You've been training AI for free
Вставка
- Опубліковано 5 вер 2024
- Modern AI needs tons of data for training and quality control -- and a lot of it is coming from human beings, whether it's contract workers recruited from Amazon's Mechanical Turk or regular users filling out a CAPTCHA.
Learn more: bit.ly/2X8e9AW
Photo of the reconstruction of the Turk, Carafe via CC SA: bit.ly/39k0JGN
Subscribe: goo.gl/G5RXGs
Like The Verge on Facebook: goo.gl/2P1aGc
Follow on Twitter: goo.gl/XTWX61
Follow on Instagram: goo.gl/7ZeLvX
Why'd You Push That Button Podcast: pod.link/12952...
The Vergecast Podcast: pod.link/43033...
More about our podcasts: www.theverge.c...
Read More: www.theverge.com
Community guidelines: bit.ly/2D0hlAv
Wallpapers from The Verge: bit.ly/2xQXYJr
Subscribe to Verge Science on UA-cam, a new home base for our explorations into the future of science: bit.ly/2FqJZMl
What other topics would you like to see?
Surprise me 🧐
A video about the monopolies of giant tech companies.
Hey
Hello
More AI topics. Thanks!
All I’m hearing is that Google owes me a boat load of cents.
A literal boat load
Yeah, here's it's what they say on the website it's actually hidden in plain sight. 'Millions of CAPTCHAs are solved by people every day. reCAPTCHA makes positive use of this human effort by channeling the time spent solving CAPTCHAs into digitizing text, annotating images, and building machine learning datasets. This in turn helps preserve books, improve maps, and solve hard AI problems.'
No, they don't.
You solve captcha, you get content. It's a simple exchange of information!
@@mr.norris3840 Wrong
@@mr.norris3840 wrong.. that works for the older version of captcha.. the latest ones which you need to select pictures still owes me boats of cents..
gonna put that on my resume: Trained and developed AI for Google
Jiayi Li 😂
lmao!
definitely not a lie but the truth
Me to! Im not joking !!
Should be top comment
Plot twist: google started the mannequin challenge to get the data
yes I think you're right.
Angry Ox what’s that?
@@speedzzfreak challange where you upload a pic from yourself now and one from 5 years back on fb
@@user-xo1jz7pp7e Yes they did indeed. They promoted that challenge and boosted views and got everyone to do it.
Why do you still remember that?
Now the 10 years challenge makes sense
That's exactly what I thought, too. Glad I did not participate.
Lolol yikes that's a bit uncomfortable
Except they probably already had your photo from 10 years ago and could've compared it without the need of a challenge.
It's weird how they turn these ai training courses into "fun memes" was planking a way for AI to tell if people were sleeping or laying down
Denzel Dickenson What I took from the video is that we still do these things out of our own volition but that these challenges are then used by those who want to improve AI. Even the Captcha stuff wasn’t used for AI until more recently.
"We have a resident expert here"...Expert: "So I did googling here"..
He was searching expert. :p
He did Googling on what's in the photo, not regarding machine learning itself.
😂😂😂😂😂
The bots are already among us
Everyone uses search engines, as they are virtual libraries containing information on everything ever documents and uploaded.
He is a 40 year old anti vaxx mom
Robot: “Are you a robot?”
Human: “ugh no stop asking”
Robot: “ click pictures proving you are not a robot”
Human: “ Boom, boom, boom, there you happy?”
Robot: “...thank you, you may proceed”
*Human leaves page
Robot: “AI world domination plan currently 57%...”
funny ヽ(*⌒∇⌒*)ノ
Where do I send the invoice for doing all that CAPTCHA?
Lmao
Where does Google send you an invoice for using UA-cam for free?
One of the best ones from The verge. Nice work team
I don't know why, but I really love the subtle pan-up to show the title
That's quite disgruntling for me. Interesting. The cut-away after the camera drop really ought to lead to some content other than the host, imo.
I still don’t like that they hide what the host is doing under the frame. Granted, those are just simple readjustments of things on the table, but I feel like the Verge is being dishonest for a moment simply because of that.
Edit 1: a feeling is not always a reflection of facts, however. I still trust the Verge’s journalism.
Edit 2: Maybe I should not trust the Verge. Some people say they are wrong.
The dramatic pause for added effect. I'd like to see the pan up, dramatic pause, pan down and continue, instead of the cut to nearly the same shot
I'm such a bad AI trainer. Usually it takes me 3+ attempts to get past the skewed and the image squares captchas.
null pointer times milion of people
Well somebody has to stop these robots
@@tshapedl law is too slow to react for those giants monopoly for info and resources...
I don't trust this guy one bit! I've seen him on the movie 'Split' before
Rejoice! The broken are the more evolved! Rejoice!
Was the '10 year challenge ' built for the same purpose, i.e. to improve face recognition software ?
Yes
That laptop's is only used to get closed and I'm here for it.
It was a dead water damaged one he got on ebay for 50 bucks!
Waiting for the guy to grab a mic and sing "Numb"
Lol I wrote a song called numb on my sound cloud. OG Vibe. so I totally thought you were talking about me lmao
@@theblackhole05 nice promotion
Waiting for the day when yt comments have worth rather than being a pointless and shameless way to get internet likes...
I bikam somtaim
He looks so alike
I did some googling 🌚
I guess I’m an AI specialist as well.
I watched chernobyl. I guess I'm a nuclear physisist as well
He did Googling on what's in the photo, not regarding machine learning itself.
I purposely get those picture matching grids wrong. Take that google!
God damn! Chester came back to work with verge!
Anindya Guha he’s waiting for GTA 6
Amazon's MTURK probably just got thousands of account requests for an account.
We've been training AI without our consent.
Incorrect. Don't you remember reading google's license terms?
@@jakubsebek I don't remember posting this comment, let alone watching the video.
@ Yeah I wouldn't be proud of it too
Wait wait wait.
Did google manufacture the mannequin challenge to gather depth perception data?
Great video. Really enjoying these
Wow I found your comment karl one year ago lol you love AI 😁 just was scrolling and boom no likes ah on your comment!!
So AI won't murder me because I'm basically it's daddy
Daft Funk Well no, were all its daddy and mommy
This plays a huge part in how Tesla is building the self-driving capabilities in it's cars. Every time people drive in autopilot the car is collecting more data.
"Excited A.I" = Skynet
I need this black-mirror-ified
Just to add clarity the alpha go program wasn't based off of any human data. The algorithm was only given the rules of the game and it played against itself to learn those interesting rules.
Year 2040, human ask AI how something doing,
A.I: So I'm did some googleing...
Human:
hehe It's a brazilian price checker 0:49
The reverse is also true.
AI is training on us, instead of us training it actively. It mimics, till it figures out better patterns.
1:45 that blocked webcam tho
I like the little slide on your laptop to cover your webcam.
Oh god if AI's understanding of the world comes from UA-cam videos, we're totally screwed...
Lol
Really good insights on the datasets collection for AI
I'm mostly scared of AI because they are learned by human. Meaning they will think we are the robots and them the human.
Take Cleverbot as an example.
Human: Who are you?
Cleverbot: I'm claire, u?
Human:John
Cleverbot: No you're not, you're a robot
Human:YOU'RE THE ROBOT
Cleverbot: No u
That first "parking meter" is actually a Brazilian price checker! Maybe they're making an app that checks prices or something?
You guys love the concept of showing someone talking about a subject over a video call :)
0:46 those are actually market bar code scanners, showing some prices for Brazilian products. Thought you ought know :D
It wouldn't hurt to translate "busca preço" before making the video 😒
Lol I thought UA-cam was glitching and was sending me to the wrong video until I read the description.
Okay so humans only need to make working recaptcha shirts and wear them if robots take over. Lmao
after looking at this i believe that the portrait mode of pixel phones that we love so much is just the by product of mannequin challenge
They split the camera in half, using each half of the sensor like a different camera to help do it
Great video, make more videos about AI
El North nice try Deepmind
4:27 Euskal preso eta iheslariak etxera!
Thanks The Verge :D
Jox Oiz whatever you said stfu
How does one “Troll”?
Asking for a friend...
I’m positively not a robot
But the Images on Captchas must have already been classified? Otherwise it wouldnt matter what you click on?
Rennschnekk nope they just compare the mass choice with your choice and then checks for a certain percentage of correctness
it just wrong since most people click wrong picture..... i am not using that data to teach computer since that would take forever to fix that...
Great just at always, but you could cut down on the background music a little
Well that was a dramatic change of title
What was it before?
@@NourAslaoui They've changed it again haha
@@NourAslaoui Initially it was "How this 300-year-old robot shapes AI today". Then "What Google's CAPTCHA is really for" (or something along those lines).
@@ASLUHLUHCE thank you!
I've noticed the host has a subtle notch hairstyle...
Captchas cannot be used to train AI, as it has to already be a labelled set. To check if you are human and actually able to read the correct number of a picture, it has to know the correct number, to state one example.
@ 3:22 - But that doesn't make sense in this case. The program has to know the answer otherwise your recaptcha wouldn't work in the first place.
The captcha works because it already has an answer, right? But think of thousands of people making the same mistake on a single captcha. This is where the machine thinks something is wrong. Then they reanalyze what answer is people choosing and why it's throwing an error. With captcha, we are not directly giving them the answer. What we are doing is giving the Machine confidence that, what machine has predicted is right. (sorry for my bad English)
To understand better I recommend you to watch CGP Grey video titled How Machines Learn. It'll give you an idea as to how AI were made!
I think they use a mix of verified and unverified images when they do recaptcha, because there have been times where I’ve missed images and it’s still worked fine for me, or clicked on the wrong ones. This only happens about a third of the time from what I can tell
That guy has a notch
.
.
.
.
Mr. Apple😂👍
Lol
Hey! I'm brazilian and I have to tell you that is not a parking meter. Actually, "busca preço" in portuguese can be translated to "price search". That is a machine found in supermarkets that scans barcodes and show the prices of products.
I feel like because I am in my field I take it for granted that people know about this. But more over than educating people on the fact that this is happening, I like this kind of content to remind people like me that not every body know about it happening.
*"A small price to pay for salvation"*
Amazing amazing video! We absolutely need content this every hour of every day!
catchpa: prove to us you’re not a robot
*uses data to help make robots*
You've been changing the title, Verge. First it was about Google's captcha, now it's about AI
I recently took Ocr pics in our local language for google machine learning 😅 got $20 earning for 100 pics
Where did you do this please? Mturk?
The production and quality of verge’s videos are outta this world
Thank you! Just updated my CV on *AI Tester for Google*.
Doing it since 2015...way to go!
Busca preço means search for the price in Portuguese. It’s a gadget for supermarkets in Brasil & Portugal
I appreciate that you didn't use skype or at least didn't show it in the video. The video clip of the AI expert is in fact of a high resolution. Thumbs up for that!
Great video! Gets clipped in the end there. Would love to see more of these.
Actually, Facebook has only one picture of me as a profile picture, and that picture is 10 years old. When I log in, usually once a month or so, they beg me to add info to my profile, but I never do. My profile is nice, clean, and minimalistic.
On the other hand, google photos has every vacation photo I have ever taken since iPhone 4s as well as tons of document scans from school.
Welcome to the Matrix, whole food man
Yea I’ve kinda suspected this for a while, sometimes captchas feel a little sketchy, I’ll click a square that isn’t quite correct and it’ll pass just fine.
umm... yeah CGPGrey said that... ages ago
But how does captcha know when you don't click right ?
This kind of videos is why I subscribed to The Verge! great video, good Job!
SanraS Uzumaki They stole another episode from Planet Money: episode 908 aired on 4/24/19. I would have been awesome if they hadn’t stolen the idea.
6:45 So pretty lean description of this 300-year-old machine.
Ivo Poblete I guess that’s why they changed the title
@@GodOfDonuts Yeah, right?! 😯
They just did.
This is machine learning. Not AI !!
There's no ai without machine learning. Otherwise you're referring to something that isn't ai.
Yeah I know
If you click wrong picture in captcha it knows that you picked wrong.
For free in exchange for information (content on website) and service.
Doesn't sound free to me...
around a year ago i saw a post like this on 4chan talking about training AI's while doing capcha thought it was a troll to get us to get them wrong on purpose.
mechanical turk is probably the best name that someone could come up with for this job
In the End, humans are now capable of creating other humans
Jes Vinay we already done that since first human existed. By having sex.
Making weird faces again ruthlessly
😂😂
we don't have self driving cars because i deliberately select the wrong tiles on captcha.
You guys changed the title and the thumbnail! 😅
What do you mean?
I demand a raise
1:04
Man, Chester Bennington is alive!
That was so eye opening and informative. Thanks verge
I guess the same happens when Google asks you those little vocabulary quizzes like synonyms or sentences
I was just talking to a few friends about this recently. Would you guys say that for the convenience we get from technology (specifically Google and Facebook) it’s worth it for the compromise of our privacy??
In the early 2000s there were sites on which you could check your paranormal abilities. You were shown a sequence of pictures and (it was alleged that) at the same time there was someone else in the world looking at the same pictures. The task was to enter as many appropriately descriptive keywords for each picture that you'd imagine the other would enter too. The more keywords you and the other "player" had in common, the higher your paranormality scores. To make it addictive, there was a leaderboard and you could invite others to test their paranormal abilities with you. It also had pseudo-scientific articles (aka nonsense) about extra-sensory perception, global consciousness, famous incidents etc. Facebook wasn't invented yet.
It turned out that Google was behind it all. They needed to accurately tag the images they were storing in Google Photo, which the players did for them. And the players verified AI results at the same time. Millions of people eagerly played that "game" for years, just seduced by being able to check how well they could the "read the mind" of a random someone else somewhere in the world. It was successful crowd-sourcing in its infancy.
Anyone else notice the title change?
abysmal truant What was the previous?
What was the previous??
.
@bobbob123ful I honestly preferred the other one... It felt less clickbait-y. It was something along the lines of “the history of AI machine learning 300 years ago” . Now I wish I had screenshoted it XD
love the verge channel so versatile
0:35 WTF?!
Shocker. It’s great to know that we are providing free curriculum based ML training.
Companies are ripping off users and not paying there work basically.
Thats why you get to use it for free! You are unknowingly making a deal and probably even sign for it in the small print you didn't read when you hit "I agree"
human beings building their own demise one captcha at a time
that mturk robot looks creepy af
I like that twist he threw in at the end. 👌🏽
Did they change the description of this video or am I experiencing deja vu?
Captcha needs abolishing. I had one a couple of nights ago that took me through about 5 different sets of images before telling me that I had failed, even though I had tapped on the correct images. On my next attempt at logging in, it let me log in to a website after just 1 set of images. It's slowly getting worse and worse.
Amazing video. Great job verge!
im ready for the Singularity, are you?
CAPTCHA is an acronym for “completely automated public Turing test to tell computers and humans apart.” This can be with images/audio or whatever we will see in the future.
Why do sites use it?
CAPTCHA is used to prevent bots from automatically submitting forms with SPAM or other unwanted content. Google and other companies use it to prevent bots from creating multiple Gmail accounts.
finally we're not training here any ML model.
get better on solving the capatachas.
So we're calling machine learning AI now?
When we select the pictures with car and it tells us that we have selected the right picture, but if the system already knows that which pictures are of car why do they need us to train AI then.