I’ve never been a huge fan of the google recaptcha as I tried it years ago and was instantly bypassed (I assume it is probably better now).
My biggest problem with it however is that from my limited understanding the ‘select all the squares with …’ is nothing to do with security and is just used for machine learning.
I regularly select the wrong tiles and after a couple of different challenges (all of which selected incorrectly) it just lets me pass.
A quick look on the recaptcha page doesn’t obviously mention machine learning.
Am I wrong on this?
if this is the case shouldnt it be stated more clearly?
No conspiracy theory here… I think they have publicly came out and said that in fact it serves two purposes. One is to act as a deterent for bots by making sure you are human, but the second is in fact they have said they use it in machine learning to help with pattern recognition. I don’t think they are trying to hide the fact.
In fact here it is stated on their website…
reCAPTCHA offers more than just spam protection. Every time our CAPTCHAs are solved, that human effort helps digitize text, annotate images, and build machine learning datasets. This in turn helps preserve books, improve maps, and solve hard AI problems.
reCAPTCHA’s risk-based bot algorithms apply continuous machine learning that factors in every customer and bot interaction to overcome the binary heuristic logic of traditional challenge-based bot detection technologies.
That always gets me. Sometimes I’ll have to click every square containing a bike and there will be some tiny weensy speck of handlebar in a square on its own and I’m never sure if I should click it or not. I remember someone mentioning that the correct answer was what most people had agreed upon, but while searching just now I could find nothing to back this up.
What I did find though was an interesting post from Reddit:
If you’re looking at one of those picture grids where it wants you to do something like picking all the traffic lights, then you have 9 pictures to start with.
There’s at least 1 picture that it definitely knows has a traffic light.
There’s at least 1 picture that it definitely knows doesn’t have a traffic light.
Then there are up to 7 pictures that it isn’t sure whether or not they have traffic lights.
When you make your selection, the system is making sure you selected the positive control, making sure you didn’t select the negative control, and assuming those are correct, it passes your CAPTCHA, and it also adds the data about the unknown pictures that you entered.
I did encounter a captcha recently that had two boxes. On the left it showed an arrow pointing a specific direction. In the second it had an animal figure and told you to rotate the animal figure around to face the same direction as the arrow.
Now the issue is, there are people in the world who have trouble solving spatial problems like this and hopefully they are not going to keep going down this route and alienate people with learning disabilities such as impact to their spatial recognition part of their brains.
I knew a guy who had something similar where he would walk down the street and have to gauge the cracks in the sidewalk to determine if he could safely traverse it because he couldn’t gauge how wide or deep the crack was. Fun fact though, his disability made him a savant in 2-D space and was a genius engineer on paper. I think the government had him drawing up blueprints for stuff.
Secretly, this is Google’s way of determining the collective average intelligence of humans so that its AI can determine the correct time to strike.
By the way, i’ve got some nice shiny hats for sale…
In all seriousness though, of course they’re using the data to machine learn from. Even if for no other reason than to know whether/when/how quickly etc the bots crack through their measures.
I guess its more a case that for the majority of people who have to pass one of them on a website, they have no idea they are actually doing work for google.
I reckon people would be slightly peeved if their favourite website started forcing users to fill in a short marketing survey on behalf of another company before they could log in in the name of security.
This is in essence what is happening with recaptcha though but we aren’t wired to think of it as a survey and we aren’t told by the website what it’s secondary purpose is. Users are only told they need to complete it for security reasons.
I guess that’s mostly why I find it slightly annoying when presented with one.
