SoylentNews is people
SoylentNews
from the can-it-tell-which-personality-is-currently-active? dept.
This psychologist's "gaydar" research makes us uncomfortable. That's the point.
Michal Kosinski used artificial intelligence to detect sexual orientation. Let him explain why.
By Brian Resnick@B_resnickbrian@vox.com Jan 29, 2018, 12:00pm ESTIn September, Stanford researcher Michal Kosinski published a preprint of a paper that made an outlandish claim: The profile pictures we upload to social media and dating websites can be used to predict our sexual orientation.
Kosinski, a Polish psychologist who studies human behavior from the footprints we leave online, has a track record of eyebrow-raising results. In 2013, he co-authored a paper that found that people's Facebook "likes" could be used to predict personal characteristics like personality traits (a finding that reportedly inspired the conservative data firm Cambridge Analytica).
For the new paper, Kosinski built a program with his co-author Yilun Wang using a common artificial intelligence program to scan more than 30,000 photos uploaded to an unnamed dating site. The software's job? To figure out a pattern about what could distinguish a gay person's face from a straight person's.
I hate the terms "Must see TV" and "must read" and similar terms. But, this article comes pretty close to "must read" for those who wish to understand where computer are going to take us. Especially read the conversation between Resnick and Kosinski - the research is not really about homosexuality, but about analyzing people in general.
Michal Kosinski
Exactly.
It proves to be uncomfortably accurate at making predictions.
We know that companies are already collecting this data and using such black boxes to predict future behavior. Google, Facebook, and Netflix are doing this.
Basically, most of the modern platforms are just virtually based on recording digital footprints and predicting future behavior.
Psychologists would say, "Oh, yes, that's true, but not personality. This is just pseudoscience." I'm like, wait. You can accept that you can predict 57 things, but if I say, "What about 58?" you say, "This is absolutely theoretically impossible. This is pseudoscience. How can you even say that?"
Science or pseudoscience, we can bet that corporate America and the government are going to be using this.
A smart person with a computer and access to the internet can judge sexual orientation of anyone in the world, or millions of people simultaneously with very little effort, which makes lives of homophobes and oppressive regimes just a tiny bit more easy.
(Score: 1, Insightful) by Anonymous Coward on Thursday February 01 2018, @04:42PM (10 children)
This proves that the cause of homosexuality is in nature, not nurture. Facial structure isn't set by choice (unless you see a plastic surgeon), so it must be set by either genes or environment. Thus, attempts to turn gays straight are pointless.
On the other other hand, this may lead opressive regimes to decide that gays are irredeemable and do worse things to them than they otherwise would.
(Score: 0) by Anonymous Coward on Thursday February 01 2018, @04:47PM
This only proves that gay people upload different photos than straight people when the purpose is attracting a partner. Colour me surprised.
(Score: 4, Insightful) by Anonymous Coward on Thursday February 01 2018, @04:56PM (2 children)
No, it doesn't. Note that it didn't analyse random pictures of the person, but pictures that person uploaded to the social media/dating site. That is, the picture not only tells something about the person's look, but also about the person's personality. You wouldn't update a random picture to the site, but one which you think gives a favourable impression of yourself.
Also note that profile pictures not only reveal biological traits, but also your hair style, and typically some of your clothing. The background may also be telling.
That is, if the system really does what is claimed, and not just on the training data. Remember, with neural networks, we never know what the network really learns.
There's the nice case of a system that was meant to distinguish own tanks from enemy tanks. It worked quite fine on the test data, but then failed spectacularly when applied in realistic conditions. It turned out that the images of the own tanks in the test set were all photographed at perfect weather, while the enemy tanks weren't, so what the system really had learned was to distinguish weather conditions.
(Score: 4, Interesting) by DannyB on Thursday February 01 2018, @05:03PM
The real test of your theory will be if an effective working AI Gaydar Camera can be developed. If so, and it is accurate on any random picture of a person, then the choice of picture uploaded would be irrelevant.
That would be an interesting experiment to identify if it is the selection of picture that is really what identifies orientation, or if any random picture of a person would identify that (and maybe other!) characteristics.
What about an AI algorithm that can identify from a photo whether a person is likely to be a serial killer?
The lower I set my standards the more accomplishments I have.
(Score: 2) by nobu_the_bard on Thursday February 01 2018, @09:04PM
The tank thing's an urban legend, though it remains a useful way of illustrating some of the limitations. As best as various investigators have been able to tell, it was merely a possibility that was discussed.
https://www.gwern.net/Tanks [gwern.net]
(Score: 3, Informative) by DannyB on Thursday February 01 2018, @05:48PM (1 child)
Two sibling replies state that this does not prove what it says. Let me quote from TFA . . .
This would suggest it has nothing to do with the photo one self selects to upload to a dating website.
The lower I set my standards the more accomplishments I have.
(Score: 4, Interesting) by AthanasiusKircher on Thursday February 01 2018, @08:53PM
I might encourage you to read some critiques of the study's interpretation and conclusions here [medium.com]. There have been responses to these criticisms too, but if you read this stuff, it becomes much less clear-cut.
For example, the study claims a 71% success rate on female faces for determining homosexuality. The critical response I linked above showed that they could achieve a 63% success rate just by creating a naive algorithm on the basis of whether a woman wears eyeshadow. (Lesbians apparently are statistically less likely to wear eyeshadow.)
If they include sex more yes/no questions about presentation (“Do you ever use makeup?”, “Do you have long hair?”, “Do you have short hair?”, “Do you ever use colored lipstick?”, “Do you like how you look in glasses?”, and “Do you work outdoors?”), they increased the success rate of their simplistic algorithm to 70%, which is nearly identical to the performance of the algorithm based on a photo. Note that all of these questions are based on superficial characteristics that have nothing to do with "innate" facial features, but rather are about stylistic or cultural choices: the first four obviously are superficial things that would influence a photo appearance. Lesbians are apparently more likely to wear glasses. And the "outdoors" question comes from the fact that a poll showed those with homosexual interests had a slightly less likely chance of working outdoors, which correlated with paler skin... i.e., whether they had a "tan" or not. (The study authors from TFA had suggested a difference in skin color for homosexuals due to hormones, but it might have simpler origins...)
That link also notes that in some other studies involving photos taken in controlled lab conditions (same angle, same lighting, no glasses consistently, no make-up), previous AI algorithms basically dropped to chance in terms of ability to predict homosexuality.
That doesn't deny that there couldn't be SOME level of judgment based on facial characteristics that are more "innate" in homosexuals. But it seems there's a pretty strong component of these judgments that could be based on superficial characteristics, style choices, and "social signaling" in the way photos are chosen (lighting, angle, pose, etc.).
Let the authors repeat their study with lab controlled photos taken under the same lighting, pose, and other superficial characteristics controlled for. Then we can reevaluate how strong their claims that their algorithm is picking up on fundamental aspects of facial structure (perhaps determined by more basic physiological characteristics like hormones, etc.) are.
(Note, by the way, that there's good reason to be suspicious of such claims. There is a VERY long history in science of trying to identify various "degenerates" based on basic physical characteristics. In the late 19th and early 20th centuries this was particularly pronounced, with lots of scientists claiming to be able to identify those with criminal disposition, homosexuals, etc. just by measuring various elements of their heads or bodies. These sorts of studies were often praised by eugenicist movements, which even forced sterilization on some such individuals with such defining characteristics. I'm NOT being alarmist here -- I'm just noting that there are still those who want to claim homosexuals are "degenerate" in some way, and they'd love physical data proving they are different in measurable, perceptible ways. And maybe they are... maybe there are some basic physical characteristics that can make the difference. But this study hasn't really met the bar for telling us how much of its algorithm -- if any -- is actually differentiating on that basis... rather than more superficial characteristics of the photos determined by cultural signalling.)
(Score: 2) by HiThere on Thursday February 01 2018, @06:41PM
Actually, I *think* this is a reference to research I read about a month or so ago in Science News. The AI learned to recognized dress styles and hair styles, and similar things, not physical characteristics. And there was the comment that the photos that people upload to a dating site aren't random choices. It wasn't tested against random pictures. My guess is because nobody felt comfortable about walking up to a stranger and saying "Pardon me, are you gay?".
The recognized characteristics are likely to be something that changes as rapidly as hair styles go in and out of fashion. (I haven't seen a lime-green Mohawk cut in quite awhile.)
Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
(Score: 2) by Runaway1956 on Thursday February 01 2018, @11:48PM
Nonsense - this proves absolutely nothing. You don't know - you can't know - whether being gay makes a mark on your, or the marks exist before you become gay. Which is the cause, and which is the effect? Worse, we don't even have a proven correlation yet - just a single study that suggests some of this stuff are correlated.
Science, or pseudo-science? If it's actual science, it needs a good bit more work before it's really useful. (Really useful, as opposed to other pseudosciences admisable in court, and government proceedings, such as lie detectors.)
(Score: 3, Insightful) by VLM on Friday February 02 2018, @12:00AM (1 child)
Two separate and interesting "not quite" comments:
1) The soy-boy meme where high soy consumption among people not genetically predisposed to soy intake (urban white males, perhaps?) results in very low T levels and increased estrogen levels and bodily changes to an aesthetic somewhat appealing to gay folks. Force feed a borderline bi male enough soy, the hormonal changes will push them gay. At least thats the "soy boy" meme. I only believe it about 60% to 80%, I think it more likely right than wrong but I'm not betting the farm on it. It smells believable in the sense that there's plenty of non-controversial science about amphibians and other animals where environmental contaminants can really fuck up their sexuality, and I'm not enough of a creationist to think human animals are magically protected from chemical toxins by some higher progressive power, so even if it turns out not to be soy, it seems realistic there exists some organic chemical out there that can screw up sexuality from the point of view of successful evolutionary reproduction.
2) Classic facial structure issue is fatty face vs skinny face. And there's a subculture of gay men who like the emaciated aesthetic, so you could do worse for an algo with something like bulimic level skinny-face is a somewhat higher than average chance of gay. We're talking about a game of small percentages here, not "all skinny men is gay" level stupidity.
I would be interested to know if the study corrected for hormonal imbalances and simple variation in body weight (so a subpopulation preselected of gay and straight 200 pound men with T levels around 800, lets say, with the implication the higher T means 200 pounds of muscle not lard)
(Score: 2) by Runaway1956 on Friday February 02 2018, @10:38AM
Anyone looking for a link between low testosterone in males, and high estrogen in males and females, need look no further than dairy products. Agrobusiness keeps the dairy cows loaded with hormones, antibiotics, and whatever else they deem "profitable". Milk and cheese is all-pervasive in the American diet, so you'll have a hard time avoiding all that.