-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
New AI can guess whether you're gay or straight from a photograph, 91% of the time
- Thread starter ghostofsparta
- Start date
Finale Fireworker
Member
SCIENTIST 1: We have done it. This machine can detect homosexuality with 91% accuracy.
SCIENTIST 2: It has been a pleasure working with you. This has truly been a dream come true.
SCIENTIST 1: I am almost sad that it's over.
(Suddenly, the machine activates.)
MACHINE: t w o t a r g e t s i d e n t i f i e d. s c a n n i n g......
SCIENTIST 2: Oh no!
SCIENTIST 1: Turn it off!
MACHINE: s c a n c o m p l e t e. r e s u l t:
t a r g e t s s h o w a t t r i b u t e s o f h o m o s e x u a l i t y w i t h i n a 9 % m a r g i n o f e r r o r
(The scientists are stunned, but also, in love.)
SCIENTIST 1: Did... did you know?
SCIENTIST 2: No... I mean, maybe. But I thought I just loved science.
SCIENTIST 1: It seems the answer we were looking for was right in front of us.
SCIENTIST 2: All along...
SCIENTIST 1: Perhaps our experiment is just beginning.
SCIENTIST 2: I am eager to work with you again in this pursuit.
(They hold hands.)
(The robot beeps, knowingly.)
SCIENTIST 2: It has been a pleasure working with you. This has truly been a dream come true.
SCIENTIST 1: I am almost sad that it's over.
(Suddenly, the machine activates.)
MACHINE: t w o t a r g e t s i d e n t i f i e d. s c a n n i n g......
SCIENTIST 2: Oh no!
SCIENTIST 1: Turn it off!
MACHINE: s c a n c o m p l e t e. r e s u l t:
t a r g e t s s h o w a t t r i b u t e s o f h o m o s e x u a l i t y w i t h i n a 9 % m a r g i n o f e r r o r
(The scientists are stunned, but also, in love.)
SCIENTIST 1: Did... did you know?
SCIENTIST 2: No... I mean, maybe. But I thought I just loved science.
SCIENTIST 1: It seems the answer we were looking for was right in front of us.
SCIENTIST 2: All along...
SCIENTIST 1: Perhaps our experiment is just beginning.
SCIENTIST 2: I am eager to work with you again in this pursuit.
(They hold hands.)
(The robot beeps, knowingly.)
TheRagnCajun
Member
If you select a sample from the population where 50% of them are gay then you can test your computer's accuracy relative to random guessing which would have a 50% success rate. That seems like the only way to do it.
HeisenbergLannister
Member
Oh man, a lot of politicians "against homosexuality" are quickly trying to delete every photograph of themselves that has ever existed.
Ether_Snake
å®å®å®å®å®å®å®å®å®å®å®å®å®å®å®
It only comes off as that as a result of the comments posted here.
Enter The 36 Chambers
Member
I give it 2 years to before this becomes an app, and an inevitable snapchat filter
Lotta dudes finna be outed
Lotta dudes finna be outed
PdotMichael
Banned
Not a deep learning expert, but that appears to be quite the subpar base for such an algorithm.
Also absurd how angry people can get over such trivial application of deep learning.
Icyflamez96
Member
I mean, that's on you then no?
If you're concerned about gaydar you could probably trick the system by wearing makeup in a certain pattern.
https://medium.com/@ageitgey/machin...entionally-trick-neural-networks-b55da32b7196
https://medium.com/@ageitgey/machin...entionally-trick-neural-networks-b55da32b7196
Iva Demilcol
Member
So, the real question here is where can I start to upload pictures of my friends?
okay, that was a bad joke,I don't have friends in real life 
boiled goose
good with gravy
So many large foreheads now explained...
Interesting research
Interesting research
You know what. This literally proves that being "gay" isn't a "choice". The physical body shows "traits" we may not even notice...and I find that pretty awesome.
Rather than this PROVING that to horrible human beings they would use it against people I suppose....derp.
Like it said, it's not 100%...but I find the things the AI picks up on and uses facinating (and pretty obvious).
90+% is darn impressive.
Rather than this PROVING that to horrible human beings they would use it against people I suppose....derp.
Like it said, it's not 100%...but I find the things the AI picks up on and uses facinating (and pretty obvious).
90+% is darn impressive.
They don't always do, but as a bisexual man, I'm not oblivious to gay trends. For example, there are hairstyles popular with gay men.
1.) It's not accuracy, it's AUC.
2.) The dataset is balanced, so a default classifier would only have an accuracy of 50% anyway.
hmmm
What if the 9% are just gay people in denial of their sexual orientation? AI could still be right...
Moose Biscuits
It would be extreamly painful...
Hasn't that been the stance for the Church for the longest time anyway?
Wildo09
Member
It's right there in the OP.
They want to find out if sexuality is correlated to the exposure to certain hormones
What if the AI is so smart it knows that data set is balanced ahead of time, because that would be the easiest way to have a proper sample, and is basically just guessing half gay half straight?
Yeah I know it wouldn't actually work, would still just come out at 50% or so.
What is interesting here, did the AI come up with the criteria for classifying or did the people who programmed the AI? Because then it's really just people deciding what features all anecdotally more likely to have people consider that person gay upon close observation, and then the AI is just good at discerning those features from photos. Essentially just doing the work more quickly and crunching the numbers with less hemming and hawing than people.
Baraka Obama
Banned
It doesn't matter to me but I would imagine a good amount of people wouldn't like to be classified anything by a machine.
Kinda pointless, but apparently it is somewhat of a look if the thing works with that high of an accuracy from a photograph. Personally, I'm kinda glad there might be some physiological signs to combat the idiots that think sexuality is a choice.
50% is the lowest you can go in 2 class classification. If you get less than 50% you can just tell the classifier to choose the opposite of what it was going to chose and get more than 50%.
Not sure specifically how features were handled in this case, since I just glanced over the study, but deep learning is generally capable of coming up with features without human help, which is one of its biggest strengths. There's also a lot of algorithms used in face recognition that do extract specific features (things like cheek shapes and nose shape) but even then you usually extract a whole bunch of features and then let the AI figure out which subset to use. You also definitely never explicitly state "this feature is more likely to be present in this specific class of images", because then that's not really AI anymore.
UncleSporky
Member
I'm guessing this primarily works because the people in the photos confirm whether they are gay or straight, and are not closeted or in denial in any way. A closeted person might dress or groom or carry themselves differently in an effort to hide it.
In other words I would guess this isn't a "gay detection machine" you can point at any photo to determine secrets, something that would make politicians or religious people nervous. They're likely not going to look gay to the AI, and will say "nope I'm not gay, see, it works." The people who look gay to the AI are the kind of people who would tell the scientists and researchers "yes, I am gay."
In other words I would guess this isn't a "gay detection machine" you can point at any photo to determine secrets, something that would make politicians or religious people nervous. They're likely not going to look gay to the AI, and will say "nope I'm not gay, see, it works." The people who look gay to the AI are the kind of people who would tell the scientists and researchers "yes, I am gay."
CampbellzSoup
Member
Oh great before you're hired now someone can your photo through this. This helps no one.
neverknowsbest
Member
Damn progressive GAF, when did you get so anti-science? The OP lists valid scientific reasons for doing this kind of research. It strengthens support for the claim that homosexuality is by birth, not choice. Plus, it's interesting to see that humans do so poorly compared to the AI.
Is it?
Very few gaffers even understand science so no surprise most people's opinions on here are shocking.
Oppo
Member
pretty good.
I was going to do a much darker version of this taking place in a surveillance control room in Saudi Arabia
They actually address this in the author's notes:
Like they say, it's obviously very hard to test on closeted people with a statistically significant sample size, but it's not unreasonable to assume it would work.
Oppo
Member
Nonetheless, let's run it on this random shot of a megachurch crowd. Just for funsies.
ChuckHoneywood
Banned
What's so shocking about it?
Awesome, thanks for the info. The article was really bad at describing the actual underlying stats work.
The Hermit
Member
It will save a lot of therapy... Or will increase.
I have no idea, but I know a lot of people that wouldn't pass this test
hendersonhank
Member
"The data also identified certain trends, including that gay men had narrower jaws, longer noses and larger foreheads than straight men"
It's not just looking at "gender expression" but actual physical traits.
Finale Fireworker
Member
I try to give my queer fiction happy endings.
So, the real question here is where can I start to upload pictures of my friends?
okay, that was a bad joke,I don't have friends in real life
I'll be your friend :3
Most gaffers' relationship with science can be best summed up in this gif
Stumpokapow
listen to the mad man
Quick followup -- I had missed the preprint link initially when I commented so I was going off the coverage and the abstract. Thanks for the followup, t does sound like the inference is closer to correct (and I'm glad they include a discussion of how thresholding impacts the tradeoff between precision/recall, which is a basic thing you'd think every ML paper does). I'd have to think a little more about how the design of the paper as a balanced sample would impact true OOS testing -- for example, it's not clear the AI would guessing everyone is straight in a design without a balanced sample.
I also maintain their dismissal of the sampling concerns as ungrounded. This is an AI that can tell the difference between dating site photos of gay and straight persons, not between photos of gay and straight persons. Even granted that they've identified morphological features as being predictive, I would like to see true OOS with, say, randomly harvested non-dating photos.
Not even close
Look at another way it does prove that it isn't just a choice.
Though I'm worried what would happen if less desirable people got it
How is that depressing? Isn't it liberating?
It had been argued for decades that sexual orientation has biological origins and is not a "life choice". This data certainly supports such arguments.
I also read research that showed that homosexual men consistently have a more nuanced and refined sense of smell (like women) than heterosexual men. That data also supports the same argument.
People cannot control their genetics and so it makes the case that homosexuality is not a choice and also that people should not be penalized or disadvantaged in life for that which they didn't choose (such as sex, race, physical impairments, etc)
This appears to be just one of many types of genetic traits that deserves equal treatment to that of the majority trait (heterosexuality).