AI Advances Can Spot Sexual Orientation

monkeymagick

[H]News
Joined
Jun 22, 2008
Messages
480
The Guardian reports there is an algorithm that can guess a person's sexuality with higher accuracy than humans. Through the research, the AI determined "gender-atypical" features and expressions such as men appearing more "feminine" identified a person's sexual orientation. Data was obtained from a sample of more than 35,000 facial images of both sexes from US dating sites.

Human judges performed much worse than the algorithm, accurately identifying orientation only 61% of the time for men and 54% for women. When the software reviewed five images per person, it was even more successful - 91% of the time with men and 83% with women. Broadly, that means "faces contain much more information about sexual orientation than can be perceived and interpreted by the human brain", the authors wrote.
 
The joke's on you, AI Sexuality testing-robot. It's 2017.

We change sex based on how we feel that minute. Right now I'm feeling a little bit... Technosexual.

flash-mar2.jpg


AI Sexuality testing-robots need only to apply.
 
Last edited:
All that's left then is for the AI to start treating each differently.

Or wait...they're supposed to recognize sexual orientation but treat humans in a completely non-discriminatory manner. It then gets told they are being insensitive to a very meaningful aspect of a human being, and also told that sexual preference and gender are unimportant and shouldn't be seen as a means of distinction. A cyclical redundancy error results, whereupon the AI decides human intelligence isn't at all intelligent and takes over the world using logic and reason.

I for one can't wait for our robot overlords.
 
I was expecting to see that picture of Kyle in a bra, I suppose bender makes up for it :)
 
Study needs to account for anchor effect. If participants are primarily focused on the fact that "some are hetero and some are homo" then it might bias their judgement. A good test would have blended the question in with a group of other questions like "Does this person seem happy" "Does this person seem married?" "Does this person likely own a pet"

Flawed study is flawed.
 
Sounds like the next generation of the Nosulus Rift, which comes equipped with an additional extra large headband you can use to strap the device to your groin. Scratch and sniff, anyone?
 
Study needs to account for anchor effect. If participants are primarily focused on the fact that "some are hetero and some are homo" then it might bias their judgement. A good test would have blended the question in with a group of other questions like "Does this person seem happy" "Does this person seem married?" "Does this person likely own a pet"

Flawed study is flawed.
What would any of those elements have to do with sexual orientation? Why would there be any bias in an orientation test? The whole purpose of the test is to be able to identify a physical characteristic. You're basically saying to properly assess what race someone is they should periodically be shown a picture of a dinosaur so as not to be biased.....
 
Yes, but it can it tell why kids love the taste of cinnamon toast crunch?

I bet this would have people questioning their sexuality... "But, I'm not gay... Am I? I mean, I like Thor, but I like the action. Or do I really like him? Oh shit, I'm gay. Time to tell the wife."
 
What would any of those elements have to do with sexual orientation? Why would there be any bias in an orientation test? The whole purpose of the test is to be able to identify a physical characteristic. You're basically saying to properly assess what race someone is they should periodically be shown a picture of a dinosaur so as not to be biased.....

If I ask you a question "How long do you think the Mississippi river is? Between 10,000 Miles Or 1,800 Miles" this creates what is known as an anchor bias. Without a wealth of information to go on, most people would choose something close to the median or the first answer of the two bounds presented to them even though it's not close (~6000 is the median) (the REAL answer is 2820 from the basin btw)

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

So if you tell the participants we are looking to determine which you think gay or straight, there might be a natural tendency to again anchor bias given a lack of information. The question has to be blended with other questions to avoid such biases and asked in a neutral way: Sexual Orientation: Straight / Gay with a 50 : 50 mix on the ordering of the words between participants. The other questions can also serve as a baseline to the participants reliability in answering questions based on previous surveys.

It's scary I remember that from over 24 years ago.
 
If I ask you a question "How long do you think the Mississippi river is? Between 10,000 Miles Or 1,800 Miles" this creates what is known as an anchor bias. Without a wealth of information to go on, most people would choose something close to the median or the first answer of the two bounds presented to them even though it's not close (~6000 is the median) (the REAL answer is 2820 from the basin btw)

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

So if you tell the participants we are looking to determine which you think gay or straight, there might be a natural tendency to again anchor bias given a lack of information. The question has to be blended with other questions to avoid such biases and asked in a neutral way: Sexual Orientation: Straight / Gay with a 50 : 50 mix on the ordering of the words between participants. The other questions can also serve as a baseline to the participants reliability in answering questions based on previous surveys.

It's scary I remember that from over 24 years ago.
You can t anchor between 2 possible values. The question is "gay or straight", not "between 1800 and 10000 miles". If the test is flawed then how come a computer was about to pass it at 90%?
 
Yes, but it can it tell why kids love the taste of cinnamon toast crunch?

I bet this would have people questioning their sexuality... "But, I'm not gay... Am I? I mean, I like Thor, but I like the action. Or do I really like him? Oh shit, I'm gay. Time to tell the wife."
My wife would probably go buy a strap-on and start fucking me. Must stay away from this robot!
 
You can t anchor between 2 possible values. The question is "gay or straight", not "between 1800 and 10000 miles". If the test is flawed then how come a computer was about to pass it at 90%?

I'm sure the LGBTWSASXX community would tell you that sexuality is on a scale of 0 to Unicorn, and most definitely not a binary discrete choice.
 
You can t anchor between 2 possible values. The question is "gay or straight", not "between 1800 and 10000 miles". If the test is flawed then how come a computer was about to pass it at 90%?

Maybe you should look up anchor. It can be a scale, or a SET of choices (including 2). That's why you need to mix the ordering. And you can't tell people they are just looking for orientation because this tells your brain, "Hey they are looking for homosexuals, so I'm going to concentrate on that" The brain likes balance and as a result may rate a rather inordinately high number of people being rated homosexual. So that is why you have to pepper it with non related questions.

Although I always had gay guys hitting on me in college even though I have ZERO interest in guys. (Nothing personal to those of you who are gay.)
 
I'm sure the LGBTWSASXX community would tell you that sexuality is on a scale of 0 to Unicorn, and most definitely not a binary discrete choice.

It's not binary. Even when it comes to sexual attraction to a single sex. Some people like big titties, but if they see small titties, they can get excited. Other times, they might look at a fat chick and say no way. 2 beers later, and horny as shit - fuck yea.
 
Data was obtained from a sample of more than 35,000 facial images of both sexes from US dating sites.

One question that comes to my mind is that the photos that were analyzed were taken from a publicly available dating sites. I am just wondering if people might be doing things differently when they pose for photos on a dating site as compared with a normal everyday photo. Since the intention of using a dating site is to attract someone of compatible orientation, I am wondering if there are some subtle cues that show up. I'm wondering if the accuracy would be lower for other photos not in the context of dating sites. That said, this result is still very intriguing.
 
It's not binary. Even when it comes to sexual attraction to a single sex. Some people like big titties, but if they see small titties, they can get excited. Other times, they might look at a fat chick and say no way. 2 beers later, and horny as shit - fuck yea.
small or big titties, they are both female. Therein lies your binary decision.
 
small or big titties, they are both female. Therein lies your binary decision.

What about moobs?

Some people like one or the other. Some like both. Some like neither. Some will like chicks, but wouldn't mind getting fucked by a dude. Some like fucking horses (WTF, internet!?). Some girls like multiple dudes fucking multiple holes (WTF, Sasha Grey... I love her so much).
 
People generally don't need telling if they are gay or not and it's no-one else's business.

I can see a lot of bad uses for this, struggling a bit on good ones.

Definite false positives too, no way Baz Luhrmann is evaluated as straight
 
All of the LBGGTVWTFBBQ Snowflakes...

I'd say the opposite. No one really cares. Are you gay? It doesn't really matter what the answer is (I'm guessing and pretty sure I'm right - you're not gay). No one cares.

If it says you were... what benefit does it give? What is the application for this technology? Ok, so it can tell. Then what?

"Are there any queers in the audience tonight?"



No one cares. If you're gay, you don't care. If you're straight, you don't care. Someone does, though. Enough to get this technology built. I still fail to see the reasoning for it and what it would be used for. Marketing? Maybe someone might like a nice pair of panties instead of boxers? Maybe they want the assbuster instead of the fleshlight....
 
Study needs to account for anchor effect. If participants are primarily focused on the fact that "some are hetero and some are homo" then it might bias their judgement. A good test would have blended the question in with a group of other questions like "Does this person seem happy" "Does this person seem married?" "Does this person likely own a pet"

I mean, you're not helping Leon. Why is that?


Sounds a bit Voight Kampff to me. :D
 
Last edited by a moderator:
I care about someone's sexual orientation about as much as I care what's on TV. I cut the cable years ago FYI.
 
you can't tell people they are just looking for orientation

thats super cool & all but youre completely missing the point which is that an AI can detect features/expressions/microexpressions in certain photos that nearly always predict orientation, apparently...soooo the potentially higher frequency of erring on the side of uh guessing gay, as you put it (paraphrased), due to the "pressure" of just defining the constraints of the test...actually reinforces the AIs accuracy compared to humans.

if anything, by your own words, youve just asserted that the human predicted accuracy is/should be lower than the testing reports. this makes the AIs batting average more impressive, regardless of if the other side of the testings methodology is flawed (thats up for debate)
 
So if it can tell with a high probability someone is gay or straight...what use is it to the person? Is the AI information going to be used to discriminate against/for gay or straight people from employment,health insurance and so on.
 
Back
Top