AI can tell who is gay and straight

[x65943]

What are the political implications of a photo revealing one's sexuality 91% of the time??

Facial recognition predicts cis male sexuality accurately 91% of the time

https://www.theguardian.com/technol...ether-youre-gay-or-straight-from-a-photograph

AI trained on brain wave patterns predicts cis male sexuality accurately 83% of the time

https://www.quora.com/Where-can-I-f...tion-that-identifies-gay-people-through-faces

 

[Veho]

Well that sounds a lot like phrenology with extra steps.

I wonder if the sample used to train the AI was the same one that gave us this gem:

Anyway, as always with this type of "research", I want to know how they compensated for the sample being extremely skewed.

What are the political implications of a photo revealing one's sexuality 91% of the time??

Depends on the people writing the policy, doesn't it?
The "political" implications are that some might seek to prevent this technology for invading privacy, others might use it to keep tabs and not give a shit that it's not at all accurate.

 

[SeventhSon7]

 

[Xdqwerty]

Bender from Futurama???

 

[Spring_Spring]

Those faces are so generic that they don't really indicate anything. Also, why does any of this even matter?

 

[x65943]

Well that sounds a lot like phrenology with extra steps.

I wonder if the sample used to train the AI was the same one that gave us this gem:

View attachment 404146


Anyway, as always with this type of "research", I want to know how they compensated for the sample being extremely skewed.




Depends on the people writing the policy, doesn't it?
The "political" implications are that some might seek to prevent this technology for invading privacy, others might use it to keep tabs and not give a shit that it's not at all accurate.

sexuality is theorized to be linked to in utero hormone exposure, which would obviously alter more than brain anatomy - which might explain the differences in face shape

phrenology is horse wash obviously, but it's silly to act like there aren't any anatomic differences which might be related to a variety of different hormones or other factors that might affect one's personality

It's important to note although there was a high level of accuracy, still about 1/10 times the algorithm was not able to properly predict sexual orientation

There are a ton of random things like this tho - like for instance there is a small but statistically significant relationship between nose length and IQ

Those faces are so generic that they don't really indicate anything. Also, why does any of this even matter?

Well imagine a country like Iran got ahold of this (or another state in the ME) and started using it to harass people - or ad agencies start using facial recognition to assess sexuality and send targeted ads

unfortunately if it's possible, it will be done, and may lead to negative outcomes

 

[Veho]

phrenology is horse wash obviously, but it's silly to act like there aren't any anatomic differences which might be related to a variety of different hormones or other factors that might affect one's personality

Maybe.
But we're getting back to sampling bias, and what exactly the AI was looking at (and for) in training and in testing.

First of all the pictures were taken from a dating site, so they're filtered and photoshopped to hell and back (or at least heavily made up and contoured) (to hell and back) according to some perceived ideal of beauty that the intended sex will find attractive. Instagram face and so on. Does it really represent that person's facial features? You be the judge.

The sample is bogus.

Second:

The research found that gay men and women tended to have “gender-atypical” expressions and “grooming styles”

There are "gender typical expressions"? I think I understand the "gender-typical grooming styles"; women tend to wash. But "expressions"? Being very generous I'll say that means men tend to scowl and women tend to smile in photos on dating sites. Fellas, smiling is gay. Take note.

So the AI was trained on a sample with a significant slant, and then tested on photos from that same sample with the same slant, but I'm wondering how they accounted for the bias, and how the test would work on randos in the street.

And I wonder if the results can be replicated by someone else and with a different training set.

https://en.wikipedia.org/wiki/Replication_crisis

Methodology pls.

 

[impeeza]

 

[Veho]

Animu is gay confirmed.

 

[Spring_Spring]

sexuality is theorized to be linked to in utero hormone exposure, which would obviously alter more than brain anatomy - which might explain the differences in face shape

phrenology is horse wash obviously, but it's silly to act like there aren't any anatomic differences which might be related to a variety of different hormones or other factors that might affect one's personality

It's important to note although there was a high level of accuracy, still about 1/10 times the algorithm was not able to properly predict sexual orientation

There are a ton of random things like this tho - like for instance there is a small but statistically significant relationship between nose length and IQ



Well imagine a country like Iran got ahold of this (or another state in the ME) and started using it to harass people - or ad agencies start using facial recognition to assess sexuality and send targeted ads

unfortunately if it's possible, it will be done, and may lead to negative outcomes

What kind of "targeted advertising" could that possibly be? And who really pays attention to targeted advertising in the first place? Also, do advertisers just have free access to your camera? I thought that required an explicit permission, unless you believe that is all a ruse.

 

[impeeza]

But sometimes that is the problem of deny visual evident truths is a complex matter.

Is like when you point to sky and tell everyone "my sky is green" some people will believe you and modify their personal truth to accommodate your change on reality even if their senses tell them the sky is blue. Others will deny your reality and others will give a rat ass. That is how the humanity works.

I love the movie "Don't Look Up" even if is Leonardo on it, because is a reflection of the actual way of relative truths

 

[x65943]

What kind of "targeted advertising" could that possibly be? And who really pays attention to targeted advertising in the first place? Also, do advertisers just have free access to your camera? I thought that required an explicit permission, unless you believe that is all a ruse.

For instance some stores already use AI to catch shoplifters when they enter stores using the in house security cameras. They could identify a customer, make a generalization of their sexuality and then if it's a store where they have access to your address or phone number etc they could target advertising (like for instance a grocery store that someone signed up for a discount card for that has the customer's address)

As for how Iran could do it I think that's obvious

Maybe.
But we're getting back to sampling bias, and what exactly the AI was looking at (and for) in training and in testing.

First of all the pictures were taken from a dating site, so they're filtered and photoshopped to hell and back (or at least heavily made up and contoured) (to hell and back) according to some perceived ideal of beauty that the intended sex will find attractive. Instagram face and so on. Does it really represent that person's facial features? You be the judge.

View attachment 404189

The sample is bogus.

Second:



There are "gender typical expressions"? I think I understand the "gender-typical grooming styles"; women tend to wash. But "expressions"? Being very generous I'll say that means men tend to scowl and women tend to smile in photos on dating sites. Fellas, smiling is gay. Take note.

So the AI was trained on a sample with a significant slant, and then tested on photos from that same sample with the same slant, but I'm wondering how they accounted for the bias, and how the test would work on randos in the street.

And I wonder if the results can be replicated by someone else and with a different training set.

https://en.wikipedia.org/wiki/Replication_crisis

Methodology pls.

I think there is likely more to it than we are giving credit given it was >90% accurate

The fact brainwaves could lead to identification also at >80% tells us there is clearly something there

 

[Spring_Spring]

For instance some stores already use AI to catch shoplifters when they enter stores using the in house security cameras. They could identify a customer, make a generalization of their sexuality and then if it's a store where they have access to your address or phone number etc they could target advertising (like for instance a grocery store that someone signed up for a discount card for that has the customer's address)

As for how Iran could do it I think that's obvious

I think there is likely more to it than we are giving credit given it was >90% accurate

The fact brainwaves could lead to identification also at >80% tells us there is clearly something there

Does targeted advertising even work?
Iran is indeed pretty obvious.

 

[x65943]

Does targeted advertising even work?
Iran is indeed pretty obvious.

I'm no expert on targeted advertising but the millions companies already spend on it daily online (on YouTube, Facebook etc) tells me smarter people than us have probably concluded it does

 

[DinohScene]

I'd be pretty disappointed if the AI would label me straight, ngl.

 

[Spring_Spring]

I'm no expert on targeted advertising but the millions companies already spend on it daily online (on YouTube, Facebook etc) tells me smarter people than us have probably concluded it does

It has never worked for me, so I have my doubts. And I doubt my experience with it is extremely unique.

Post automatically merged: Nov 14, 2023

Anyway, just be ace and you won't have to worry about a thing!

Post automatically merged: Nov 14, 2023

Of course its not really a choice, but I digress.

 

[CoolMe]

Who cares, it's no one's business if you're gae, straight or bent etc. Why can't we just move on from this gender racism already, to more pressing matters? Smh

 

[Jayro]

Just looks like another Chinese database face collector to me.

 

[Veho]

I think there is likely more to it than we are giving credit given it was >90% accurate

Maybe.

I went and read the paper (because I couldn't coax a tl;dr out of you ) and:

first of all, the paper has a list of caveats longer than Ea Nasir's copper outlet.

b) the training sample is skewed, the test sample as well (they took pictures from a dating website, used 80% to train the AI and the remaining 20% to test it... my previous comments apply).

Four, I'm not sure how much of the "facial morphology" traits are genuine and how many literally due to head tilt:

Averages show that smiling is feminine and squinting is masculine I don't make the rules.


d) Quoting the study itself:

It is possible, however, that facial images posted on a dating website are particularly revealing of sexual orientation. Perhaps the users selected the photos that their desired partners might find the most appealing.

Perhaps.


Anyway. I will not discount the study out of hand but the results are tentative at best, and it will bring more harm than good if the idea that "science has proved" that "small jaw = gay" and so on takes root in some people's heads.

 

[AncientBoi]

unfortunately if it's possible, it will be done, and may lead to negative outcomes

It already has. Your posting that proves this. duh