With all the high technology development, why can’t bots check boxes that say “I am not a robot”?
In: 649
The creepy answer to this question is “They can (or one day they will be able to).”
Simplifying things a lot. when you click that checkbox you’re usually presented with some kind of challenge to “prove that you’re a human.”
We used to ask you to do something really simple like answer “What is 1 + 3?” and expect you to enter 4 in a box.
Bots “learned” to read those questions and answer them pretty quickly so we made the challenge harder: Here’s some letters and numbers, but they’re funny colors, and there’s lines through them, and maybe we warped the image a bit. Tell me what the letters and numbers say.
People who work on document recognition love problems like that, and so eventually bots learned how to do that too, and we had to make the challenge harder again.
We went through that process of making the challenge harder a few times, and now we the modern challenge is to answer questions like “Which of these images contain tractors?” (or traffic lights, or mountains, or motorcycles, or busses….) – Bots aren’t great at that yet, so most time when you complete one of these challenges you’re identifying some images that a human has classified – we’ll call this person Hugh.
We know those images contain the thing we’re looking for because a man named Hugh said they do, and Hugh is an expert at classifying images. Hugh is right something like 99.999999% of the time, so if you agree with Hugh you’re a human – we let you in.
Now here’s the rub: Sometimes – not always, but maybe once out of every dozen challenges – Hugh didn’t classify all the images. Bob did some.
Bob is a bot.
Now you, Prospective Human Number 368472, will classify the same image Bob did. And if you agree with Bob we let you in.
If you don’t agree with Bob we make you try again, this time on a different challenge that Hugh classified (because we don’t want to make you mad if the bot is dumb and can’t tell a tractor from a taco truck).
We then take those images that both you and Bob classified and we show them to a few thousand other people. If all the Prospective ~~Hugh Mans~~ Humans tell us the image is not a tractor then we tell Bob it got that one wrong. We don’t knoow what it is, but we know what it *isn’t* and it is NOT a tractor.
Similarly if all the Prospective Humans tell us it *IS* a tractor then we tell Bob “This is definitely a tractor. All the humans said so.”
That feedback gets incorporated into the bot, which gets better at spotting tractorsif you got it right, and we use that information to train the bot further, until one day the bot will be able to answer the challenges we’re presenting with accuracy approaching that of a human.
. . . and then we start all over again with something harder.
Latest Answers