Captcha Is Asking Users To Identify Objects That Don't Exist (vice.com) 68
An anonymous reader quotes a report from Motherboard: People trying to use Discord are being asked to identify an object that does not exist. The object in question is a "Yoko," which appears to be a kind of mix between a snail and a yoyo. Multiple people have reported seeing a prompt to identify a Yoko when asked to solve a simple captcha prompt while trying to use Discord. The picture of the Yoko, as well as the other images in the captcha, appear generated by AI. Another user complained on Twitter that they'd failed to pass a captcha to log into Discord when it asked him to identify images of a puzzle cube. Again, the pictures appeared to be AI generated.
Discord's captchas are run by a company called hCaptcha. "The technology that generates these prompts is proprietary to our third-party partner and Discord does not directly determine what is presented to users," Discord told Motherboard. "While most hCaptcha interactions do not result in a visual challenge, many variants are used at any given time," a spokesperson for hCaptcha told Motherboard. "This particular question was a brief test seen by a small number of people, but the sheer scale of hCaptcha (hundreds of millions of users) means that when even a few folks are surprised by a challenge this often produces some tweets."
The issue with hCaptcha's strange AI generated prompts highlights two issues with machine learning systems. The first is that the AI systems require an enormous amount of human input to not be terrible. Typically image labeling is outsourced to foreign workers who do it for pennies on the dollar. The other is the issue of data drift. The longer these machine learning systems run, the more input they require. Inevitably, they begin to use data they've generated to train themselves. Systems that train on themselves long enough become AI Hapsburgs, churning out requests to identify incomprehensible objects like "Yokos."
Discord's captchas are run by a company called hCaptcha. "The technology that generates these prompts is proprietary to our third-party partner and Discord does not directly determine what is presented to users," Discord told Motherboard. "While most hCaptcha interactions do not result in a visual challenge, many variants are used at any given time," a spokesperson for hCaptcha told Motherboard. "This particular question was a brief test seen by a small number of people, but the sheer scale of hCaptcha (hundreds of millions of users) means that when even a few folks are surprised by a challenge this often produces some tweets."
The issue with hCaptcha's strange AI generated prompts highlights two issues with machine learning systems. The first is that the AI systems require an enormous amount of human input to not be terrible. Typically image labeling is outsourced to foreign workers who do it for pennies on the dollar. The other is the issue of data drift. The longer these machine learning systems run, the more input they require. Inevitably, they begin to use data they've generated to train themselves. Systems that train on themselves long enough become AI Hapsburgs, churning out requests to identify incomprehensible objects like "Yokos."
We know where AI-generated captchas will end up... (Score:5, Funny)
Re:We know where AI-generated captchas will end up (Score:4, Funny)
https://xkcd.com/2415/ [xkcd.com]
Eventually it will be for screening out humans (Score:5, Funny)
The captcha will be so difficult that only a bot will be able solve it. And deliberately so, the AI will probably be the most valued consumer.
Re: (Score:2)
Since it's only a matter of time until bots get to make purchase decisions, and only a matter of a little more time until such important decisions cannot be left to the fickle minds of humans, it's only logical.
Re: (Score:2)
People who use Discord deserve to lose
Captchas must die a fiery death (Score:5, Interesting)
Firstly, they're an insult to a man's intelligence. Seriously, identifying fire hydrants and traffic lights to visit websites... What is this? A training center for the mentally feeble?
Secondly, always remember that each time you "solve" a captcha, you help a big data company fine-tune an image recognition algorithm that will generate millions for the company for free. Or said another way, not only do you have to suffer the ignominy of those stupid micro-tasks over and over, but you actually work for free for Google each and every time.
Fuck that! where's my salary? I don't work for free! Even Amazon's Mechanical Turk pays whoever does the tasks.
The only redeeming grace of hCaptcha is that it's not Google. But that doesn't detract from the fact that they've chosen the most hateful business model there can be on the internet, 3 rungs below data broker and advertiser.
Re: (Score:2)
Indeed, just show the frigging content the user is asking for. No cloudflare nor stupid captcha in any site I host, there are many other ways to filter out bad request at the reverse-proxy WAF level.
I guess they use captcha only to train their stupid AI, just like you said...
Re:Captchas must die a fiery death (Score:4, Insightful)
You're not doing it for free - you're doing it for access to a website. If that value is insufficient, walk away.
And how is the fact that AI's are still too stupid to solve such captchas an insult to your intelligence?
I *want* captchas to be as easy to solve as possible while still doing their job - their existence is annoying enough, I don't need any extra grief. And I'd rather keep the value I provide to the data company behind it to a minimum.
Re: Captchas must die a fiery death (Score:3)
I'm going to have your city install spikes on the roads in your city and you have to prove that you live there for the spikes to lower so your tires aren't punctured. If you don't like it then move to another city.
I'm getting real tired of 'moral' goody two shoes who say we should just put up with abuse because "we are getting a free product". You folks are like the little girl in kindergarten who screams and points out the other kids kids in class to the teacher for every tiny infraction.
Re: (Score:2)
And we get tired of the extreme analogies that prove nothing other than how extreme you see everything
Re: Captchas must die a fiery death (Score:2)
EXTREEEME! ::gruff low voice:: YEAHH! (heavy metal guitar riff)
Re: (Score:2)
Re: Captchas must die a fiery death (Score:2)
I wonder where the AI got a "K" from the word "snail"?
Re: (Score:2)
Maybe we should start AI out watching Sesame Street.
Re: Captchas must die a fiery death (Score:2)
What AI should be processing first
"Hooked on Phonics"
"Webster's Dictionary and Thesaurus"
What AI should stay away from
"The Cat in the Hat (or other Suess books that have words like floobuls and flubbles)"
"Alice in Wonderland/Through The Looking Glass"
If we are going to start treating AI as a replacement for people, it will need a well mentally balanced diet.
Re: (Score:2)
The squares are the worst ones. Select all the squares with a road sign in them. Does that include the pole it's on? Or the squares were the sign extends 3 pixels into them? Click on the motorbikes, does that include the rider?
Re: (Score:2)
Firstly, they're an insult to a man's intelligence. Seriously, identifying fire hydrants and traffic lights to visit websites... What is this? A training center for the mentally feeble?
If it's Facebook, then yes.
Re: Captchas must die a fiery death (Score:2)
Re: (Score:2)
What is this? A training center for the mentally feeble?
Forum spam is real and we do need to make something that is smarter than a spammer's scripts but dumber than 95% of computer users.
Re: (Score:2)
Just do what /. does, offer a free pentest along with the results posted as comments to everyone spamming their webpage.
Re: Captchas must die a fiery death (Score:2)
We're reaching the point where its more like 50% of users who can still solve the puzzles. "Human detected! You have failed the Reverse Turing Test and are therefore considered human. Humans are no longer allowed on this website. Have a nice day, meatsack."
Re: Captchas must die a fiery death (Score:1)
Re: (Score:2)
The alternative to captchas would be strongly authenticating all platform users, such as having you to produce ID during account creation and forced 2FA on each login. The do-nothing alternative is getting downed in spam and bots.
Not hard at all (Score:3)
The object in question is a "Yoko,"
This is really easy: If the picture includes a famous rock star who looks like Jesus hanging out in a bed, simply select the squares that contain the long-haired Asian woman sitting next to him.
Re: (Score:2)
They're asking us to identify a Yoko? Oh, no!
Re: (Score:2)
That's easy. It's number 9.
product of a deranged imagination (Score:2)
Also GIGO. I dread to see the unanticipated side-effects of people intentionally lying to the AI. "You will witness man-made horrors beyond your comprehension.", indeed.
Re: (Score:2)
The object in question is a "Yoko,"
Maybe the capcha squares should have sound tracks. Then you could pick the one that sounds like a cat being tortured.
If you don't like captchas, pollute the well (Score:5, Interesting)
Most people never dare to try solving captchas like shit because they want to get past them as fast as possible. But if you want to get back at companies that run those awful captchas a little bit each time - mostly Google of course, who else... - you can.
Here's what you do:
- When the captcha asks you to identify "all XYZ" in an image, only click one square then Verify. Half of the time, the damn thing passes and you haven't actually tagged the entire image properly. If it doesn't pass, select another square - one that is close by but not even correct - and Verify again: that usually passes also. If it doesn't, rinse and repeat, identifying what you're asked to identify as badly as the captcha will allow.
- When the captcha asks to pick squares containing something, pick only one square and Verify. The captcha will tell you to pick ALL the square. But only pick one more and Verify. After 1 or 2 more, usually it's ripe so pick a wrong square and it usually passes anyway.
Why does this work? Because people fail to realize that the system isn't actually checking that you know what a fire hydrant or a motorcycle is to determine that you're human, it's actually asking you to tag images to train a fucking AI for free under the guise of a humanity test. So if you provide a not-too-wrong but still wrong answer, it lets you in and you pollute the sumbitches' training data.
Everybody should do this and solve captchas as poorly as possible as a way of protesting them, because if enough people did this, the captchas would become useless for the purpose of training AI, would cease to be of any value and would stop being used.
Re: (Score:2)
if enough people did this, the captchas would become useless for the purpose of training AI, would cease to be of any value and would stop being used.
I suspect much of this is being used for training self-driving cars. I mean... what other use is being able to recognize traffic lights, fire hydrants, cars, busses, taxis and the occasional bridge or mountain? Sure, stairs and trees pop up occasionally that are off-theme, but most of the time... it seems very, very FSD-focused.
So... do you really want us to make the model worse?
Re: (Score:3)
So... do you really want us to make the model worse?
No. I want to get paid for the work I do. Or I want someone to get paid to do that work. When companies virtually enslave me - because I have to solve a fucking captcha to get to the website I want to see - I will sabotage the work as much as possible. That's the only pushback I have left, albeit a very weak and probably not very effective one.
You're suggesting that I should perform forced labor properly because the result of my labor is positive to society. That is an incorrect premise. The correct premise
Re: (Score:2)
It is not forced labour since your payment is allowing you to visit the website. Will you compare it to holocaust next?
Re: (Score:2)
So where do I put in my credit card number to buy a year's subscription to the website?
Re: (Score:2)
The vendor chooses the accepted payment methods.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Yes, a very frustrating one that.
They're also inconsistent as to whether they accept the back of traffic lights/pedestrian crossing signals or not.
Re: (Score:2)
If you want to send them a message everyone has to simply close the web page whenever one pops up. Good luck with that. But if you think wasting your own time is doing something, have at it. You are only wasting your own time.
pffft that's nothing (Score:2)
Try finding all the Onos instead,
On second thought they're noisy, unintelligible, and tend to break things up. Better just ignore them.
Re: (Score:3)
This one is so easy. You choose the picture that hurts your ears the most.
Y'all are just trying to fuck with me... (Score:1)
Wow (Score:2)
First you have to create an account, send them your phone number, and give your whole fucking life info over to them and they still treat you like a criminal.
And then they baffle you with bullshit by making you click on all of the woozle wuzzles just to do basic tasks.
They should just shut down the servers and get into the basket weaving business. The fear of the big bad penis spam boogyman is too much for them.
Is the AI exhibiting imagination? (Score:1)
Is this actually a feature not a bug?
Lovely turn of phrase (Score:1)
Identify a Yoko? (Score:2)
Is it anything like a scrupt? (Score:2)
Is it anything like a scrupt? Because if it's like a scrupt then I'd want to know.
We would ALL want to know if it's anything like a scrupt.
(My apologies to Stanislaw Lem)
Re: Is it anything like a scrupt? (Score:2)
(From the intro to one of his books)
Re: (Score:2)
Better a scrupt than twonky
What a BULLSHIT answer! (Score:4, Insightful)
If you outsource something, you are still responsible for the performance of that outsourced component.
We have to get away from the management perspective of "outsourcing means never having to say 'it's my responsibility'!"
Re: (Score:2)
Responsibility is always towards something or someone. So in the context of the agreement between Discord and the captcha provider, the captcha provider is responsible for their service. In the context of the agreement between discord and their paying customers, discord is responsible for making sure the system works (including captcha) as stipulated in the agreement.
You and me aren't any such parties so discord or the captcha provider isn't responsible to us at all. Given that the error clearly lies within
Bicycles Anyone? (Score:2)
Weâ(TM)re training the AI (Score:1)
Re: Weâ(TM)re training the AI (Score:1)
The irony (Score:2)
This is happening at the same time artists are claiming that AIs can't create and are clearly copying.
Here's help (Score:2)
A yoko looks like this:
https://cdn.theatlantic.com/th... [theatlantic.com]
I got hit with the yoko. (Score:2)
Getting tricky! (Score:2)
Yesterday afternoon, Twitter decided to lock my old account during a DM and gave me a tricky captcha. I had to select dice images with their total numbers to 14. It was quite annoying!
hCaptcha is violating wage laws (Score:2)
They are selling your labor to theird parties without paying you at least minimum wage. They are also exploiting child labor, as they do not care who they burden with their criminal conduct.
Yoko! (Score:2)
Ono!
For either sense of "Yoko", I guess the more appropriate form would be "Oh no!"
In either case, poor Marco would be embarrassed by the association.
Good ole captchka (Score:2)