Stanford Researchers tried to create a "Gaydar Machine" using neural networks

[captain_to_fire]

Sources: New York Times, Author's Note (via Google Docs), OSF (pre-print) wang_kosinski.pdf

 

Quote

Deep neural networks are more accurate than humans at detecting sexual orientation from facial images.

 

ABSTRACT

We show that faces contain much more information about sexual orientation than can be perceived and interpreted by the human brain. We used deep neural networks to extract features from 35,326 facial images. These features were entered into a logistic regression aimed at classifying sexual orientation. Given a single facial image, a classifier could correctly distinguish between gay and heterosexual men in 81% of cases, and in 74% of cases for women. Human judges achieved much lower accuracy: 61% for men and 54% for women. The accuracy of the algorithm increased to 91% and 83%, respectively, given five facial images per person. Facial features employed by the classifier included both fixed (e.g., nose shape) and transient facial features (e.g., grooming style). Consistent with the prenatal hormone theory of sexual orientation, gay men and women tended to have gender-atypical facial morphology, expression, and grooming styles. Prediction models aimed at gender alone allowed for detecting gay males with 57% accuracy and gay females with 58% accuracy. Those findings advance our understanding of the origins of sexual orientation and the limits of human perception. Additionally, given that companies and governments are increasingly using computer vision algorithms to detect people’s intimate traits, our findings expose a threat to the privacy and safety of gay men and women. (Emphasis is mine)

Comments could get nasty here. I need your moderating powers @iamdarkyoshi for anyone with comments against the CS (political/religious rants) or derails the topic.

Quote

Whether he has now created “A.I. gaydar,” and whether that’s even an ethical line of inquiry, has been hotly debated over the past several weeks, ever since a draft of his study was posted online. Presented with photos of gay men and straight men, a computer program was able to determine which of the two was gay with 81 percent accuracy, according to Dr. Kosinski and co-author Yilun Wang’s paper.

Teaching a Machine to ‘See’ Sexuality

Dr. Kosinski and Mr. Wang began by copying, or “scraping,” photos from more than 75,000 online dating profiles of men and women in the United States. Those seeking same-sex partners were classified as gay; those seeking opposite-sex partners were assumed to be straight. Some 300,000 images were whittled down to 35,000 that showed faces clearly and met certain criteria. All were white, the researchers said, because they could not find enough dating profiles of gay minorities to generate a statistically valid result. The images were cropped further and then processed through a deep neural network, a layered mathematical system capable of identifying patterns in vast amounts of data. Dr. Kosinski said he did not build his tool from scratch, as many suggested; rather, he began with a widely used facial analysis program to show just how easy it would be for anyone to pull off something similar.

The software extracts information from thousands of facial data points, including nose width, mustache shape, eyebrows, corners of the mouth, hairline and even aspects of the face we don’t have words for. It then turns the faces into numbers. “We showed that this model produces slightly different numbers for gay and straight faces,” Dr. Kosinski said. The authors were then ready to pit their prediction model against humans in what would become a notorious gaydar competition. Both humans and machine were given pairings of two faces — one straight, one gay — and asked to pick who was more likely heterosexual. The participants, who were procured through Amazon Mechanical Turk, a supplier for digital tasks, were advised to “use the best of your intuition.” They made the correct selection 54 percent of the time for women and 61 percent of the time for men — slightly better than flipping a coin. Dr. Kosinski’s algorithm, by comparison, picked correctly 71 percent for of the time for women and 81 percent for men.

When the computer was given five photos for each person instead of just one, accuracy rose to 83 percent for women and 91 percent for the men. After the study was referenced in an article in The Economist, the 91 percent figure took on a life of its own. News headlines “made it sound almost like an X-ray that can tell if you’re straight or gay,” said Dr. Jonathan M. Metzl, director of the Center for Medicine, Health, and Society at Vanderbilt University. Yet none of the scenarios remotely resembled a scan of people “in the wild,” as Ms. Garvie put it. And when the tool was challenged with other scenarios — such as distinguishing between gay men’s Facebook photos and straight men’s online dating photos — accuracy dropped to 74 percent. There’s also the issue of false positives, which plague any prediction model aimed at identifying a minority group, said William T.L. Cox, a psychologist who studies stereotypes at the University of Wisconsin-Madison. Let’s say 5 percent of the population is gay, or 50 of every 1,000 people. A facial scan that is 91 percent accurate would misidentify 9 percent of straight people as gay; in the example above, that’s 85 people (0.91 x 950). The software would also mistake 9 percent of gay people as straight people. The result: Of 130 people the facial scan identified as gay, 85 actually would be straight. 

 

“When an algorithm with 91 percent accuracy operates in the real world,” Dr. Cox said, “almost two-thirds of the times it says someone is gay, it would be wrong.” He noted in an email that “the algorithms were only trained and tested on white, American, openly gay men (and white, American, presumed straight comparisons),” and therefore probably would not have broader implications.

It resembles a lot of a topic I posted before about an AI predicting the outcome of a relationship.

I won't lie that it got me concerned for a bit but reading the actual paper and the FAQ (Google Docs) the authors provided made me think it's indeed a way to warn people about the repercussions of artificial intelligence and machine learning. It reminds me when Elon Musk said that Mark Zuckerberg doesn't know the dangers of AI if left unregulated. I'm with Elon Musk on this one. 

On the lighter side of things, this technology can be used by dating apps targeted to gay individuals but then, I'd rather see the person face to face rather than letting an AI do it for me. Looking at the comments of other experts, it's obvious that the algorithm is very limited on their sample size and subjects since they only used gay white men and women. The thesis authors said in their Google document that:

Quote

 

Despite our attempts to obtain a more diverse sample, we were limited to studying white participants from the U.S. (Unfortunately, this seems to be the problem affecting most other studies of sexual orientation.)

 

This does not invalidate the results of the study in any way. The study shows that you can distinguish between white gay and straight individuals.

 

It does not show that the same applies to other ethnicities - but the our findings suggest that this, unfortunately, is likely.. The same biological, developmental, and cultural factors—which are responsible for differences between gay and straight individuals—are likely to affect people of other races as well.  

 

Just imagine you're a tourist in a country and a facial scanner at the airport identifies someone as gay or lesbian, they can immediately put you in a database. Just imagine oppressive countries who have staunch religious doctrines using this to identify closeted gay people and have them thrown from top of buildings or just be shamed by society into committing suicide. Or better yet, just imagine North Korea developing an AI who determines if someone has doubts on the regime or is planning to defect to South Korea. 

Quote

Recent press reports, however, suggest that governments and corporations are already using tools aimed at revealing intimate traits from faces. Facial images of billions of people are stockpiled in digital and traditional archives, including dating platforms, photo-sharing websites, and government databases. Profile pictures on Facebook, LinkedIn, and Google Plus are public by default. CCTV cameras and smartphones can be used to take pictures of others’ faces without their permission.

 

We felt that there is an urgent need to make policymakers and LGBTQ communities aware of the risks that they are facing. Tech companies and government agencies are well aware of the potential of computer vision algorithm tools. We believe that people deserve to know about these risk and have the opportunity to take preventive measures.

Knowledge is power and just like a knife that can be used to stab someone or cut that delicious medium rare steak, this algorithm is also a double edged sword. At the moment, it's not ready for prime time as it has limited sample diversity and the authors admit that. In fact as stated in their Google Doc that they want to be wrong and they are terrified of the results. More refinement is definitely needed and just like any scientific theory, it needs to be replicated. At the moment, all humanity knows about the potentials and risks of AI and machine learning is just at the tip of the iceberg. So there's no need to be concerned or alarmed just yet but we also shouldn't dismiss a scientific finding just because we disagree with it or it could pose harm to us and in fact, I would like for this paper to be published soon in the Journal of Personality and Social Psychology. I'm not a Psychology major but I encourage people with that degree to replicate this thesis but with much diversified subjects. 

 

Also, it should be kept in mind that "peer-review" is not always cranked up to be as a method of validating or dismissing a theory. Peer-review is important in academiai but most of the time, it's just validating research methodology, checking grammar and look for suspicious and interpolated/forged results. Peer-review doesn't determine if a thesis' conclusion is right or wrong. In fact, there are a lot of bogus theses that are "peer-reviewed". The only way to determine if a thesis' conclusion is right or wrong is if it's replicated and check the results to see if they match up or not. [See NCBI, Nature, CBC, WSJ)

 

Again, I highly encourage everyone reading this thread to read this Google Doc as it clarifies a lot of concerns on privacy, human rights, etc. 

https://docs.google.com/document/d/11oGZ1Ke3wK9E3BtOFfGfUQuuaSMR8AO2WfWH3aVke6U/edit?usp=sharing

[Ashley MLP Fangirl]

i personally think this is a good thing. i'm gay, but i had to figure it out myself, and was in doubt for a long time. (years)

 

if they can build an algorithm that is 100% accurate, it could help a lot of people who are unsure of their sexuality.

 

[captain_to_fire]

1 minute ago, firelighter487 said:

i personally think this is a good thing. i'm gay, but i had to figure it out myself, and was in doubt for a long time. (years)

Same

1 minute ago, firelighter487 said:

if they can build an algorithm that is 100% accurate, it could help a lot of people who are unsure of their sexuality.

But even that, I don't think one's personal issues especially sexual orientation should be based on an AI alone. The paper is a bit subjective since it relies on facial stereotypes of white gays and lesbians on online dating apps and feed it to a neural network. 

[Shimejii]

1 minute ago, firelighter487 said:

i personally think this is a good thing. i'm gay, but i had to figure it out myself, and was in doubt for a long time. (years)

 

if they can build an algorithm that is 100% accurate, it could help a lot of people who are unsure of their sexuality.

 

I can forsee this being used as a weapon in other countries, those that kill gay people. And its interesting to read, but i dont think anything like this based on physical features could ever be 100% Its a tool, but a dangerous and careless one in the wrong hands.

[Ashley MLP Fangirl]

5 minutes ago, hey_yo_ said:

But even that, I don't think one's personal issues especially sexual orientation should be based on an AI alone. The paper is a bit subjective since it relies on facial stereotypes of white gays and lesbians on online dating apps and feed it to a neural network. 

that's true. but if they "could" make one that's pretty reliable, it could still help people.

4 minutes ago, Shimejii said:

I can forsee this being used as a weapon in other countries, those that kill gay people. And its interesting to read, but i dont think anything like this based on physical features could ever be 100% Its a tool, but a dangerous and careless one in the wrong hands.

yes, that is a big concern.

[Ashley MLP Fangirl]

5 minutes ago, VegetableStu said:

there was a gaydar article a month back wasn't it o_o

maybe a different one.

 

i've seen loads of articles that claim that gay people have a "gaydar"..

[captain_to_fire]

9 minutes ago, VegetableStu said:

there was a gaydar article a month back wasn't it o_o

Which one? The OP was a fairly new one.

[Taf the Ghost]

Hormonal expression effects facial features, along with musculature and neurological response pathways.

 

It also works in a lot of other chronic disorders, as the body responds to issues forced upon itself. The most stark one is always Down's Syndrome, but it applies to most genetic disorders. Though it's normally a lot more subtle, so in normal interaction you don't notice it. There's a big, future market for diagnosing genetic disorders this way. (Or at least flagging off there is something wrong.) 

 

The other application will be things like pedophiles and sociopaths. That type of dysfunction also shows up in the face as well.

[Taf the Ghost]

5 minutes ago, firelighter487 said:

maybe a different one.

 

i've seen loads of articles that claim that gay people have a "gaydar"..

In people it's not really "gaydar". It's pheromone response criteria. Ovulating Women are most sensitive to it, as a homosexual male will generally not be showing interesting a Woman.

 

The facial side of things is a pattern recognition that humans do. You're about 70% likely to pick out a co-religionist as well. Humans "know their own" at a fairly sophisticated & instinctual level.

[Ashley MLP Fangirl]

2 minutes ago, Taf the Ghost said:

In people it's not really "gaydar". It's pheromone response criteria. Ovulating Women are most sensitive to it, as a homosexual male will generally not be showing interesting a Woman.

 

The facial side of things is a pattern recognition that humans do. You're about 70% likely to pick out a co-religionist as well. Humans "know their own" at a fairly sophisticated & instinctual level.

oh okay. i get it.

[captain_to_fire]

24 minutes ago, Taf the Ghost said:

n people it's not really "gaydar".

To be fair, what is considered as "gaydar" in people is mostly based on stereotypes and observation of people's body language. But just because a man is a hair stylist in a salon doesn't mean he's gay nor does a soldier in a war zone mean he's straight.

Spoiler

Back in college, I have a personal way of discerning if someone is gay or not. I look into the movement of their eyes when a good looking man is passing by or talking to them. More often than not, their eyes are fixated to always look on that good looking guy and that's how I spot if someone is gay or not. 

 

[Guest]

Poll question is odd. I am not comfortable nor uncomfortable with such a test without a reason to knowing why and what will happen with such information. 

 

I would take it for funsies if it were private information, not data logged in the nether somewhere to used for or against me at some point. I value people knowing as little as possible about myself. 

[Bananasplit_00]

this is creepy AF

[PlayStation 2]

This is the money we waste, guys. I dunno about you but I’d rather not have an AI tell me my sexuality. 

[captain_to_fire]

17 minutes ago, Dan Castellaneta said:

This is the money we waste, guys. I dunno about you but I’d rather not have an AI tell me my sexuality. 

But it's worth exploring. I just don't want to see a perfected version of this in the hands of wrong people. 

[PlayStation 2]

Just now, hey_yo_ said:

But it's worth exploring. I just don't want to see a perfected version of this in the hands of wrong people. 

The thing with this is that it could very well go down a slippery slope path. Give them one thing and next thing you know they have the whole pie. 

[TheUncannyScrub]

AI NN to tell if you're gay or not. Thats just weird and i'm pretty sure impossible or at least inaccurate. Everyone face is unique and if it is going off what I think it would be then it would just be a machine that stereotypes people for how they look.

[JoseGuya]

Now your iPhone X could know if you're gay and auto buy that new season of RuPaul for you!

[captain_to_fire]

1 minute ago, ivan134 said:

-snip-

Please don't start a debate in violation with the CS

3 minutes ago, TheUncannyScrub said:

AI NN to tell if you're gay or not. Thats just weird and i'm pretty sure impossible or at least inaccurate. Everyone face is unique and if it is going off what I think it would be then it would just be a machine that stereotypes people for how they look.

It is inaccurate as admitted by the authors in their Google Doc I attached in the OP

[captain_to_fire]

Just now, JoseGuya said:

Now your iPhone X could know if your gay and auto buy that new season of RuPaul for you!

Love RuPaul's Drag Race!  Though I am not buying an iPhone X because I think something creepy might be going on with the Face ID sensor.

 

[DildorTheDecent]

This is hilarious! 

 

2017 we almost did it. 

[TheCherryKing]

The folks in Saudi Arabia will be very happy with this! 

[TheCryoknight]

Pure pseudoscience, PHT is entirely unproven, also while not only does it not include bisexuality, Pansexuality transgender, and non-binary. It entirely goes against of what we know of human sexuality being fluid. Some people's sexuality changes throughout their lifetime. This test as it is entirely broken by that fluidity; their faces do not physically change when their sexuality shifts. The test defines sexuality as a constant something it just is not.

[mr moose]

I long for a time when a device like this is as benign as a computer that tells you what color suit you are wearing.   Basically I want to live in a time when sexual preference is as benign as the color of your jumper.

[InertiaSelling]

10 hours ago, Taf the Ghost said:

The other application will be things like pedophiles and sociopaths. That type of dysfunction also shows up in the face as well.

What if pedophiles and sociopaths were actually more common than we think, and a lot of them were able to refrain from committing crimes ? We would get an endless watchlist of probably-but-not-undoubtedly-innocent people. It would be a mess.