People generally don't take pictures of rolling green hillsides. But they very often take pictures of rolling green hillsides with sheep on them. So if you ask the robot to draw a picture of rolling green hillsides, it will include sheep. Or, if you ask it to draw a picture of the savanna, it will want to include giraffes.
Now you're being asked to find a bus in a photo without a bus because it's a street scene, and every street scene has a bus in it.
I haven't read her book, yet, but her Twitter[2] is often full of amusing anecdotes like this.
Don't humans do this too, though? If I asked someone to draw a picture of rolling green hills, they may well add sheep as an additional detail.
I haven't done the experiment, but I'd posit that if you walked up to a group of 8-year-old children, gave them crayons, and asked them to draw pictures of "rolling hills", a significant portion would add sheep, cows, flowers, or some other details—even though a majority of rolling hills in the world don't have any of these features.
Your goal shouldn't be to answer the question earnestly, but to confirm the machine's biases. Going with the flow is expedient, as well as giving Google less support.
It gets more frustrating when it's less clear. Is "click the hills" with a picture of a mountain a mistake, or a trick? Should I click all the tiles that contain a bus, even if it's only one pixel, or should I only click the ones that mostly contain a bus? etc.
Hopefully these rules of thumb will help someone reading this find these captchas less frustrating.
Also when it asks to select all the cars. Is a bus a car? Is a truck a car? I really don't know what it is expecting and I must pick wrong as I often fail.