problem of pictures being taggeable / limited supply

i would suggest use this method for account 'activation' (when it's implement it should be forced upon everyone once)

another example is 3D captcha / isometric (where the text is picture 3D render / angled )
http://code.google.com/p/3dcaptcha/
but even that might be broken if the source generator is 'known' for reverse engineering ...

another try http://spamfizzle.com/CAPTCHA.aspx via 3D generated pictures and tagging

different angle is usage of Human made question and Human defined answer for that Question
the only partially working system is IBM's Watson http://www.ibm.com/innovation/us/watson/index.html

how would that work ?
as author of giveaway i fill up two fields

1. Question
2. Answer

Example, simple:

1. What's name of Earth moon?
2. Moon

Example, more complex

1. Who was popular actor of captain Jean-Luc Picard in StarTrek: TNG?
2. Patrick Stewart

of course You as author of giveaway may decide how complex this question will be
(if it needs brain or just search to solve)

please realize that even the Question and Answer can be rigged (if the answer is known it can be used on N bots)
combination of at least 2 methods would be needed to show some 'results' (isometric / picture recognize + question/answer)

p.s. i wrote this as reaction on the endless amount of useless ideas with captcha/re-captcha (hint, OCR vs re-try)
http://www.darkreading.com/authentication/167901072/security/vulnerabilities/226700514/index.html
read http://www.allspammedup.com/2011/01/google-recaptcha-cracked/