5 ms·

A.I ‘Gaydar’ Could Be the Start of Something Much Worse

38 points by sus_007 9y ago

amiga-workbench 9y ago
I don't see where the panic is coming from, I'm almost certain Google, Facebook and several advertising agencies know I'm gay from my browsing history, and all that data is already made available to governments.
- QAPereo 9y ago
  (And to be clear, based on this work alone, AI can’t tell whether someone is gay or straight from a photo. The panic is being manufactured, probably for little more than grant money, recognition, and clicks.
  - tspike 9y ago
    It's amazing how short our memories are.
- chaostheory 9y ago
  > warns that facial recognition will soon be able to identify not only someone’s sexual orientation, but their political views, criminality, and even their IQ. With statements like these, some worry we’re reviving an old belief with a bad history: that you can intuit character from appearance. This pseudoscience, physiognomy, was fuel for the scientific racism of the 19th and 20th centuries, and gave moral cover to some of humanity’s worst impulses: to demonize, condemn, and exterminate fellow humans.
  - wyager 9y ago
    > With statements like these, some worry we’re reviving an old belief with a bad history: that you can intuit character from appearance. This pseudoscience, physiognomy If it actually works, it's not pseudoscience, it's just science. (Assuming you don't understate the variance in generated intelligence/sexuality/whatever distributions.)
- tgb 9y ago
  "You're not really gay, you're just acting gay for attention - look the AI says you're not gay!" "I can't believe you're still seeing John! You know he's gay, right? I tested his profile picture - it came back 98%!" The whole point of the LGBTQ movement was to let people decide for themselves what they are, not to label them based off what society says they ought to be.
  - 21 9y ago
    I'm male. Next time at the gym showers I'll identify as a female, for about an hour. No problem, right.
    - ehnto 9y ago
      Probably not a helpful tangent of the debate.
      - 21 9y ago
        This will become more and more of an issue if we allow anyone to identify at any time as anything. There was an article recently about males which identify as females wanting in on female-only music festivals (which were a response to frequent sexual assaults at mixed festivals)
        ehnto 9y ago
        I understand, but it's a complicated debate and it's full of strong opinions and subjectivity. Definitely not one we can solve with a one liner about gym showers, and I think a tangent based on that is unlikely to be anything but antagonistic.
    - Toboe 9y ago
      Ok, assessments from 2 independent psychiatrists and a history of passing please. Then no problem.
  - wyager 9y ago
    Just because people might choose to misinterpret the results of a test doesn't mean the test is invalid. People already do an acceptable job understanding the existence of false positives and negatives on other tests.
- protomyth 9y ago
  For some countries (e.g. Saudi Arabia, Iran), the government knowing might not be very healthy for you.
basicplus2 9y ago
<And to be clear, based on this work alone, AI can’t tell whether someone is gay or straight from a photo.> What can i say?
dogruck 9y ago
It's illogical that the Verge article takes it as fact that it's impossible to infer someone's character from photos or videos. If that's true, then the only concerning outcome is that we build and operate software that falsely purports to provide insight. To me, the worrying outcome is that it is feasible. Before you know it, SAT tests are replaced by photos. Of course I can think of good applications, such as scanning for terrorists by identifying abnormal signs of stress.
- TheSpiceIsLife 9y ago
  Not all abnormal signs of stress are indicative of terrorist tendencies.
  - dogruck 9y ago
    Agreed. I think about that sometimes, for example, when I'm stressed at the airport just because I'm late or headed to an important business meeting. That said, I can imagine a AI-driven screening mechanism that was rooted in a causal relationship to data, instead of random nonsense.
bitL 9y ago
It's a major can of worms. If any of these classifiers work well on intelligence, orientation, character, capabilities, it might give scientific backing to some horrible practices like eugenics, abortions/extermination of undesirables etc. This is a truly horrifying scenario, an ultimate black pill and humanity will need to confront it soon it seems.
- adventured 9y ago
  Apparently we can accurately determine a child is a psychopath from an extremely young age now: http://www.nytimes.com/2012/05/13/magazine/can-you-call-a-9-year-old-a-psychopath.html http://www.nytimes.com/2012/05/13/magazine/can-you-call-a-9-... There's no question that the next 20-30 years will be overflowing with cans of worms. The next step after early (fetus or embryo stage) diagnosing of X thing, is that we're going to soon have the genetic alteration power to do something about it courtesy of CRISPR. It seems practically guaranteed that authoritarian societies will begin forcing parents to alter their unborn children to eliminate all sorts of attributes that are considered undesirable by whatever regime is in power.
  - stevenwoo 9y ago
    Would you believe 20 months old? It seems preposterous but this one girl was pretty antisocial as a toddler. https://www.theatlantic.com/magazine/archive/2017/06/when-your-child-is-a-psychopath/524502/ https://www.theatlantic.com/magazine/archive/2017/06/when-yo...
  - TheSpiceIsLife 9y ago
    It won't take an authoritarian regime to force people to alter their embryonic children. I'm certain people will be willingly lining up at the clinics.
    - pizza 9y ago
      Heh, I'm.. wary of that, I guess? The people who alter their embryonic children during the first 100 years of that technology's existence - I assume - will mostly get a raw deal until embryonic alteration as a field goes through a billion 'Anna Karenina principle' failures. We have the ability to adapt by simulating nature in our minds, and science too comes from this - but we are the result of so much time and such harsh filtering. In other words, I don't think we will see as much interest ten years into it after as we would during the first year.. we would only, in such little time, pick the low-hanging fruit changes. But these are also the ones nature has not propagated - which would be odd if they are both simplest to achieve (including my accidental mutation) and also known to be beneficial on average. I'd imagine hypothetically that the first wave of children would start exhibiting genetic illnesses will drive down the willingness to sign children up, out of simple precaution that parents are unlikely to have ever seen evidence that the procedure does not have a risk of ruining the child's life.
caconym_ 9y ago
I don't really understand the article's position. Comparing physiognomy to a computer program that can actually make accurate predictions about people at a statistically significant rate based on their facial features is like comparing alchemy to chemistry. They might be superficially similar, but one actually works and one doesn't. edit: Of course, I think this technology is horrifying and that a) it will never be 100% accurate, and b) it is just begging to be abused. I just don't think denying the reality of its existence will be a good way to fight against its misuse. Pandora's box has been opened.
- alva 9y ago
  The position of the article is one which will become extremely prevalent if this area of research gets results. If certain behavioural indicators can be identified reliably with this method, it will be the atomic bomb of political-incorrectness.
  - candiodari 9y ago
    It could be even worse if it becomes self-reinforcing. One could see this being developed in a way that "allows HR to make better decisions", and let's say it predicts purple people with yellow spots have a tendency to steal. Which of course will lead to exactly what it predicts. And we all know HR departments would gladly use stuff like this. Especially for the jobs where it would matter most: underpaid entry-level jobs that a lot of people get as either a last resort or a first job.
bitL 9y ago
More coming from China: "Automated Inference on Criminality using Face Images" https://arxiv.org/pdf/1611.04135v2.pdf https://arxiv.org/pdf/1611.04135v2.pdf
- R_haterade 9y ago
  IIRC, this was widely discredited due to problems with the sampling procedure. Can't be arsed to dig up the criticisms I found. Confirmed some of my long-held biases about physiognomy though, so I didn't dismiss it outright.
tomrod 9y ago
Interesting -- why is the composite skin tone so light? I've heard ML algorithms can have a hard time with darker skin tones, though I'm still not sure of the reasons. Is the research here subject to these same issues?
- zitterbewegung 9y ago
  It could be that their sample is biased to start with. According to one of the studies they took it from an online dating website.
- stevenwoo 9y ago
  Maybe related to the bias of film and digital cameras for caucasian skin tones? https://priceonomics.com/how-photography-was-optimized-for-white-skin/ https://priceonomics.com/how-photography-was-optimized-for-w...
cerealbad 9y ago
seems like some kind of face camouflage is necessary for the future.
- hacoo 9y ago
  Relevant paper (attacks facial recognition, but same concept): https://www.cs.cmu.edu/~sbhagava/papers/face-rec-ccs16.pdf https://www.cs.cmu.edu/~sbhagava/papers/face-rec-ccs16.pdf
zitterbewegung 9y ago
Can a model predict someone is Gay? Who knows. Can someone create a model that looks legitimate and then has false negatives and positives? Absolutely. Also, they will probably be able to sell it to countries. This reminds me of a movie called The Final Cut by Robin Williams. Its where everyone has an implant for recording everything and you get a face tattoo to bypass it. In this case you would get a face tattoo that would intentionally misclassify you as whatever orientation that you want.
- cool_shit 9y ago
  Then they would give their algorithms more power and memory, upgrade their cameras to capture more spectrum and/or resolution and learn who has used a tattoo to cover up their face.
Animats 9y ago
A system to identify concealed gun carriers from video should be possible. It's not hard if you can see someone step up or down; the inertial effects of carrying a big weight show.
- icelancer 9y ago
  There'd have to be "truth" gait data for the person. Gait is highly variable between subjects, and if the weapon is worn in a centered position (small of the back is a common concealment position), it's unlikely this will work. Holsters are not always worn in the lower extremities, either. I'm not sure this would be all that successful. It seems quite difficult.
0xbear 9y ago
As a human I can’t predict whether someone is gay or not most of the time. Most gay people I know don’t show any of the stereotypical “gay” cues. I don’t see how a machine would be able to determine something like this, so this story just seems like meaningless clickbait to me.
- stevenwoo 9y ago
  The test seems kind of bogus but I don't know anything about statistics or machine learning - they always processed two photos at a time and one was guaranteed to be self identified as gay and the other was always heterosexual. It seems like one would really need to have the test work in any combination of cases, not just the one they used where there was always one of each type of sexuality.
- icelancer 9y ago
  >>As a human I can’t predict whether someone is gay or not most of the time Why would your success as a human have anything to do with whether or not an algorithm could have success in any field?
  - 0xbear 9y ago
    Because it’s really freaking hard to beat humans on visual tasks.
    - barry-cotter 9y ago
      The fact that you can't do it doesn't mean humans can't do it.
      - 0xbear 9y ago
        This is modern day phrenology. It _cant_ work.
fosco 9y ago
I must say I remember reading Blink[0] by Malcolm gladwell which spoke of Sylvan Tomkins[1] looking at people's faces (and animals) and have extremely high success rates of describing their personalities to such a degree that by looking at the faces of a tribe in a rural area of Africa he was able to determine parts of their cultural rituálé and tendencies.[2] while i still find this far fetched, considering people are born a certain way it does make sense to me that like a fingerprint certain types of brain structures may have an impact on how the face looks. Please take a look at my second article that had malcolm gladwell discussing Suliban Tomkins, specifically towards the end of it. [0]https://en.m.wikipedia.org/wiki/Blink_(book) https://en.m.wikipedia.org/wiki/Blink_(book) [1]https://en.m.wikipedia.org/wiki/Silvan_Tomkins https://en.m.wikipedia.org/wiki/Silvan_Tomkins [2]http://gladwell.com/blink/the-mysteries-of-mind-reading/ http://gladwell.com/blink/the-mysteries-of-mind-reading/
omalleyt 9y ago
The funny thing is the author thinks having an 81% success rate of predicting which image out of a pair of images is of a gay person, is not as hard as predicting with 81% accuracy whether any given individual is gay or straight. Meaning the author really doesn't understand the concept of unbalanced classes

38 points by sus_007 9y ago

I don't see where the panic is coming from, I'm almost certain Google, Facebook and several advertising agencies know I'm gay from my browsing history, and all that data is already made available to governments.

QAPereo 9y ago
(And to be clear, based on this work alone, AI can’t tell whether someone is gay or straight from a photo. The panic is being manufactured, probably for little more than grant money, recognition, and clicks.
- tspike 9y ago
  It's amazing how short our memories are.
chaostheory 9y ago
> warns that facial recognition will soon be able to identify not only someone’s sexual orientation, but their political views, criminality, and even their IQ. With statements like these, some worry we’re reviving an old belief with a bad history: that you can intuit character from appearance. This pseudoscience, physiognomy, was fuel for the scientific racism of the 19th and 20th centuries, and gave moral cover to some of humanity’s worst impulses: to demonize, condemn, and exterminate fellow humans.
- wyager 9y ago
  > With statements like these, some worry we’re reviving an old belief with a bad history: that you can intuit character from appearance. This pseudoscience, physiognomy If it actually works, it's not pseudoscience, it's just science. (Assuming you don't understate the variance in generated intelligence/sexuality/whatever distributions.)
tgb 9y ago
"You're not really gay, you're just acting gay for attention - look the AI says you're not gay!" "I can't believe you're still seeing John! You know he's gay, right? I tested his profile picture - it came back 98%!" The whole point of the LGBTQ movement was to let people decide for themselves what they are, not to label them based off what society says they ought to be.
- 21 9y ago
  I'm male. Next time at the gym showers I'll identify as a female, for about an hour. No problem, right.
  - ehnto 9y ago
    Probably not a helpful tangent of the debate.
    - 21 9y ago
      This will become more and more of an issue if we allow anyone to identify at any time as anything. There was an article recently about males which identify as females wanting in on female-only music festivals (which were a response to frequent sexual assaults at mixed festivals)
      - ehnto 9y ago
        I understand, but it's a complicated debate and it's full of strong opinions and subjectivity. Definitely not one we can solve with a one liner about gym showers, and I think a tangent based on that is unlikely to be anything but antagonistic.
  - Toboe 9y ago
    Ok, assessments from 2 independent psychiatrists and a history of passing please. Then no problem.
- wyager 9y ago
  Just because people might choose to misinterpret the results of a test doesn't mean the test is invalid. People already do an acceptable job understanding the existence of false positives and negatives on other tests.
protomyth 9y ago
For some countries (e.g. Saudi Arabia, Iran), the government knowing might not be very healthy for you.

basicplus2 9y ago

<And to be clear, based on this work alone, AI can’t tell whether someone is gay or straight from a photo.> What can i say?

dogruck 9y ago

It's illogical that the Verge article takes it as fact that it's impossible to infer someone's character from photos or videos. If that's true, then the only concerning outcome is that we build and operate software that falsely purports to provide insight. To me, the worrying outcome is that it is feasible. Before you know it, SAT tests are replaced by photos. Of course I can think of good applications, such as scanning for terrorists by identifying abnormal signs of stress.

TheSpiceIsLife 9y ago
Not all abnormal signs of stress are indicative of terrorist tendencies.
- dogruck 9y ago
  Agreed. I think about that sometimes, for example, when I'm stressed at the airport just because I'm late or headed to an important business meeting. That said, I can imagine a AI-driven screening mechanism that was rooted in a causal relationship to data, instead of random nonsense.

bitL 9y ago

It's a major can of worms. If any of these classifiers work well on intelligence, orientation, character, capabilities, it might give scientific backing to some horrible practices like eugenics, abortions/extermination of undesirables etc. This is a truly horrifying scenario, an ultimate black pill and humanity will need to confront it soon it seems.

adventured 9y ago
Apparently we can accurately determine a child is a psychopath from an extremely young age now: http://www.nytimes.com/2012/05/13/magazine/can-you-call-a-9-year-old-a-psychopath.html http://www.nytimes.com/2012/05/13/magazine/can-you-call-a-9-... There's no question that the next 20-30 years will be overflowing with cans of worms. The next step after early (fetus or embryo stage) diagnosing of X thing, is that we're going to soon have the genetic alteration power to do something about it courtesy of CRISPR. It seems practically guaranteed that authoritarian societies will begin forcing parents to alter their unborn children to eliminate all sorts of attributes that are considered undesirable by whatever regime is in power.
- stevenwoo 9y ago
  Would you believe 20 months old? It seems preposterous but this one girl was pretty antisocial as a toddler. https://www.theatlantic.com/magazine/archive/2017/06/when-your-child-is-a-psychopath/524502/ https://www.theatlantic.com/magazine/archive/2017/06/when-yo...
- TheSpiceIsLife 9y ago
  It won't take an authoritarian regime to force people to alter their embryonic children. I'm certain people will be willingly lining up at the clinics.
  - pizza 9y ago
    Heh, I'm.. wary of that, I guess? The people who alter their embryonic children during the first 100 years of that technology's existence - I assume - will mostly get a raw deal until embryonic alteration as a field goes through a billion 'Anna Karenina principle' failures. We have the ability to adapt by simulating nature in our minds, and science too comes from this - but we are the result of so much time and such harsh filtering. In other words, I don't think we will see as much interest ten years into it after as we would during the first year.. we would only, in such little time, pick the low-hanging fruit changes. But these are also the ones nature has not propagated - which would be odd if they are both simplest to achieve (including my accidental mutation) and also known to be beneficial on average. I'd imagine hypothetically that the first wave of children would start exhibiting genetic illnesses will drive down the willingness to sign children up, out of simple precaution that parents are unlikely to have ever seen evidence that the procedure does not have a risk of ruining the child's life.

caconym_ 9y ago

I don't really understand the article's position. Comparing physiognomy to a computer program that can actually make accurate predictions about people at a statistically significant rate based on their facial features is like comparing alchemy to chemistry. They might be superficially similar, but one actually works and one doesn't. edit: Of course, I think this technology is horrifying and that a) it will never be 100% accurate, and b) it is just begging to be abused. I just don't think denying the reality of its existence will be a good way to fight against its misuse. Pandora's box has been opened.

alva 9y ago
The position of the article is one which will become extremely prevalent if this area of research gets results. If certain behavioural indicators can be identified reliably with this method, it will be the atomic bomb of political-incorrectness.
- candiodari 9y ago
  It could be even worse if it becomes self-reinforcing. One could see this being developed in a way that "allows HR to make better decisions", and let's say it predicts purple people with yellow spots have a tendency to steal. Which of course will lead to exactly what it predicts. And we all know HR departments would gladly use stuff like this. Especially for the jobs where it would matter most: underpaid entry-level jobs that a lot of people get as either a last resort or a first job.

bitL 9y ago

More coming from China: "Automated Inference on Criminality using Face Images" https://arxiv.org/pdf/1611.04135v2.pdf https://arxiv.org/pdf/1611.04135v2.pdf

R_haterade 9y ago
IIRC, this was widely discredited due to problems with the sampling procedure. Can't be arsed to dig up the criticisms I found. Confirmed some of my long-held biases about physiognomy though, so I didn't dismiss it outright.

tomrod 9y ago

Interesting -- why is the composite skin tone so light? I've heard ML algorithms can have a hard time with darker skin tones, though I'm still not sure of the reasons. Is the research here subject to these same issues?

zitterbewegung 9y ago
It could be that their sample is biased to start with. According to one of the studies they took it from an online dating website.
stevenwoo 9y ago
Maybe related to the bias of film and digital cameras for caucasian skin tones? https://priceonomics.com/how-photography-was-optimized-for-white-skin/ https://priceonomics.com/how-photography-was-optimized-for-w...

cerealbad 9y ago

seems like some kind of face camouflage is necessary for the future.

hacoo 9y ago
Relevant paper (attacks facial recognition, but same concept): https://www.cs.cmu.edu/~sbhagava/papers/face-rec-ccs16.pdf https://www.cs.cmu.edu/~sbhagava/papers/face-rec-ccs16.pdf

zitterbewegung 9y ago

Can a model predict someone is Gay? Who knows. Can someone create a model that looks legitimate and then has false negatives and positives? Absolutely. Also, they will probably be able to sell it to countries. This reminds me of a movie called The Final Cut by Robin Williams. Its where everyone has an implant for recording everything and you get a face tattoo to bypass it. In this case you would get a face tattoo that would intentionally misclassify you as whatever orientation that you want.

cool_shit 9y ago
Then they would give their algorithms more power and memory, upgrade their cameras to capture more spectrum and/or resolution and learn who has used a tattoo to cover up their face.

Animats 9y ago

A system to identify concealed gun carriers from video should be possible. It's not hard if you can see someone step up or down; the inertial effects of carrying a big weight show.

icelancer 9y ago
There'd have to be "truth" gait data for the person. Gait is highly variable between subjects, and if the weapon is worn in a centered position (small of the back is a common concealment position), it's unlikely this will work. Holsters are not always worn in the lower extremities, either. I'm not sure this would be all that successful. It seems quite difficult.

0xbear 9y ago

As a human I can’t predict whether someone is gay or not most of the time. Most gay people I know don’t show any of the stereotypical “gay” cues. I don’t see how a machine would be able to determine something like this, so this story just seems like meaningless clickbait to me.

stevenwoo 9y ago
The test seems kind of bogus but I don't know anything about statistics or machine learning - they always processed two photos at a time and one was guaranteed to be self identified as gay and the other was always heterosexual. It seems like one would really need to have the test work in any combination of cases, not just the one they used where there was always one of each type of sexuality.
icelancer 9y ago
>>As a human I can’t predict whether someone is gay or not most of the time Why would your success as a human have anything to do with whether or not an algorithm could have success in any field?
- 0xbear 9y ago
  Because it’s really freaking hard to beat humans on visual tasks.
  - barry-cotter 9y ago
    The fact that you can't do it doesn't mean humans can't do it.
    - 0xbear 9y ago
      This is modern day phrenology. It _cant_ work.

fosco 9y ago

I must say I remember reading Blink[0] by Malcolm gladwell which spoke of Sylvan Tomkins[1] looking at people's faces (and animals) and have extremely high success rates of describing their personalities to such a degree that by looking at the faces of a tribe in a rural area of Africa he was able to determine parts of their cultural rituálé and tendencies.[2] while i still find this far fetched, considering people are born a certain way it does make sense to me that like a fingerprint certain types of brain structures may have an impact on how the face looks. Please take a look at my second article that had malcolm gladwell discussing Suliban Tomkins, specifically towards the end of it. [0]https://en.m.wikipedia.org/wiki/Blink_(book) https://en.m.wikipedia.org/wiki/Blink_(book) [1]https://en.m.wikipedia.org/wiki/Silvan_Tomkins https://en.m.wikipedia.org/wiki/Silvan_Tomkins [2]http://gladwell.com/blink/the-mysteries-of-mind-reading/ http://gladwell.com/blink/the-mysteries-of-mind-reading/

omalleyt 9y ago

The funny thing is the author thinks having an 81% success rate of predicting which image out of a pair of images is of a gay person, is not as hard as predicting with 81% accuracy whether any given individual is gay or straight. Meaning the author really doesn't understand the concept of unbalanced classes