Discussion
gmaz said:
67Dino said:
Yep...Elon Musk: "We'll have fully self driving cars in 5 years"
Internet: "Prove you're not a robot by clicking the boxes with a roadsign"
When you do one of those "click all the squares with a roadsign" human tests, your clicks are literally being used to train the machine learning that is used in self-driving cars vision systems. That's the very reason we have those particular tests and not, say, click all the birds in this picture of a tree, which would be just as effective at checking you're a human, but the data produced from it would be far less useful to Google and others.
Google et al need massive amounts of data to feed into these algorithms to "teach" them what a roadsign/traffic light/etc looks like in all weather and light conditions and all countries. They could never produce all this photo data by hand, so they came up with this clever "crowdsourcing" idea to make us do it for them. The more roadsigns we click, the better the algorithms get at recognising road signs. We're all doing our bit for road safety!
TartanPaint said:
gmaz said:
67Dino said:
Yep...Elon Musk: "We'll have fully self driving cars in 5 years"
Internet: "Prove you're not a robot by clicking the boxes with a roadsign"
When you do one of those "click all the squares with a roadsign" human tests, your clicks are literally being used to train the machine learning that is used in self-driving cars vision systems. That's the very reason we have those particular tests and not, say, click all the birds in this picture of a tree, which would be just as effective at checking you're a human, but the data produced from it would be far less useful to Google and others.
Google et al need massive amounts of data to feed into these algorithms to "teach" them what a roadsign/traffic light/etc looks like in all weather and light conditions and all countries. They could never produce all this photo data by hand, so they came up with this clever "crowdsourcing" idea to make us do it for them. The more roadsigns we click, the better the algorithms get at recognising road signs. We're all doing our bit for road safety!
TartanPaint said:
Apologies if you already knew this, and it's a bit off topic because it's not funny, but it's definitely geeky and somebody might find it interesting.
When you do one of those "click all the squares with a roadsign" human tests, your clicks are literally being used to train the machine learning that is used in self-driving cars vision systems. That's the very reason we have those particular tests and not, say, click all the birds in this picture of a tree, which would be just as effective at checking you're a human, but the data produced from it would be far less useful to Google and others.
Google et al need massive amounts of data to feed into these algorithms to "teach" them what a roadsign/traffic light/etc looks like in all weather and light conditions and all countries. They could never produce all this photo data by hand, so they came up with this clever "crowdsourcing" idea to make us do it for them. The more roadsigns we click, the better the algorithms get at recognising road signs. We're all doing our bit for road safety!
Um... if no-one (human) has previously marked which grid squares actually contain the traffic signals (etc), then how can the captcha check your (i.e., you the human web browser) answers to let you proceed?When you do one of those "click all the squares with a roadsign" human tests, your clicks are literally being used to train the machine learning that is used in self-driving cars vision systems. That's the very reason we have those particular tests and not, say, click all the birds in this picture of a tree, which would be just as effective at checking you're a human, but the data produced from it would be far less useful to Google and others.
Google et al need massive amounts of data to feed into these algorithms to "teach" them what a roadsign/traffic light/etc looks like in all weather and light conditions and all countries. They could never produce all this photo data by hand, so they came up with this clever "crowdsourcing" idea to make us do it for them. The more roadsigns we click, the better the algorithms get at recognising road signs. We're all doing our bit for road safety!
Escapegoat said:
Um... if no-one (human) has previously marked which grid squares actually contain the traffic signals (etc), then how can the captcha check your (i.e., you the human web browser) answers to let you proceed?
If 95 out of 100 people click a particular square to say there is a road sign in it, there's a good probability there is a road sign in it.jammy-git said:
Escapegoat said:
Um... if no-one (human) has previously marked which grid squares actually contain the traffic signals (etc), then how can the captcha check your (i.e., you the human web browser) answers to let you proceed?
If 95 out of 100 people click a particular square to say there is a road sign in it, there's a good probability there is a road sign in it.jammy-git said:
If 95 out of 100 people click a particular square to say there is a road sign in it, there's a good probability there is a road sign in it.
Well yes, but (as described) the captcha-thingy doesn't know the answer before it poses the puzzle. So when the image is first shown, how does it know which set of squares chosen by the user successfully pass the test (of not being a robot)?It's not logically possible to do both things - discriminate human from robot AND gather human pattern recognition data - at the same time.
Think it through from the robot perspective: thousands of robots randomly choosing grid square sets = getting the answer wrong. If the captcha-thingy didn't already know the correct answer:
1 - it couldn't tell humans and robots apart (its purpose for the user)
2 - it would create a dataset that is full of noise, making it way harder to discern human pattern-recognition for the AI (its purpose for the Big Data Bods)
(Apologies for not even dissecting an actual kitten, etc)
Gassing Station | The Lounge | Top of Page | What's New | My Stuff