I taught an AI to stan BTS
Vložit
- čas přidán 21. 07. 2020
- The first 1000 people who click the link in the description will get 2 free months of Skillshare Premium: skl.sh/sabrina2
Have you ever wondered how face id worked? Have you ever worried about the future of facial recognition? In this video, Sabrina attempts to teach a machine to recognize kpop idols and musical groups like BTS, EXO, Blackpink, and Seventeen but eventually stumbles into the far darker world of the ethics and safety of facial recognition.
In this video, you’ll learn a bit about computer vision, facial recognition, and the debate around the t future.
FOLLOW US ONLiNE
Sabrina
Twitter: / nerdyandquirky
Instagram: / nerdyandquirky
Taha
Twitter: / khanstopme
Instagram: / khanstopme
SUGGESTED READING
A straightforward walkthrough of facial recognition implementation
/ machine-learning-is-fu...
Summary of the ethical debate
towardsdatascience.com/how-et...
Example of facial recognition failing
www.nytimes.com/2020/06/24/te...
Some applications of the technology
www.theguardian.com/technolog...
www.wired.com/story/delicate-...
www.fastcompany.com/90310803/...
MUSIC
Epidemic Sound. Get started today: share.epidemicsound.com/answer...
EQUIPMENT
The following links are generated with the Amazon affiliate program, meaning we get a small kickback for any purchase.
Canon EOS 70D Digital SLR Camera with 18-55mm STM Lens
amzn.to/2Y9tNfz
Canon EF 50mm f/1.8 STM Lens
amzn.to/2VZk0FW
Rokinon FE14M-C 14mm F2.8 Ultra Wide Lens for Canon (Black)
amzn.to/3cQTkyd
DJI Osmo Pocket - Handheld 3-Axis Gimbal Stabilizer with integrated Camera 12 MP 1/2.3” CMOS 4K Video, Attachable to Smartphone, Android, iPhone, Black
amzn.to/2Y3gLAb
Blue Yeti USB Mic for Recording & Streaming on PC and Mac, 3 Condenser Capsules, 4 Pickup Patterns, Headphone Output and Volume Control, Mic Gain Control, Adjustable Stand, Plug & Play
amzn.to/2x6sUsR
Neewer 2 Packs Dimmable Bi-Color 480 LED Video Light and Stand Lighting Kit Includes: 3200-5600K CRI 96+ LED Panel with U Bracket, 75 inches Light Stand for CZcams Studio Photography, Video Shooting
amzn.to/3bxCDaH
Dell XPS 15 laptop 15.6", 4K UHD InfinityEdge Touch, 9th Gen Intel Core i7-9750H, NVIDIA GeForce GTX 1650 4GB GDDR5, 1TB SSD storage, 16GB RAM, XPS7590-7565SLV-PUS
amzn.to/2VYl0tR
--------------------------------
Welcome to the joke under the fold! Ready to face a horrible pun?
Your nose is in the scenter of your face.
Comment NOSE down below to let me know you were here ;-)
You trained an ai to recognize k-pop idols and thus trained yourself to better recognise school buses, incredible.
And provided a human check to google's school bus recognizing machine learning software. So at least two AI were being worked on there.
I wonder if there’s a site where you just do captchas to train AI, maybe the site could keep track of which captchas were done most often and you could maybe choose a less common captcha like…trees, haven’t seen one for trees yet
@@daniellewilson8527 q
School buses and deeply entrenched racism!
@@elliotmcdeville1117 yikes you're definitely reaching there pal
sabrinas fbi agent is really questioning why shes downloading so much k-pop
She's Canadian, so it would be the CSI agent. Just like in my case, I would have an M15 agent.
mine is wondering the same thing about me but i have a less legit reason
Then they see her making thia video and tgey're like "Ah, that makes sense"
pretty sure that's the normal amount of kpop
@@carolinemcgovern4488 for americans it'd actually be an nsa agent not an fbi agent
“Everyone is some member of bts” is honestly so correct about facial recognition like in my google photos library, it says like 90% of kpop idols are V even when they look nothing like him
Is my dad this facial recognition software?
I was kinda expecting my iPhone photos to be like that but it seems to do pretty well. a couple wrong faces here and there and one album of 3 different people each with a pretty equal amount of pictures but it generally seems to do pretty well so far
@@phoenixfritzinger9185 didn't it tell you it's a cyborg?
I guess I'm not that bad at recognising people after all.
@@lightishredgummi9654 so what I'm hearing is that the google facial recognition doesnt know asian faces but the apple facial recognition does. Interesting
6:00 "Consider this photo of Seventeen. A group with, arguably, too many people."
Or, arguably, five too few people.
there's actually 13 members bahahaha. need 4 more of them boys
@@cherrywyrm same. I’ve always felt that since the name is seventeen they should have just committed to actually having seventeen members lol
@@michellepauling7449 Yup, that picture she found was missing Seungkwan 😅
Doyoon, Ming Ming, Dongjin and Samuel moment
@@saveloona4695 no, they have 3 sub-units. the picture sabrina was using was missing a member.
In the wise words of Michael Reeves: "Facial detection? Nah, I prefer racial detection"
They tried to train a neural network to differentiate between wolfes and dogs.
In the end the neural network spit out "wolf" whenever snow showed up on the pictures because of bad dataset. So they failed at racial detection xD
@@benjaminmeusburger4254 who is they?
YES! Best quote ever
and she is also Filipino so...
I was just thinking about Michael Reeves lol
"It just assumes that everybody is some member of BTS"
Why is this so my friends and family
I was thinking the same thing
Why is it so everyone on Omegle when they see an Asian boy 🤩
Yeah it probably is akward too for other male kpop groups that everyone just goes ”BTS? Is it BTS?
This was also me when I first got into KPOP lol
wear a face mask:
Protects your identity
Protects you from COVID
with glasses
Face masks don't protect you, they protect others from you. If you want to protect yourself you also want to wear a face shield alongside your mask. Or glasses lol
@Alastair Valyocsik obviously but they would look slightly strange
@@miglek9613 You misspelled 'awesome'.
Steampunk ftw!
Also protects you from conveniently unlocking your phone with Face ID T^T
“They all look like a random member of BTS”
So apparently my dad is the facial recognition software
nah more like my uncle
@@shutup3378 Who knows, your uncle could secretly be Phoenix's dad.
“Seventeen a group with too many members”
All units of NCT: “am I a joke to you?”
*AKB48 intensifies*
Sabrina 😨 135 members! I never even heard of that group until now, granted I don’t listen to J-pop.
lmao rip new stans
seventeen, a group with THIRTEEN members, with an incomprehensible name and an equally incromprehensible amount of members. nct makes a tiny bit more sense because you're not expecting to see all members but even then it's ??
@@lovex3842 back to nct 2018 its hard lol
What I got from this vid:
1. Facial recognition is terrifying and I hate that I live in the future
2. Sabrina's graphics are GORGEOUS
Well, I watch two-minute papers so none of this scares me, even the code.
I think facial detection is great and I'm glad to live in the future
@@Zaurkax14 You do? that's cool. What do you like about it? The criminals/missing persons detection?
@@Cationna its nice in every day objects like phones, or cameras that detect if someone new entered your house, so it prevents burglaries, and obviously as you mentioned missing people and criminals. with crime its not as simple, because these people would just cover their faces, but with people who are missing, if they are not locked somewhere there is a change to actually detect their faces somewhere in a crowd. a lot of women forced into prostitution could be found. these chances arent high yet, because they are usually from third world, so we dont have their photos, but the more it all will be developing the more common it will be to have that stuff up to date.
some girls ger kidnapped and after years when they get a stockholm syndrome their kidnapper starts to go out with them, that could also get detected. i am sorry that my answer is so chaotic... i cant really gather my thoughts :)
@@Zaurkax14 no no, I absolutely get what you're saying. And I agree those things are great in many ways. If even one missing or enslaved person was found this way, it will have been worth the invasion of privacy. I just think it's not very likely because as the technology develops, so does the criminals' inegnuity in avoiding it. On the other hand, in the wrong hands it can be used to extremely grave and harmful effects - it can aid traffickers and murderers as well as help catch them. Not to mention how it can be used by totalitarian governments (which, we do tend to forget, are a reality currently in many places in the world). So to me it's still terrifying, even if - or, all the more so because, it's already unavoidable. At this point it already exists and is widely available and being worked on, so the question is only how far do we spread its use, how much do we normalise it. I know I personally would rather not have the comfort of using facial recognition in everyday life, if it also creates a record of my habits and of who visits me at home etc. - because that database is stored somewhere and therefore can be hacked. Of course we can argue it's more a question of data protection, but the fact remains, as long as the data exists, it's usable - so who do we allow to use it and why? how do we enforce it? how do we absolutely eliminate the danger of it being used to ill intent? Well, I certainly don't have the answers..
As a humanities major who writes every single dissertation using an assortment of scrambled post-its and paper scraps, the fact that you, a science major, used a tidy notebook but your friend, a humanities-ish person, used post-its, is extremely hilarious. I was not expecting this video to take the turn that it took, and it learnt a lot of (scary) things about how facial recognition works. Great work!
Bruh I dont understand a single thing you sayin!
@@ferdinandinc2899 They are taking the same major as the guy in the vid. They think it’s funny how the guy writes his notes on post-it notes and this person does too, while the person who made this vid is taking a different major and taking notes in a notebook.
@@aperson2140 ohh!! Appreciated✊
But developers also use postits a lot... Have we secretly been humanities all along? :o
"As a humanities major who writes every single dissertation [...]"
_Every single dissertation_ ? Like how many PhDs do you have by now?
the "big thing about kpop stars....THEY ARE ASIAN." part had me dying, like i knew where that sentence was going and i legit said it aloud i-
"People can tell I'm Filipino. Like, look at my nose bridge. My glasses are falling." SAME GIRL SAME
LOL THIS IS ME but I’m korean... I actually wrote an essay about my nosebridge for college HAHA
Ngan Dinh haha yeah luckily some of my top schools really liked it!!! Gotta owe it to my nose bridge lol
For a Filipino that's not even a low nosebridge, mine is lower 😆
This has called me out and I am uncomfortable this is why I prefer contacts XD
rel8
So far:
Sabrina 0 - 2 Machine learning
Big oof
Third time's a charm!
Idk. Wasn't this time a draw? It wasn't her fault that the algorithm was trained on a bad dataset, but she didn't win either.
@@Kenionatus Depends. If the machine's goal is simply that Sabrina fails, then it won.
Unrelated to the actual content but I'm loving this series of "I wonder how X works, I'm going to do a huge personal experiment on it and enlist a friend to help" videos. I hope you keep making them!
love how this went from kpop fancams to machine learning to facial recognition being racist then goes to obscurity issues and chilling effect
I love hating on most k pop idols
Wait shi-
@@secondlieutenan "You are under attack by Kpop stans"
Facial recognition software: genuinely scary example of a future without anonymity
Sabrina: haha kpop idol go brrrrr
i studied cognitive science (which is basically ai, the major) and my data science professor told us a story about a facial recognition ai made for airports that began telling asian individuals to "open their eyes" because it had only been trained on white faces. my prof made a HUGE point out of the potential dangers of ai and data science, and spent several classes on the unintended consequences of developing new technologies. so not to say that it was a good thing your ai was not able to immediately recognize jungkook, sabrina, but i think talking about the ways technology is biased (bc it's made by biased ppl) and the potential consequences of any and all technology is super super important. thanks for this video, it was very informative!
and also stan bts :)
Yup, AIs trained by ignorant straight white techbros are already proving to cause harm in more than a few areas, sadly.
💜
Not the making a machine racist-
Q: Do you remember the majors points he made in those classes?? any sources and articles to read about those other-than-white points of view ??
you ignited my curiosity and left me hanging
plz make a good long list :pppp
stream butter
Honestly i came here because of BTS on the title, then i left with all these knowledge, wonder, anxieties, and gratitude
What I wouldn’t give to have friends like Sabrina‘s...
“Do you want to spend hours learning info to help me make a CZcams video?”
“Aight I can do that.”
"vibe in the horror of man" is truly the motto for our time.
I've only really thought about facial recognition as being used by the government to track your behaviour which I was barely okay with. I had never really considered that a casual murderer on the street could use it to find where you live and work. So thanks Sabrina? For new nightmare fuel.
He could just follow you home.
Some ai-developping entreprise (I think it was Clearview AI ?) was promoting its facial recognition AI with "what if" posts such as "what if you could know where a person lives just by seeing them ?" and it got some pretty huge backlashes on internet, arguing that it would only cause the death of people (especially women..), but I guess that it doesn't stop the ai to exist and be sold... really nightmarish
@@Zaurkax14 But he wouldn't have to. At least with following someone there's the potential of being noticed by the victim. With facial recognition to this extreme they could break in and wait inside
@@theo4907 they can pick a lock just as well. They can be watching you every day and you'd not know. Following you doesn't mean literally walking beside you like in movies. He could be stalking you for weeks, learn your schedule and then attack.
If someone is bad they can do bad things even without an technology at all.
And I'm pretty positive that this kind of technology can do more good than harm for majority of countries.
@@Zaurkax14 yeah but most people would not feel like following a person around for weeks, even if they want to kill them. A lot more people would murder someone if they could figure out how to strike unnoticed in a couple of hours
This is literally all I've ever wanted. Kpop, coding and sociopolitical & ethics . My perfect trifecta. This is amazing, props to you.
finally, someone who gets me!
Even Google thinks V and Jungkook are the same person on google photos, so don't feel too bad lol
ahh this is scary and i dont like it, i dont trust govs and big companies not to use this for evil and i dont trust that anyone will be motivated to regulate this properly - at least not till after a big awful thing happens and damage is done.... eek
p.s. your production and graphics and sound quality are all incredible! this is such a well made video
+
+
How could it be used for evil?
Wait nevermind, I just finished watching the video.
@@tagz5186 yeah..... :( also like general lack of privacy is really not ok, even if you consent.
"Invisible Women (Data bias in a world designed for men)"
by Carolina Cardio Perez is also a great book on the topic of bad data
Highly recommend it
Second this!
third this
what a great recommendation, i had never heard of this book!! thanks!!
I've never laughed harder at the words "THEY'RE ASIAN" in my life. From one Asian to another, thank you Sabrina!
"Don't want another mac'n'cheese without macaroni."
I understood that reference.
I was NOT expecting a taha but im HERE for it
i saw him and i audibly said "tahaaaa!!"
What the bloody hell happened to the like button?
@@hioyua650 what do you mean
@@yoavzack LOOK AT THE BUTTONS they look weird. You know the ones below your comment? Like and dislike button.
@@lauralettau Does he have his own CZcams stuff? I didn't notice it in the description, but Im also tired...
As you can see at my left, another reason why facial recognition will eventually be so advanced that plastic surgery may be the only way to avoid identity theft, and to my right... stan twitter, shield your eyes.
Sabrina: i want the machine to identify kpop idols for me
Racism: NO >:(
Sabrina: :c
It's not racism lol
this went from "oooo k-pop and ai" to "how exciting, machine learning" to "what would happen to your identity and entire existence with facial recognition" way too fast
It's scary the stuff you can do with data and statistics which pretty forms the basis of all AI systems.
The budget and effort that goes into these videos... criminally underated.
After watching your first AI video I actually started to try it out because of its accessibility and I grabbed all of the rain sounds I could find on rain "generator" websites and trained an RNN and now I can actually generate rain sounds (it's just not efficient enough to generate rain sounds in real time... by more than 1 magnitude).
do you have any inputs like thunderstorm rain. or is it all random?
@@scno0B1 Currently it only does rain, but I'm working on collecting together thunder-only sounds because with thunder it would sometimes get stuck generating endless thunder with no breaks
please make a website for this
@@precumming That's cool. I like rain sounds as background noise for relaxing but thought provoking music.
I wish there was an app that let you train an AI using sounds, I have lots of mp3s on my IPad and would be curious what the combined data would make if an AI was trained on it
The real issue with facial detection is that it has difficulty distinguishing individual faces if the pool size is too large.
we look too similar to other people.
If it's only comparing a few faces it can do fine, but at a city scale, it will falsely flag a lot of people who look extremely similar.
This. For better or worse, I have a pretty unusual-looking face. But I've heard from multiple people that they saw someone who looked almost exactly like me at the high school across town / at another company. Not common enough that I'd see them in a multi-million-resident metro area, but common enough that it happens and could have feasibly been me.
Which is not okay when we decide to criminally prosecute people based on facial recognition software.
"I got so caught up in the euphoria of robots that for a minute I lived in a world where racism didn't exist"
As a machine learning engineer, I have to say you did an impressive work. The training data is everything, the more precise it is, the better the model will perform on the task. 90% of the time should be spent in collecting meaningful data, the rest on the model. And ML models are discriminative by construction, the more you want to generalize, the lower will be the accuracy on new data. Fairness of ML is a whole field of studying and it's very interesting, aiming to be non discriminant on features that are important for us (i.e. races, gender, etc.) while still trying to keep the prediction of the model meaningful.
Keep up with the great work!
I have hundreds of photos of BTS saved on Google Photos from across all seven years of their career, and I've found some interesting quirks of facial recognition that were mentioned in this video. Namjoon, the beginning example in this video, is very easy for facial recognition, 99.9% of the time the algorithm is successful. Jimin is accurately identified as himself probably 90% of the time. But Jin is identified as Yoongi a ridiculous number of times, despite none of his features even remotely looking similar?? Same with Jungkook! In fact, if it's an Asian male (other BTS members) the app more than often just lumps them all as Yoongi! It's so bad that I literally have a folder labeled "pictures of Jin, because Google is racist". But as the years go on, and the app fine tunes itself with my constant correcting of it, it's unsettling to think of the implications of this algorithm getting better. I'm just trying to organise my references for fanart, not contribute to Big Brother
i just LOVE LOVE LOVE your comment!!
So… does that mean Yoongi is the one who is actually Worldwide Handsome?
This is so true. The same thing happens to me with my Google photos.
interestingly enough, when I was getting into BTS and trying to tell them apart, that was the order I learned the members' faces in! I could recognize namjoon easily and then jimin, then V, then j-hope, then jin jk and yoongi I'd struggle just a bit (for context, im chinese-filipino that moved to Canada in my mid-teens). I wonder if its really just because the proportions of their faces for the last 3 members are quite similar?
I'm an artist and used to draw BTS fanart. While drawing them you sometimes do end up with Yoongi when you planned for Jin etc, etc. It's a bizarre theory and most people would say it's crazy, but it made me think how the older they get and the more time they spend together, the more similar they end up looking, because it doesn't happen as much with their likeness when they were younger. I know it's impossible to adjust your FACE to your friends, but I swear it's what's happening to BTS. I'm not new in the fandom, I've been here since 2016 and I swear it's true :D
Once again I accidentally got educated by virtue of stanning BTS lol. Seriously though, kudos! This video is incredibly well produced and entertaining, taking on a serious topic in an approachable way. I'll definitely check out her channel more.
LOL sameee
OMG who do you stan? They're all so unique!
What the computer sees: 🗿 🗿 🗿 🗿 🗿 🗿 🗿
"Vibing in the Horror of Man" is the 2020 mood
Do y'all remember that Amazon store where nobody checks out bc facial recognition identifies you and charges your Amazon account? Asian block party at that store.
Okay but the face when you hit the captcha that was literally just having you sort pictures for this exact purpose killed me
My dad is South Asian, and he was thrown into a folder with anime characters and K-pop photoshoots 🤧
My non identical twin sister has the same face as me, according to software. Our faces have like small differences (like different freckles and stuff) and we have always had different. BUT DOES IT MATTER, NO.
What about that time my brother shaved his beard, and the software thought it was me.
Ooh or that time I cut my previous long hair short, but on a little makeup, and lost some weight, and Apparently according to my phone I wasn’t even. Lol.
@@fabuloushatlady7812 Facebook seems to think that I am my mother or my two-year-old niece.
"im a very average person (and that's on being asian)" ASDFASDFASF
me, Asian: "I'm not exotic , I'm statistically average"
what does "ASDF..." mean?
i think i saw a zuko fancam in the blurred out scroll
edit: i watched it at half speed and its definitely a zuko, azula or iroh fancam
what a cursed image of joon we had to stare at for that whole sequence
I feel like an interesting element of this lack of privacy in many ways is being explored by celebrities and youtubers who already are widly recognized and unable to blend into the background. I'm not sure i have thoughts beyond that right now, but that connection is itnerseting to me
Oh yeah.... that’s interesting. Even without facial recognition there are still some people without an anonymous shield.
That's exactly the kind of thing the video is talking about. Public areas with surveillance cameras (which is most of them right now) AND facial recognition software will make Going Out In Public as dangerous and predatory as it is right now for celebrities Going Out In Public. That's exactly what losing anonymity is like.
Except it's not just fans and paparazzi hounding you, it's multibillion-dollar corporations and the government. So it's worse actually.
@pluspiping when taha said facial recognition could be used to locate vulnerable missing persons, my immediate thought was "and to locate vulnerable people who are "missing" for their own safety." Social media is already being used by abusive partners and family members to track down people who escaped them by making posts about their "missing" partner/relative/etc. This shit gets people killed.
We should stop tagging ourselves and other people on photos. The future is dark and full of terrors.
I like how she teaches us by learning with us so it doesn’t feel like we’re being talked at but rather talked with. If that makes any sense.
Exactly why i love this channel
i kinda feel good? that i can recognize idols better than a computer 😂
Many models have problems with dark skin, even when trained with a broad dataset bad cameras or dark lighting tend to have a stronger effect on darker skin, like when you watch a video dark gradients are way blockier than light gradients.
A better camera would use electromagnetic waves that aren't affected by melanin, outside the visual spectrum. There's just the problem that we have to deal with pretty low or relatively high energy photons then, or need a different measuring technique all along.
While we're on the subject of A.I. and facial recognition, I want to bring up somethingthat I think is interesting:
My phone allows me to tag faces in my gallery and based on who I tag as myself, it guesses who is me. It is also able to guess where faces are.
When I took a picture with my sister, the phone put a blue square around my sister's face, meaning it was guessing it was me. Me and my sister look really similar, but it isn't obvious. This crappy 2006 phone was able to pick up on all the little things that made my sister's face match mine.
It was also able to find a face in a cartoony piece of fan art that I saved, which seems like the kind of thing this old phone shouldn't be able to do.
Then again it also thought a tree branch was a face so results may vary I guess.
I think I had a dangle earring tagged as a face in the early 2000s.
sabrina: seventeen, a group with arguably too many people.
nct: *cries in 23 members*
There's already apparently a huge hidden camera problem in South Korea. Imagine hidden cams paired with facial recognition software. 💀
Eurocentrism also plays a part into AI facial recognition. Because alot of facial recognition is based in European and American countries, little data is used on Asian people.
This is not necessarily as fault to white people, but currently, AI recognition is styled more to White people not Black people, and even less so Asian people, simply because of racial populations where AI recognition takes training data from. In regards to the evolution of facial recognition, more data would need to be taken of non-white people to improve it (so MORE photo(s) (databases?) from not-North American or European countries).
Wether training it on racial minorities (in western places) is a good thing is up to you.
(It's already been said, I'm just explaining the point a little further.)
in addition, darker skin tones reflect less light, which can make facial recognition more difficult. the method sabrina uses here would get drastically different results on a black person who has much more prominent "shiny" spots on their skin, whereas a lighter skinned person would seem "shiny" all around
yeah i was thinkin the same too! similarly, for humans, there's the Cross-Race Effect. we're much more likely to tell apart faces of people from a certain race if we're more accustomed to it. this is why Asians can't really tell apart faces of POC and white people as well if they come from a homogeneous country like Korea or Japan. it's also why the stereotype that all Asian ppl look the same originated-because they haven't seen enough Asian people to tell Asian facial features apart. ask a white person in a European country if they can tell POC apart or ask a citizen from a South American or African country to tell apart white people and they won't know the difference. (side note: all the pictures should be of different people with similar facial hair and facial fat so as to not make it easier to tell them apart, obviously).
China’s facial recognition has no problem with asians.
It’s only the public open well known facial recognition softwares that have problem.
This is from a western perspective though, I am confident in saying that China isn’t training their algorithms with mostly white faces.
Guys it’s a bad idea to train AI to recognise faces in South Korea, they are still at war with North Korea.
I follow you since you won that vlogbrothers award....damn you've evolved so much
I'm so happy that her channel has grown so much!
i came here because i saw BTS but i stayed because it was interesting lol
same 😁💜
The way her search history was just kpop fancams was everything...
I'm a simple person. I see kpop, AI, and Sabrina, I freak out.
Seriously you're making videos I didn't know I needed
Kpop idols changing faces cuz of surgery is such ignorant stereotype. Im sure it’s happened but if you’ve been a fan of kpop you know how much backlash these idols would have if they do something to their face. So its very rare for them to change their facial features. Yet western celebrities get surgery all the time yet its kpop that gets pigeon hold with that idea
I agree it’s definitely an exaggerated stereotype, but South Korea is still has the biggest cosmetic surgery industry in the world, and many idols do get minor changes to their faces (usually before debut or early in their career so it’s not as noticeable). But yeah usually when an idol looks very different than they typically do it’s just because of something else like different makeup or hair
I came here to see kpop idols being recognized by ai and ended up being concerned about my bank app constantly asking me to activate the facial recognition option.
Your and your friend's voice-overs are so clean! They sound so good~
Wow never been so early before! Sabrina is a genius. A legend. We STAN!
I want to like this comment but it has 69 likes soooo....
"Vibe in the Horror of Man" is 2020s new slogan.
10:45 i love how she is playing SNAPPING by CHUNGHA.... That song is a bop !!!!!!!
I was shooketh when some downloaded childhood photos got identified by Google photos as still me... These were like 20 year old pictures and I don't look the same 😳
"its horoscope detection for wallstreet"
THIS made my day, im laughing a lot
haha XXD
Other people: K-pop idols kinda look the same
Sabrina: ...
They honestly don’t. As said in the video, the AI are not trained to Asian faces and features. Each race has some commons characteristics, so if it’s trained to be an expert in the small deferences in white people faces, it has a more generalized look for Asian and black people
Just_ktrash UwU honestly, that’s great. Now destroy facial recognition all together
this is the first video i've watched from you and i thought i was gonna be like "kpop idols haha" but instead was blown away by the gorgeous visuals and amazing editing skills, as well as some mild existentialism from the development of technology lmao. gonna check out your other videos now, this is amazing stuff!
now if school taught me things using kpop I would listen and learn so much faster T-T
lmao same
You would not. You would start to dislike kpop instead.
Wow now I’m depressed, anxious but also happy that I learned this and that I am a bit more educated on the subject of facial recognition
I loved this video! I think it raises very important questions that we should all think about as the prevalence of facial recognition becomes more widely available and governments start taking advantage of this.
I also loved your keyboard and keycaps, where can I get these? 👀
I went in expecting to enjoy myself and learn a few things but... now I can’t sleep knowing that it’s that easy to use facial recognition.
Also, much love Sabrina from the Philippines!!
I like playing with visual AI and I'm always amazed how easy it is to do so. In 2 hours I can create (from zero) an application that can detect what emotion you have. It won't work well 100%, but I can do it fast and easy. I don't think that we should debate IF AI will be used for bad things, but WHEN. There will be someone using good tools for bad, but in my opinion, it will be worse when we'll use a faulty tool without knowing it. There are already examples. The most known are the deep fakes, but perhaps worse than that is someone using face detection to detect criminals and because of the data used, it had a bias detecting black faces (you already mentioned this as a thought experiment, but this is real). As far as I know, the project was scrapped, but someday we may use such tool without knowing it, as you used this one in this video. AI will be used for bad thing, whether we know it is bad or not. We just need to be prepared and informed when that happen and at this point I think we are neither.
you deserve all the attention youtube can give! your work is amazing, thank you so much
Great video! Really interesting and also terrifying!
This was an incredibly well put together video. I cannot explain how good the script was, the editing, the animations, the structure of the story, your effort in coding (which in itself is no small thing), the way you brought in the expertise of your friend...I basically never share content with my friends on the internet but this one is worth it.
namjoon is cute even with that bad haircut and teal color 😔
NOSE anyways so far i am really amused that the first iteration of the machine was saying everyone was a bts member and also saw jungkook and was like, who?
Awesome vid! Chilling, entertaining, educational... I love the prospects of machine learning but conversations around the ethics are so important while the technology is so young. Glad to see you bringing more light to the topic for laypeople like me
"The Asian problem" sounds what a middle ages white european ruler would say
this is so cool! i love how this video started with my personal interest (k-pop) and then delved into my.professional/academic life (public policy + smart cities). questions like "how could facial recognition technology be abused" are incredibly important questions to ask, and we need to put policies in place to make sure those worst case scenarios never happen.
This is some really good content!!
I love how you dove deeper into the fundamental issues of facial recognition instead of focused solely on the code.
As cheesy as it sounds I'm super glad I subbed to you!
Epic stuff, doing cool things that I’d be curious on but wouldn’t be able to do if I thought of it myself!
yes so this is one of the best videos i’ve seen in quite awhile. the animations, the information, the levity as well as the gravity, it was all so freaking enjoyable. thank you for the awesome content!
this twenty something video completely flew by?? the animation was so nice and everything was informative and this was really cool to watch overall lol
The GDPR implementation is so so poor and here's the real reason why: it should be an automatic opt-out, i.e. when you click the "do i have a choice?" option they should be off by default and THEN you can click "confirm". but a lot of sites do not do this and are actually breaking the lawwww!
I love Sabrina, she’s awesome!!!
this was genuinely SUCH a good video oh my god!! it's been a while since i've last sat through an entire 25min long video without going on my phone or anything but like. i was basically glued to the screen the entire time!!
I'm watching this thinking someone can theoretically make a chrome extension that grabs a few frames from fancams on Twitter and tells you who that idol is from which group.
"Don't want another mac and cheese without macaroni" lmao
This video is way more serious than what I expected from the title, i almost skip it. Great content!
Thank you for sharing great content!!!
aaaah your motion graphics are so memorizing !! great job 💖
This is awesome! I never realized this was how facial recognition worked.
This is interesting because for the vast majority of human evolutionary history, people probably did live in a context where every individual actually knew a lot about every other individual. People couldn't live in the kind of densities you need to be anonymous as an obscure "one of many" until like 10k years ago max. So in some ways, this technology is actually taking us backwards rather than forwards. It just feels particularly jarring to those of us who experienced what recently came before, especially for those of us who fondly experience the extreme anonymity represented by a certain stage of the internet.
Probably our societies will adapt and develop modern versions of the cultural technologies people have used in the many other contexts where everybody knows everybody. Not the same thing, but look how fast we all adapt to talking around words and phrases that are commonly censored or controversial. In contexts where we know that machines are closely observing our every move, we use peculiarly human leaps of symbolic thinking and communication to make ourselves unintelligible to those machines.
Hey Sabrina, your animation skills have really improved! It's been a while since I watched your channel and was blown away by the quality of the content! Kudos to you!
Just wanted to say that, aside from the really cool topic, it was a joy to learn about all of this in such a well presented manner. Loving the animations! :)