7 ms·

New AI can guess a person's sexual orientation from a photograph

41 points by kawera 9y ago

YeGoblynQueenne 9y ago
>> The research found that gay men and women tended to have “gender-atypical” features, expressions and “grooming styles”, essentially meaning gay men appeared more feminine and vice versa. The data also identified certain trends, including that gay men had narrower jaws, longer noses and larger foreheads than straight men, and that gay women had larger jaws and smaller foreheads compared to straight women. Besides the fact that this sounds deeply offensive for gay people, it also sounds terribly, awfully unlikely. In fact, it reminds of the recent "phrenology" paper, where deep nets were used to, allegedly, "identify criminals from their faces": https://www.technologyreview.com/s/602955/neural-network-learns-to-identify-criminals-by-their-faces/ https://www.technologyreview.com/s/602955/neural-network-lea... Nowadays we know that sexual orientation is something that is neither set in stone, nor absolute (people are known to change throughout their lives, as well as have occasional partners of the "other" sex, whichever is opposite to their normal orientation). So the very idea of an accurate "gaydar" is ludicrous in and of itself. Finally, I wonder why anyone with an ounce of brains would seriously consider training a machine learning algorithm to recognise peoples' orientation. What morally justifiable practical purpose could there possibly be?
- kowdermeister 9y ago
  > Besides the fact that this sounds deeply offensive for gay people, it also sounds terribly, awfully unlikely. Offensive or not, sexual orientation is DNA based and so does your facial structure. What do you make of the claim that based on 5 photos they achieved a 90% success rate? That's significant.
  - DanBC 9y ago
    > What do you make of the claim that based on 5 photos they achieved a 90% success rate? I'm reminded of Gerd Gigerenzer's book "Reckoning with risk". It's about medical testing, so forgive the language used: > A condition exists. There is a test for that condition. The test is good, but not perfect. If someone has the condition there is a 90% chance they'll return positive. If someone does not have the condition there is a 1% chance they'll return positive. About 1% of the population have the condition. Bob has the test, and it comes back positive. What's the probability Bob has the condition? Most people, even doctors who give these types of tests, get this wrong. The answer is about 50%, but most people put it much higher at 99% or 90%. I think the Guardian reporting is irresponsible because the general public do not understand percentages, and any reporter using percentages is misleading (albeit inadvertantly) the public. Here's a link to the book: https://www.amazon.co.uk/Reckoning-Risk-Learning-Live-Uncertainty/dp/0140297863/ https://www.amazon.co.uk/Reckoning-Risk-Learning-Live-Uncert...
  - junto 9y ago
    > Offensive or not, sexual orientation is DNA based Sorry, but do you have a source for that claim?
    - always_learning 9y ago
      There is no source, the gender spectrum is fundamentally based on rejecting science
      - hankhill 9y ago
        Sooooo your source is gamergate/redpill bullshit propaganda. Okay. Have fun with that.
      - junto 9y ago
        I think you're confusing biological sex and sexual orientation. The poster above stated that sexual orientation is based off your DNA. There is no gene that determines whether you heterosexual or gay. It has nothing to do with gender. You're just trying to proliferate your opinion that gender = biological sex. Anyone that is transgender would disagree with your opinion.
    - kowdermeister 9y ago
      Yes, about 114 sources on Wikipedia. "As of 1999, about 500 species, ranging from primates to gut worms, have been documented engaging in same-sex behaviors.[3][page needed][4] According to the organizers of the 2006 Against Nature? exhibit, it has been observed in 1,500 species.[5]" So if not DNA, then what else? https://en.wikipedia.org/wiki/Homosexual_behavior_in_animals https://en.wikipedia.org/wiki/Homosexual_behavior_in_animals
      - foldr 9y ago
        That's missing the point. The question was whether there's any evidence that homosexuality (in humans) is the result of a particular genetic specification. We don't know whether or not that's the case for the worms. All we know is that some worms exhibit homosexual behavior.
  - YeGoblynQueenne 9y ago
    >> What do you make of the claim that based on 5 photos they achieved a 90% success rate? I'm assuming their model is badly trained and overfitting or they're misinterpreting (though I hope not misrepresenting) their results.
- Houshalter 9y ago
  How is this any more offensive or unlikely than stuff like digit ratio research? The idea that sexual orientation correlates with other features is very old, and this paper would seem to prove it. We've known for awhile that identical twins were much more likely than chance to share sexual orientations. https://en.wikipedia.org/wiki/Biology_and_sexual_orientation https://en.wikipedia.org/wiki/Biology_and_sexual_orientation >Nowadays we know that sexual orientation is something that is neither set in stone, nor absolute I don't think that's really correct. Sure there are exceptions, but most people stick to one sexual orientation, or at least somewhere in between on a spectrum. Attempts at converting gay people to become straight were unsuccessful. In fact this was one of the biggest arguments for gay rights, that it isn't a choice and can't be changed. >What morally justifiable practical purpose could there possibly be? This proves a biological origin for sexuality. Which is both scientifically interesting, and a strong argument for gay rights in places where it is still seen as some kind of sin or mental illness. Anyone hypothetically using this tool to discriminate against homosexuals would have to admit the tool works of course. And that would mean homosexuality is biological, which would contradict their own beliefs. Lastly do you have any actual argument against the papers methodology? Or the phrenology paper you link for that matter. Your comment is basically "I don't like the conclusions so I won't accept them". That's not terribly scientific.
  - nostromo123 9y ago
    > Your comment is basically "I don't like the conclusions so I won't accept them". That's not terribly scientific. The whole "gender is a social construct" movement is based on rejecting actual scientific evidence, so don't be surprised...
    - yorwba 9y ago
      The distinction between sex as biological and gender as social seems useful to me. It makes "gender is a social construct" a tautology, but doesn't in itself reject scientific evidence. It's just that inasmuch you have evidence that some behavior is biologically determined, the behavior is less amenable to analysis in terms of only gender.
  - 9y ago
- mobiletelephone 9y ago
  I also find it ethically troubling. The commercial application is probably advertising.
- nostromo123 9y ago
  > Nowadays we know that sexual orientation is something that is neither set in stone, nor absolute At best, the only thing we know is that this is true for some people. You say it yourself: "people are known" to change / have sexual partners of both genders -- because such people are an exception! These people are a minority, and you really can't generalise and affirm that this is the rule for the majority.
- hankhill 9y ago
  Sorry, you got downvoted to oblivion because HN now believes phrenology is a valid science. Also, good job downvoters for making yourself look like idiots.
musashizak 9y ago
And if one is bisex? Or in a moment of life have a particolar sexual orientati on?
- lithos 9y ago
  You'll probably change your grooming habits, hairstyle, and similar to be attractive to who you want to be. The bot didn't just go for fixed features.
Houshalter 9y ago
The image from the paper is very interesting and I'm surprised they didn't include it in the article: https://i.imgur.com/zs8RWIz.png https://i.imgur.com/zs8RWIz.png
- Aron 9y ago
  Those images appear (just going by my perception) to be dominated substantially by a single person, presumably the person with the most extreme score on the scale. The gay male even has the echo of glasses on, which almost certainly can't be common enough in presence and shape to be visible over a lot of images. Anyone offer an explanation?
  - Houshalter 9y ago
    Perhaps glasses are more likely on homosexuals for whatever reason, and so the images it classifies as "most likely to be gay" are much more likely to have glasses. Another interesting thing is that both the straight faces appear to be older and heavier set than the gay faces. I can't even imagine a possible explanation for that. I'd like to see the true composite of faces, rather than this one reweighted by their algorithm.
notahacker 9y ago
A big problem with trying to do draw conclusions about groups' typical physiological features from research using dating profiles is that there's a selection bias in which users will and won't post clear facial shots from neutral angles suited to processing by this algorithm (for a variety of reasons varying from perceived beauty norms of gay subcultures to social acceptance within their ethnic group) and that might well dominate any actual difference between the average physiology of gay and straight men.
Houshalter 9y ago
They claim 91% accuracy on a balanced dataset. E.g. where there the ratio of gays to straights is 50:50. To get a ratio of correct:incorrect of 91:9 on such a dataset, their test must increase or decrease the odds a person is gay by 10. Now in the general population, the ratio of gays to straights is about 16 to 984 (1.6%)†. So if their test gives someone a positive reading, that increases the odds to 162 to 984, or 14%. So you can't use this test to accurately guess someone's sexual orientation. Simply because gay people are so rare that even a few percent of straight people misclassified will overwhelm the number of actual gay people. But nonetheless it is much more accurate than humans or the base rate. And it is scientifically interesting that there are physical features that strongly correlate with sexual orientation. EDIT: another article claims this: >However, when asked to pick out the ten faces it was most confident about, nine of the chosen were in fact gay. If the goal is to pick a small number of people who are very likely to be gay out of a large group, the system appears able to do so. †These estimates vary by a lot and could be 3 times that, which would make the test much more accurate. Perhaps as high as 42% accuracy. And some people believe sexuality is more of a spectrum than a binary straight/gay. If so the test might be identifying straight individuals that lean more gay/bisexual than normal, and it's misclassifications wouldn't be so unreasonable. The test may also give varying degrees of confidence, and it may give much higher confidence to some people than others. The edit I made confirms that, and there are some individuals that it can tell are definitely gay or straight. But for most it is more uncertain.
- svantana 9y ago
  You're assuming every trial is of equal difficulty, which seems unlikely. I would guess that the system can classify some images with very good confidence, while others are more of a coin toss. But you're correct that the 'straight' prior is very strong.
- YeGoblynQueenne 9y ago
  Before you bother to look at all this, you have to ask: why would a person's orientation (or identity?) be evident in their pictures, or their face? This is the fundamental assumption the paper makes- and then they go about "proving" it by training an algorithm. That's all shades of biased. If you're starting from a premise that's very likely to be rooted in social prejudice, you're equally likely to get results that confirm the same, but are entirely wrong. There's no point in trying to reason rationally about work that's already so far past the gates of unreason. This is completely unnecessary, completely pointless research that is most probably flawed in ways that are not evident to someone simply reading the paper.
prestonpesek 9y ago
Sexuality is fluid, non-binary, with most of humanity probably some degree of bisexual. The categorization of one or the other is a cultural superimposition championed by the ancient stupidity of world religions. Let's try not to program our machines with these ludicrous biases please.
- TheRealDunkirk 9y ago
  > most of humanity probably some degree of bisexual Citation most definitely needed.
- dang 9y ago
  Would you please not do religious flamewar on HN? https://news.ycombinator.com/newsguidelines.html https://news.ycombinator.com/newsguidelines.html
fallingfrog 9y ago
I think they missed something. I read the paper, and partway through they have a set of merged images of all the gay and straight people of each sex. One thing that jumps out at you is that the gay dataset is on the whole composed of people who are at least 20 years younger than the straight dataset. They're not seeing a horizontal difference, they're looking at a generational difference- my guess is that younger people feel more open about homosexual feelings, while older people see those feelings as shameful, and that's mostly what they've identified. It's the tank problem all over again, they're not measuring what they think they're measuring.

41 points by kawera 9y ago

>> The research found that gay men and women tended to have “gender-atypical” features, expressions and “grooming styles”, essentially meaning gay men appeared more feminine and vice versa. The data also identified certain trends, including that gay men had narrower jaws, longer noses and larger foreheads than straight men, and that gay women had larger jaws and smaller foreheads compared to straight women. Besides the fact that this sounds deeply offensive for gay people, it also sounds terribly, awfully unlikely. In fact, it reminds of the recent "phrenology" paper, where deep nets were used to, allegedly, "identify criminals from their faces": https://www.technologyreview.com/s/602955/neural-network-learns-to-identify-criminals-by-their-faces/ https://www.technologyreview.com/s/602955/neural-network-lea... Nowadays we know that sexual orientation is something that is neither set in stone, nor absolute (people are known to change throughout their lives, as well as have occasional partners of the "other" sex, whichever is opposite to their normal orientation). So the very idea of an accurate "gaydar" is ludicrous in and of itself. Finally, I wonder why anyone with an ounce of brains would seriously consider training a machine learning algorithm to recognise peoples' orientation. What morally justifiable practical purpose could there possibly be?

kowdermeister 9y ago
> Besides the fact that this sounds deeply offensive for gay people, it also sounds terribly, awfully unlikely. Offensive or not, sexual orientation is DNA based and so does your facial structure. What do you make of the claim that based on 5 photos they achieved a 90% success rate? That's significant.
- DanBC 9y ago
  > What do you make of the claim that based on 5 photos they achieved a 90% success rate? I'm reminded of Gerd Gigerenzer's book "Reckoning with risk". It's about medical testing, so forgive the language used: > A condition exists. There is a test for that condition. The test is good, but not perfect. If someone has the condition there is a 90% chance they'll return positive. If someone does not have the condition there is a 1% chance they'll return positive. About 1% of the population have the condition. Bob has the test, and it comes back positive. What's the probability Bob has the condition? Most people, even doctors who give these types of tests, get this wrong. The answer is about 50%, but most people put it much higher at 99% or 90%. I think the Guardian reporting is irresponsible because the general public do not understand percentages, and any reporter using percentages is misleading (albeit inadvertantly) the public. Here's a link to the book: https://www.amazon.co.uk/Reckoning-Risk-Learning-Live-Uncertainty/dp/0140297863/ https://www.amazon.co.uk/Reckoning-Risk-Learning-Live-Uncert...
- junto 9y ago
  > Offensive or not, sexual orientation is DNA based Sorry, but do you have a source for that claim?
  - always_learning 9y ago
    There is no source, the gender spectrum is fundamentally based on rejecting science
    - hankhill 9y ago
      Sooooo your source is gamergate/redpill bullshit propaganda. Okay. Have fun with that.
    - junto 9y ago
      I think you're confusing biological sex and sexual orientation. The poster above stated that sexual orientation is based off your DNA. There is no gene that determines whether you heterosexual or gay. It has nothing to do with gender. You're just trying to proliferate your opinion that gender = biological sex. Anyone that is transgender would disagree with your opinion.
  - kowdermeister 9y ago
    Yes, about 114 sources on Wikipedia. "As of 1999, about 500 species, ranging from primates to gut worms, have been documented engaging in same-sex behaviors.[3][page needed][4] According to the organizers of the 2006 Against Nature? exhibit, it has been observed in 1,500 species.[5]" So if not DNA, then what else? https://en.wikipedia.org/wiki/Homosexual_behavior_in_animals https://en.wikipedia.org/wiki/Homosexual_behavior_in_animals
    - foldr 9y ago
      That's missing the point. The question was whether there's any evidence that homosexuality (in humans) is the result of a particular genetic specification. We don't know whether or not that's the case for the worms. All we know is that some worms exhibit homosexual behavior.
- YeGoblynQueenne 9y ago
  >> What do you make of the claim that based on 5 photos they achieved a 90% success rate? I'm assuming their model is badly trained and overfitting or they're misinterpreting (though I hope not misrepresenting) their results.
Houshalter 9y ago
How is this any more offensive or unlikely than stuff like digit ratio research? The idea that sexual orientation correlates with other features is very old, and this paper would seem to prove it. We've known for awhile that identical twins were much more likely than chance to share sexual orientations. https://en.wikipedia.org/wiki/Biology_and_sexual_orientation https://en.wikipedia.org/wiki/Biology_and_sexual_orientation >Nowadays we know that sexual orientation is something that is neither set in stone, nor absolute I don't think that's really correct. Sure there are exceptions, but most people stick to one sexual orientation, or at least somewhere in between on a spectrum. Attempts at converting gay people to become straight were unsuccessful. In fact this was one of the biggest arguments for gay rights, that it isn't a choice and can't be changed. >What morally justifiable practical purpose could there possibly be? This proves a biological origin for sexuality. Which is both scientifically interesting, and a strong argument for gay rights in places where it is still seen as some kind of sin or mental illness. Anyone hypothetically using this tool to discriminate against homosexuals would have to admit the tool works of course. And that would mean homosexuality is biological, which would contradict their own beliefs. Lastly do you have any actual argument against the papers methodology? Or the phrenology paper you link for that matter. Your comment is basically "I don't like the conclusions so I won't accept them". That's not terribly scientific.
- nostromo123 9y ago
  > Your comment is basically "I don't like the conclusions so I won't accept them". That's not terribly scientific. The whole "gender is a social construct" movement is based on rejecting actual scientific evidence, so don't be surprised...
  - yorwba 9y ago
    The distinction between sex as biological and gender as social seems useful to me. It makes "gender is a social construct" a tautology, but doesn't in itself reject scientific evidence. It's just that inasmuch you have evidence that some behavior is biologically determined, the behavior is less amenable to analysis in terms of only gender.
- 9y ago
mobiletelephone 9y ago
I also find it ethically troubling. The commercial application is probably advertising.
nostromo123 9y ago
> Nowadays we know that sexual orientation is something that is neither set in stone, nor absolute At best, the only thing we know is that this is true for some people. You say it yourself: "people are known" to change / have sexual partners of both genders -- because such people are an exception! These people are a minority, and you really can't generalise and affirm that this is the rule for the majority.
hankhill 9y ago
Sorry, you got downvoted to oblivion because HN now believes phrenology is a valid science. Also, good job downvoters for making yourself look like idiots.

musashizak 9y ago

And if one is bisex? Or in a moment of life have a particolar sexual orientati on?

lithos 9y ago
You'll probably change your grooming habits, hairstyle, and similar to be attractive to who you want to be. The bot didn't just go for fixed features.

Houshalter 9y ago

The image from the paper is very interesting and I'm surprised they didn't include it in the article: https://i.imgur.com/zs8RWIz.png https://i.imgur.com/zs8RWIz.png

Aron 9y ago
Those images appear (just going by my perception) to be dominated substantially by a single person, presumably the person with the most extreme score on the scale. The gay male even has the echo of glasses on, which almost certainly can't be common enough in presence and shape to be visible over a lot of images. Anyone offer an explanation?
- Houshalter 9y ago
  Perhaps glasses are more likely on homosexuals for whatever reason, and so the images it classifies as "most likely to be gay" are much more likely to have glasses. Another interesting thing is that both the straight faces appear to be older and heavier set than the gay faces. I can't even imagine a possible explanation for that. I'd like to see the true composite of faces, rather than this one reweighted by their algorithm.

notahacker 9y ago

A big problem with trying to do draw conclusions about groups' typical physiological features from research using dating profiles is that there's a selection bias in which users will and won't post clear facial shots from neutral angles suited to processing by this algorithm (for a variety of reasons varying from perceived beauty norms of gay subcultures to social acceptance within their ethnic group) and that might well dominate any actual difference between the average physiology of gay and straight men.

Houshalter 9y ago

They claim 91% accuracy on a balanced dataset. E.g. where there the ratio of gays to straights is 50:50. To get a ratio of correct:incorrect of 91:9 on such a dataset, their test must increase or decrease the odds a person is gay by 10. Now in the general population, the ratio of gays to straights is about 16 to 984 (1.6%)†. So if their test gives someone a positive reading, that increases the odds to 162 to 984, or 14%. So you can't use this test to accurately guess someone's sexual orientation. Simply because gay people are so rare that even a few percent of straight people misclassified will overwhelm the number of actual gay people. But nonetheless it is much more accurate than humans or the base rate. And it is scientifically interesting that there are physical features that strongly correlate with sexual orientation. EDIT: another article claims this: >However, when asked to pick out the ten faces it was most confident about, nine of the chosen were in fact gay. If the goal is to pick a small number of people who are very likely to be gay out of a large group, the system appears able to do so. †These estimates vary by a lot and could be 3 times that, which would make the test much more accurate. Perhaps as high as 42% accuracy. And some people believe sexuality is more of a spectrum than a binary straight/gay. If so the test might be identifying straight individuals that lean more gay/bisexual than normal, and it's misclassifications wouldn't be so unreasonable. The test may also give varying degrees of confidence, and it may give much higher confidence to some people than others. The edit I made confirms that, and there are some individuals that it can tell are definitely gay or straight. But for most it is more uncertain.

svantana 9y ago
You're assuming every trial is of equal difficulty, which seems unlikely. I would guess that the system can classify some images with very good confidence, while others are more of a coin toss. But you're correct that the 'straight' prior is very strong.
YeGoblynQueenne 9y ago
Before you bother to look at all this, you have to ask: why would a person's orientation (or identity?) be evident in their pictures, or their face? This is the fundamental assumption the paper makes- and then they go about "proving" it by training an algorithm. That's all shades of biased. If you're starting from a premise that's very likely to be rooted in social prejudice, you're equally likely to get results that confirm the same, but are entirely wrong. There's no point in trying to reason rationally about work that's already so far past the gates of unreason. This is completely unnecessary, completely pointless research that is most probably flawed in ways that are not evident to someone simply reading the paper.

prestonpesek 9y ago

Sexuality is fluid, non-binary, with most of humanity probably some degree of bisexual. The categorization of one or the other is a cultural superimposition championed by the ancient stupidity of world religions. Let's try not to program our machines with these ludicrous biases please.

TheRealDunkirk 9y ago
> most of humanity probably some degree of bisexual Citation most definitely needed.
dang 9y ago
Would you please not do religious flamewar on HN? https://news.ycombinator.com/newsguidelines.html https://news.ycombinator.com/newsguidelines.html

fallingfrog 9y ago

I think they missed something. I read the paper, and partway through they have a set of merged images of all the gay and straight people of each sex. One thing that jumps out at you is that the gay dataset is on the whole composed of people who are at least 20 years younger than the straight dataset. They're not seeing a horizontal difference, they're looking at a generational difference- my guess is that younger people feel more open about homosexual feelings, while older people see those feelings as shameful, and that's mostly what they've identified. It's the tank problem all over again, they're not measuring what they think they're measuring.