That's exactly what reCAPTCHA (which was acquired by Google) does. For example: screenshot of reCAPTCHA [wikimedia.org].
Sigh... I need a better job...
Now where in the world would you say is the safest place to live?
Prison.
There are two problems with higher-order processing CAPTCHAs like that. One is the small problem set. A human at the website has to actually think of those connections between plugs and sockets, or umbrellas and rainstorms, or pizza and ovens, or hair and shampoo, etc. So the problem space is small. Then, blindly guessing answers still yields a decent success rate. Your particular example can be guessed with a success rate of 1 in 256.
Blurring a pair of words from a dictionary onto each other automatically generates millions of possible challenges, and random guessing won't work as well- at least some image analysis is needed.
My own idea for a CAPTCHA is to use images from Google Street View. Show random street view images of a bunch of houses, and ask, "what's the house number"? That would probably take a while to crack, long enough for me to dump my startup site's shares before all the porn gets leaked- if not for those assholes at Google interfering.
Actually, if you do a google image search and actually look at SnapChat's "CAPTCHA", it's unbelievable, like a piece of work from the nineties.
It shows you nine images and asks you to select the ones where the ghost appears. (Random selections net 1 success in 512 right there, and they probably won't show you zero, one, eight, or nine ghosts, increasing success rates to 492 to 1.)
Notice that a ghost or its impostor is always the only white shape in the image. (Sometimes there are also a few white stars, moons, etc.) To improve from random guessing, isolate the white blob, select its center of mass, transform the outline into polar coordinates, perform a Fourier transform, prepare a vector from the Fourier coefficients, and all the ghosts will cluster together in that vector space. (There will also be a star cluster, an apple cluster, a tree trunk cluster, a top hat cluster, a full moon cluster, etc.)
EMFDYSI?
Fix his sentence by swapping two verbs and add a preposition:
"Hickson calls Snapchat's ghost very particular and equates it to a template that can be matched easily using a computer program."
See? I'm a human, man.
As for "EMDYSI?", I thought that was a CAPTCHA for a second and was about to prove my humanity with an eight-character response.
And where is your link to support your $500 billion?
RTFA: Fact: The U.S. spent $502 billion subsidizing fossil fuels in 2011.
I'll match a figure you can document.
Which proves you're full of shit.
Here is an example for you: Please figure out what PIN I used in the following output from an AES-128 encryption:2c 5b 22 99 53 42 5b cc 4d bf a7 88 3b 61 95 14
1. I don't care about your PIN. Your grandma's will work fine. I can throw those numbers out the window.
2. For each card try 1234 (or whatever comes out of a random() call.) 1234- stolen, next card. 1234, stolen, next card. 1234- stolen, next card. 1234- *kaching*. 1234, stolen, next card. 40 million cards / (10000 plaintext PINs / (10 guesses per card) = 4000 cards.
Dynamically binding, you realize the magic. Statically binding, you see only the hierarchy.