This is a roundabout way of just saying the most obvious explanation (sites with US-centric captchas are presumably mostly interested in US users anyway) but wrapped up in an unnecessary conspiracy layer to add creepiness.
That is one way of running an captcha: set a task that is not possible for a human. Anyone who answers correctly must be a bot.
"How many flowers in this field?" "Solve for X: 3567/134 = X" Anyone getting those correct in a reasonable time is probably not a human and so has failed the captcha.
It would be different if the captcha would look at the incorrectness. I.e. the bot would now have to make the same mistakes that humans would make, which might be a workable path for captchas, however its unclear what benefit this would bring compared to answering correctly.
Would take me, a human, about 3 seconds to
1) highlight 3567/134
2) Press Alt-Escape
3) type "bc<enter>"
4) middleclick
5) Doubleclick answer
6) Middle click in answer box
26.61940298507462686567
The point of the GP was, it's training these AIs on specifically _American_ hydrants, store fronts, crosswalks and traffic lights, which sucks not only for those of us unpaid mechanical Turks who aren't Yanks, but also makes you wonder if nobody cares that these coming AI-controlled vehicles won't be so great at discriminating NON-American hydrants, store fronts, crosswalks and traffic lights from _non-American pedestrians._
And no, America does not equal “the West”; most of your street furniture looks slightly“ off” compared to, say, that of most European countries. Anyway, the part of the “West” that will probably be at the forefront of this wave will probably, as usual, be the most hyper-advanced: Japan, Hong Kong, Singapore, Taiwan...
That isn't what captchas are used for. There are two side to every captcha, what the website is using them for and what the captcha creator or provider is using them for. In google's case, captchas are being used to hone AI. So it is no wonder that items like hydrants, store fronts, crosswalks and traffic lights dominate. Those are the objects that any AI-controlled vehicle will have to recognize. Given that the first market for AI-driven vehicles seems to be western nations, specifically the US, it is no surprise that google gears its captcha program to objects from that area.
If google were developing robots to pick fruit captchas would ask us to identify bananas and oranges. I worry about the day that captchas start asking me to differentiate people from other objects. I don't want to help train an AI to tell people from trees, soldiers from bushes.