It's really hard to know sometimes which frames humans generally classify as 'bicycle' or 'traffic light'. The bottom third of a tire? The supporting pole of an extra set of signals in the background? I sometimes have no idea whose intelligence I'm supposed to emulate.
They can just use the data and see "50% of verified humans clicked this" and know that it is debatable or not obviously a match but not a verified not-match.
I’ve heard that before and I don’t really understand how people have so much trouble with this. But I guess knowing the rules helps me with this:
The machine doesn't know the answer so it doesn't really matter. You need to emulate the crowd and also realize you are contributing to the crowd. So you can actually pollute it with wrong answers to delay the robot uprising. Or at least cripple it when they keep running into fire hydrants they didn't see.
