Ive heard many times that google gathers data for their self driving cars through captchas.
But how does it train the AI if the AI already has the solution to the captcha im solving.
Its like a school test. The student can give right or wrong answers. No matter what answer the students gives, the teacher doesnt learn anything because he already has the answers
In the same way how does the „AI“ know my answer for the captcha is wrong
And how does my answer train anything
In: 8
From my understanding, AI does not always have the answer.
You’re not the only one it’s testing against. Sometimes it actually is a “is this person human?” question opposed to just used for learning.
They give the same captcha to multiple people and/or give one person a captcha where they know the answer (because other people gave it the answer previously) and one unknown captcha where they take your+others’ answers and, if they’re the same, treat that answer as right.
Those answers are used to train the AI; the AI’s knowledge is not used in determining whether you got the captcha right.
It doesn’t have the solution, or rather it might have a part of the solution if it gives you multiple images. This was most apparent in the early captchas that gave different qualities of texts.
It gives the image to millions of captcha users to sift out bad advice and help select for obstacles, stop signs, street lights, etc.
One can surmise that there is a dataset of known stoplights or house numbers or whatever they seem to be looking for, along with an unknown you are given. They mix in the new one with the old ones to get the trend of the new one, and of course if you don’t know which is the one you are teaching, and also don’t pick all correctly, you don’t get to the destination site.
Sometimes you are forced to do two in a row. One can be testing your honesty and bottyness, and the other can be learning.
Also, Google knows who you are, can track your mouse movements, etc.
What the AI is learning about in some cases (that obviously had to start with some human selection), [we can only guess.](https://www.reddit.com/r/funny/comments/yvfr0e/this_ai_machine_learning_by_captcha_is_getting/)
for what I know they’re mainly used to process a lot of information about images using same object in different context, color, light and shape, then they give those images and informations to an AI that, with a lot of data, can learn to recognise already analysed object.
This is useful to self driving cars bc they need to recognise rapidly static and moving object on the road and then act based on what they recognise.
Plus I read that actually Captcha arn’t used to detect bot (there are other and more useful method) it should be just a way to make a lot of people work without needing to pay them (like wage-free microworks)
Latest Answers