211 comments

[ 3.4 ms ] story [ 246 ms ] thread
Very interesting, but hangs at the last(?) step (life expectancy)
Could they be trying to make a point? About insurance abuses of this technology maybe? maybe not, could just be the algorithm failing. Or maybe they don't want to assume the liability of actually providing a number.
No, it's a javascript error. The second time around I agreed to the terms and conditions and then the site worked fine.
Yikes. Thanks, I'll look into it. The terms and conditions have no influence on this, but it could that that one of the scores required to create the life expectancy prediction is missing (age, gender, bmi, country).
My life expectancy was the only score that it didn't give me. It just stayed at ... and then marked that as not normal
I found the issue, and it should hopefully be fixed. It happened if your IP address couldn't be used to predict your location.
Does it fall back to US if the IP address can't be used? Because this would be the first time a GeoIP database would move my EU IP address to the US...
It looks like they're trying to guess at your life expectancy through your country, which they're guessing at using your IP address, which they can only get at if you agreed to the terms and conditions
That was simply amazing. Having the video guide you through all the machine learning prediction steps as they got analyzed on your video feed worked flawless, both in terms of technical execution but also presentatio nwise.

They also made some great points about how ridiculous some of these algorithms are (shake your head to lose 10 years of age, or just trying to predict BMI from a face photo)

And I was surprised to know that this has made its way into hiring! So, the job hunt is going to get even more superfluous? Really reminds me of that Episode in Black Mirror, where everything depends on your rating..
Apperantly when the website thinks you are a woman, it marks that column down as "Normal: No", because 82% of visitors are men.

Just something to think about.

Also it called me ugly :<

Edit:

I just ran the website for a second time and didn't change my clothes or anything, and it gave me very different results.

ie. gender, age, bmi, attractiveness all different.

I don't think the ratings for this is very stable & reproducible.

at least it rated my girlfriend as abnormally attractive. :)

Age and gender are derived using the FaceApiJS project. In my own experience it's doesn't vary as wildly if you're a white male. With other ethnicities the predictability levels can drop, especially if you're Asian.
Hey it's you! You wrote social cooling too, which I really enjoyed.

Great site. Thank you for sharing it.

What was the beauty score for that? Number wise. The scale is huge!
> I don't think the ratings for this is very stable & reproducible.

I think that's part of the point though -- the claim is that these are being used, regardless of how predictable they are.

I think this is entirely the point - to highlight how these algorithms are used in every day life, and how unreliable they are.

It points it out by "raising your eyebrows to make it think you're less fat" and "move your head around to change your age".

Interestingly it guessed my BMI at 38. To get to that number for my height I would have to be twice my current weight.
Yeah, if you train a model with pictures of the same person, or maybe their family, you may be able to guess the BMI from the face, after all, the face stores fat in a non linear way.

But I remember back in college a friend was severely overweight and eventually he was motivated to lose weight. Diet and exercise details aside, in a couple of years he became a BMI 28, but very low fat, ripped and chiseled muscled up individual. But interestingly, he mostly kept his fat person face, those cheeks, while they shrank, weren't going away.

"Violently average" oof

Great website to share with my non-tech friends and family, they're tired of my rants but this is very well articulated.

Any chances of including other languages for wider reach?

67% normal

abnormally young, which is a normal observation by others of me.

The life expectancy is probably fair even though

BMI is just about right, 19.6

Am not a good boy :-D

This is fascinating and I also approve of the cause and message. Yet more reasons to have a simple camera cover.

Accuracy side, I thought the presentation was excellent! I really like the explanation playing while the analysis is taking place. I’m totally going to steal this presentation idea.
The average beauty score ranges from something like -150000 to +150000. Not sure why!
I suspect someone has tried to manipulate their score. I found one database entry that had only an extreme beauty score. I've hardened the system against this now.
LOL funny how people here gibber about privacy but are more than happy to give their face to some rando website. Y'all are fucking muppets.
Is a picture without personally identifying information problematic? I assume pre-COVID you didn't walk around in a mask all the time.
I wonder how many people who actually upload a photo to this site will even bother to remove EXIF data?

If you have all the original metadata in a typical smartphone camera JPEG, then things like the info below is available:

- camera model

- phone model

- time & date of when the picture was taken

- geo-location of where the picture was taken

I would say for anyone motivated enough that is pretty easily identifiable.

edit: bullet points formatting

edit2: lewispollard rightly pointed out that I should read before writing. :)

It uses a live feed from your camera, not a photo.
I share your concern and decided to not try the website. The site said no personal data is collected, and if we opt to share our data, it would be anonimized. Well yeah, but how about browser fingerprint? is that considered as a personal data? or if it's collected at all?

edit: Okay, I succumbed to my curiosity and decided to try it but with all my connection turned off. As far as my obeservation goes, at least there are no ajax call from this site and everything is functioning even without connection.

Terms and conditions

This is an art project by Tijmen Schep that shows how face detection algoritms are increasingly used to judge you.

No personal data is sent to our server in any way. Nothing. Zilch. Nada. All the face detection algorithms will run on your own computer, in the browser.

In this 'test' your face is compared with that of all the the other people who have previously done this test. At the end of the show you can, if you want to, share some anonimized data. That will then be used to re-calculate the new average. That anonymous data is not shared any further.

Did you know that online communities are often comprised of multiple people that will act differently?
You will not find these on mainstream media. They will not be promoted.
I don't trust elites or any product which quickly gets popular on any form of mainstream media. There is so much dirty money flowing around and most of it goes to lying psychopaths (the selection bias for psychopathy is systemic and seemingly intentional). Our modern economic growth is founded on coercing people into continuing to accept currencies which keep getting dirtier and which keep losing their value. It's 100% about coercion, manipulation and gate keeping. If you can't figure out what the product is that they're selling, you're the product.
Can't switch to front camera on pixel 3a?
Same problem for me on a pixel 3.
The same in Firefox on Samsung Galaxy A40, Firefox version is 81.1.1 (Build #2015764547). I missed the puppy because of that.
Same with Firefox and Chrome on Pixel xl
Yikes. I'll look into it. I take it you tried the button in the top-left to switch?

Were you able to select the other camera if you reloaded the page?

the only data I reported was my age... and it had some big misses, like BMI being much less than it really is just because of beard. Also it said my life expentancy is going to be up to 81 years old, when in reality I will live forever :')
Nice project :) Too bad it doesn't say how old it thinks you are though (or I missed it) All I know now is that it thinks I lied about my age, which I didn't.

I'm just going to assume I look like I'm 20

Ah, this one couldn't find my face. It kept focusing on my left ear and gave it a 5 for attractiveness. Then it couldn't find my face any longer. There was another fun one posted here months ago [1].

I wanna say I'm disappointed, but I'd be lying. I'm happy that A.I. cannot detect us black people. I'll need to do something about that left ear though.

[1]: https://news.ycombinator.com/item?id=21337863

The only person I know who did something about his left ear, did admirable. Just don’t wrap the ear in a newspaper and hand it to a prostitute, as I doubt that sort of thing will work out well in these modern times.
Before googling: Is this a van Gogh reference?

After googling: Yep, that's a van Gogh reference.

https://www.vangoghgallery.com/misc/van-goghs-ear.html

> Vincent took the ear and wrapped it in newspaper. With a hat pulled down over his wound, he, with ear in hand, left the house to go to a “maison de tolerance”, a brothel close to the house. There he asked for a girl named Rachel who he gave the ear to saying “Guard this object carefully.”

Whatever drugs he was taking, I want them
It was just that he had forgotten that she had invited him, and it was a Sunday and all the gift shops were closed...
Of course, that makes it totally acceptable now haha
(comment deleted)
I don't think it worked out well in his times either
I think the crowd sourced "am I hot or not" did a better job.
I have a big beard and thick black framed glasses, often image recognition can't find my face either. And I'm also happy about it. :)

I'm extremely nearsighted so through the glasses my face looks smaller than it actually is. The outlines of my face are shifted.

Whenever my gfs daughter wanted to play with snapchat and me I had to take off the glasses for it to work.

TL/DR; it's a off-hand joke.

Cannot decide if I should respond with a meme reference to Samual L. Jackson asking "Does he look like a *?" or ask "Do you also cook? Do you have a deep voice? Are we only used to seeing your torso and your hands? Do you have a popular YouTube show?"

But frankly, that's interesting the accommodation you have to make for an algorithm to play with your daughter. I wonder what other accommodations people make in life because of algorithms.

According to this AI I am about as attractive as your ear.
It started with 5, then I kept getting closer to the camera until it was at 7.5 ¯\_(ツ)_/¯
Its probably focusing on a single part of your skin.
I started with 5, and got up to 6 by angling my head down and to the right and raising my left eyebrow.
I'm not black, but have a beard and I got the same results as you. I want to say maybe this algorithm needs some work.
Burkas, beards and sunglasses are the natural predators of facial recognition algorithms.
It's okay, the wall in front of me scored 6.4, was a 14 year old male with a BMI of 36.1, and had a happy reaction to a dog.
I kinda wanna see your ear now.
Maybe someone can enlighten me. I first heard about this problem 10 years ago. I'm sure it's been known far longer.

Is there some extremely difficult algorithmic problem preventing these systems from working on people with darker skin?

I had presumed it was a long corrected oversight but it still appears to be a problem.

Even photographic cameras have difficulty with exposure for dark skin. This is a problem as old as photography.
So the deltas are narrower creating a breakdown in the SNR for say edge detection?

Please excuse me if this is totally wrong and I sound ignorant, I assure you I am...

yikes! I seems to think I only have 20 years left to live. I'm only in my mid 40's
Meanwhile, I'm going to live to be 120! Boy am I glad I visited this site!
Tell me more about your other stats, I want to be like this. What should I focus on? My body mass lolz.
You need to move to a country with long life-expectancy. I am from Ukraine, so it has given pretty average life-expectancy for it ~69 years.
BMI was way off but the metric itself is such a crude generalization so this is probably excusable.
Mine was predicted much lower than it really is, but I'm lean with big muscles, so that is to be expected, I suppose.
Exactly! My real BMI is 25, and just like you, I am on the muscular side. The estimate was 19.
You may enjoy this blogpost I made that goes into how the BMI prediction algorithm was developed. It's the only one in the project that I trained myself.

https://www.sherpapieces.eu/overview/predicting-your-bmi-fro...

Like you said, muscles matter quite a bit. It's attractive to use photos af athletes to train the algorithm, because there are a lot of websites that show their photos along with their weight and height (which you need to calculate a BMI). But since muscles weigh so much more than fat, athletes have high BMI with slim bodies, which warps the algorithm.

I didn't use athletes for this reason (it's trained on a diet of 50% Chinese celebrities and 50% American arrest records). But I found projects that do use these photos.

My BMI is 25.2 and the webpage calculated 25.1.

It would have been interesting to see some summary statistics for each of the predictions.

What a great website and cool way to present a technical topic. Online education is still in its infancy and when we talk nowadays about good examples of online education, we usually think of videos and (in the best case) of websites where students can play with interactive models, solve exercises, etc. But this website demonstrates that you can still go mucher further in involving the listener as a person.

This of course requires an enormous effort in the preparation of the material.

Agreed. I found the presentation of the material more interesting and innovative than the topic itself.
I second this. I've just spent the summer putting together an online experiment for a student along similar lines:

https://www.doesmyrobotunderstand.me/ (please, feel free! We need subjects...)

I wish I'd known about the Javascript face analysis beforehand! And also how hard it would be to learn Javascript ;) And how to do that cool embedded video thing.

The immediate feedback about how I did is nice, but also may pollute the data in a way. Once I got it right, I new to retry a very similar expression for the next 1-2 tries. But then, most people might actually know how to make human expressions naturally. Personally, I appreciated this as a tool to help me learn how to human. :-)
This website doesn't work at all for me (Safari, iOS 13). After asking for permission the video becomes modal and impossible to close, completely hijacking my browser and forcing me to quit the app. That's the state of JS in 2020 I guess.
oh, thanks for bringing that to my attention, I'll look into it.
Interesting, it couldn't detect my face.
The interaction heat map blocks the Skip button. I just wanted to see the score so I skipped all the talk but that also meant that my skipping made me unable to skip after that step.
The AI rated me a 4 out of 10 and struggled to figure out whether I was a man or a woman. There goes my self esteem for the day.
Just shining my phone's flashlight to my face increase the score by 1 point. Maybe try with a better lighting? You might get a 10 score ;)
You're in good company: The app gave Henry Cavill 5.9 and Pierce Brosnan 5.1 in my testing.
So apparently the algorithms say I'm almost 10 years younger than I actually am (late 20s, algos say I'm just about to get my student driver's license).

I wanted to see how much the camera quality affects these scores so I wanted to try on mobile but I can't get it to switch to the front camera.

If you are on Android then you can try firefox mobile. It allowed me to select which camera to use.
The results were predictably "AI" = complete rubbish. Even though I'm a white male.

Our brave new world of "computer said so"…

Great presentation though, kudos!

Apparently I lied about my age, it predicts that I'm actually 8 years younger than I am which is fun.

It also tells me my BMI is about 9 points lower than reality which is also fun.

And that I've got as long left to live as I have lived already which is pretty awesome.

But it also tells me I'm ugly, which isn't so much fun :P

At least it correctly identified me as a man with 97% certainty. So I can stop worrying about that now :D

Yea, the AI didn't think I was attractive either, although I take care of my appearance :/
It said I was 16 years younger than I am. Which is nice.
Same here. While I know that I look younger than I'm - once many people assumed that I'm 10 years younger. So it is quite cool to fool algorithm too.

As well algorithm thinks that I'm quite ugly. Well, I don't care even if algorithm is right, as algorithm is not my second wife.

This is such a flattering website:

* 5.5 with glasses, 7.5 without. Glad I wear contacts, I guess.

* Age predicted: 20, truth: 30+

* BMI predicted: 18.7, truth: 22

This was their plan all along. To harvest the data people leave in the comments. Cunningham's Law in effect ;)