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.
> 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.
> 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.)
"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.
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)
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.
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.
If people thought about what they were doing at all times, both claims could be true but in the real world, both of those claims will almost surely be rejected. People don't choose to be certain, they are just unaware of their overconfidence.
Overconfidence doesn't change the probabilities of the rare, harmful events that they are predicting won't happen. But their bias against self-doubt does increase their willingness to bet against unprecedented events.. n=1001 safe observations (where E[payoff] > 0) cannot predict that the next will be catastrophic (E[payoff] = -inf, or at least absolute value >> O(N*E[payoff seen up till then])) with non-zero probability beforehand; the expected payoff would appear vastly different before the event than after. The bets they choose will reflect the way they underestimate the consequences of speculation.
Evidence of absence > absence of evidence, another way to put that is the weight of ¬valid(test) >> the weight of ¬¬valid(test), so we should not decide the test is invalid just because it's not yet invalidated.
My gut feeling is that the upper-bound of proven benefit (to individuals/society) of the test (even if ~99.99% accurate) << upper-bound of the harm (intentionally or not) - because 'invalidity' isn't a question of whether a test is always non-satisfiable, the burden of proof falls upon the one introducing it, rather than the one rejecting it.
> 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.
So why can't they use that experience to make the same point against a AI bot labeler. Too much of a leap?
No? The exact same reasoning works regardless of what kind of oracle the labeler is; it's not at all affected by increases in labeler accuracy over ordinary human accuracy.
I was going to mock this post because I think the lgbbq movement was a cultural Marxist conspiracy against the institution of the family unit, but your points are absolutely beautiful hypotheticals that everyone can get behind.
Individuals absolutely have to be free to determine who they are, and that means that society can only judge their determinations, not they themselves.
I will set aside my quibbles with the rest of of the package for another day, and simply extend you kudos for a well thought-out post.
What exactly are the priors for the idea that there is a group of people you call "cultural Marxists", who have done this, and further that they have done it?
The standard version of the "cultural Marxism" conspiracy theory states that the Frankfurt School, a group of European intellectuals, hatched a plot to bring down Western civilisation. If you believe this is the case, I have two questions:
- What works of the Frankfurt School have you read, and which do you think are congruent with what you are accusing them of?
- Do you engage in the anti-semitic element of the conspiracy theory?
If you think this is not the case, then tell me who exactly "cultural Marxists" are, their origins, and what basis you have for your accusations. These are very basic questions which you should not have to reference another book to answer, though you can of course provide a reference to one as part of your answer.
> I think the lgbbq movement was a cultural Marxist conspiracy against the institution of the family unit
Have you ever considered that, oh I don't know, maybe it's just that people would like to live their lives in peace, the way they want, fucking whomever they want without bothering anyone else or being bothered in return? What on earth possesses you to attribute this to some sinister decades long underground plot?
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.
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.
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.
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.
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.
It was a major can of worms when we started image recognition years ago. I think privacy will naturally continue to decrease if there's no regulation. More information is ammo for both good and bad applications. At first something like this could be used for affirmative action: airport security, seeking potential employees; anything where your appearance might be important. I think The West has more or less learned to ridicule the idea that "some people that naturally look a certain way should be treated a certain way". But yeah, if you're sweaty and shady at a presidential convention, it would be nice to have this kind of technology in place.
> 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.
I completely disagree -- I think the consequences of this technology could potentially alleviate much suffering, but it's important that we get ahead of it in terms of policy, so that the eugenicists can't capitalize on it.
As taboo as it is to say, I believe there are statistically significant differences in capacity and motivation between people.
If we continue to deny this (the current politically correct response), our wider social policy will assume an equality of capability that's just not true. This leads to policy solutions like re-education as a solution to homelessness or jobs lost to technological improvements, rather than accepting that some people can't hold down jobs that are valued in the 21st century and just giving them some damn housing and money to feed themselves.
Income inequality is a gigantic issue facing us in the next 20-50 years, and a sober understanding of the population distribution of talent is important as we face that. Most taxi and uber drivers displaced by self driving cars are not going be smart enough to take dev bootcamps and become software developers. It's critical that we accept this as a society so that we can get safety nets like universal income or government make-work programs in place, so that these folks can live decent lives.
> This is a truly horrifying scenario, an ultimate black pill and humanity will need to confront it soon it seems.
This was the most frustrating thing about James Damore's firing from Google, is that rather than debating the accuracy and policy consequences of Damore's point (equal intellectual capacity but differing levels of motivation for software engineering work between men and women) they stuck their head, called him a bigot, and fired him.
Technologies like this could be amazing for progressive politics, but these cherished notions of equality of capability need to die. I don't think the current zeitgeist of the left will let that happen.
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.
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.
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.
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?
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.
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.
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.
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.
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.
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.
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.
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
58 comments
[ 3.2 ms ] story [ 125 ms ] threadThe panic is being manufactured, probably for little more than grant money, recognition, and clicks.
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.)
"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.
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)
The potential for ostracism is just too great.
Overconfidence doesn't change the probabilities of the rare, harmful events that they are predicting won't happen. But their bias against self-doubt does increase their willingness to bet against unprecedented events.. n=1001 safe observations (where E[payoff] > 0) cannot predict that the next will be catastrophic (E[payoff] = -inf, or at least absolute value >> O(N*E[payoff seen up till then])) with non-zero probability beforehand; the expected payoff would appear vastly different before the event than after. The bets they choose will reflect the way they underestimate the consequences of speculation.
Evidence of absence > absence of evidence, another way to put that is the weight of ¬valid(test) >> the weight of ¬¬valid(test), so we should not decide the test is invalid just because it's not yet invalidated.
My gut feeling is that the upper-bound of proven benefit (to individuals/society) of the test (even if ~99.99% accurate) << upper-bound of the harm (intentionally or not) - because 'invalidity' isn't a question of whether a test is always non-satisfiable, the burden of proof falls upon the one introducing it, rather than the one rejecting it.
So why can't they use that experience to make the same point against a AI bot labeler. Too much of a leap?
Individuals absolutely have to be free to determine who they are, and that means that society can only judge their determinations, not they themselves.
I will set aside my quibbles with the rest of of the package for another day, and simply extend you kudos for a well thought-out post.
If this isn't sarcasm/irony, what's your evidence for this conspiracy?
(Edit: Why is this comment downvoted? Is it frowned upon to ask for evidence for a conspiracy theory?)
Fully understand that priors != evidence. Which is why I'll state it's my belief, not ground truth.
The standard version of the "cultural Marxism" conspiracy theory states that the Frankfurt School, a group of European intellectuals, hatched a plot to bring down Western civilisation. If you believe this is the case, I have two questions:
- What works of the Frankfurt School have you read, and which do you think are congruent with what you are accusing them of?
- Do you engage in the anti-semitic element of the conspiracy theory?
If you think this is not the case, then tell me who exactly "cultural Marxists" are, their origins, and what basis you have for your accusations. These are very basic questions which you should not have to reference another book to answer, though you can of course provide a reference to one as part of your answer.
>Do you engage in the anti-semitic element of the conspiracy theory?
>Oh, you don't toe the PC line? You must be a rayciswhitemale.
You're self-parodic at this point.
Apparently not being an anti-semite is being "PC"?
Have you ever considered that, oh I don't know, maybe it's just that people would like to live their lives in peace, the way they want, fucking whomever they want without bothering anyone else or being bothered in return? What on earth possesses you to attribute this to some sinister decades long underground plot?
What can i say?
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.
That said, I can imagine a AI-driven screening mechanism that was rooted in a causal relationship to data, instead of random nonsense.
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.
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.
I completely disagree -- I think the consequences of this technology could potentially alleviate much suffering, but it's important that we get ahead of it in terms of policy, so that the eugenicists can't capitalize on it.
As taboo as it is to say, I believe there are statistically significant differences in capacity and motivation between people.
If we continue to deny this (the current politically correct response), our wider social policy will assume an equality of capability that's just not true. This leads to policy solutions like re-education as a solution to homelessness or jobs lost to technological improvements, rather than accepting that some people can't hold down jobs that are valued in the 21st century and just giving them some damn housing and money to feed themselves.
Income inequality is a gigantic issue facing us in the next 20-50 years, and a sober understanding of the population distribution of talent is important as we face that. Most taxi and uber drivers displaced by self driving cars are not going be smart enough to take dev bootcamps and become software developers. It's critical that we accept this as a society so that we can get safety nets like universal income or government make-work programs in place, so that these folks can live decent lives.
> This is a truly horrifying scenario, an ultimate black pill and humanity will need to confront it soon it seems.
This was the most frustrating thing about James Damore's firing from Google, is that rather than debating the accuracy and policy consequences of Damore's point (equal intellectual capacity but differing levels of motivation for software engineering work between men and women) they stuck their head, called him a bigot, and fired him.
Technologies like this could be amazing for progressive politics, but these cherished notions of equality of capability need to die. I don't think the current zeitgeist of the left will let that happen.
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.
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.
"Automated Inference on Criminality using Face Images"
https://arxiv.org/pdf/1611.04135v2.pdf
Confirmed some of my long-held biases about physiognomy though, so I didn't dismiss it outright.
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.
Holsters are not always worn in the lower extremities, either. I'm not sure this would be all that successful. It seems quite difficult.
Why would your success as a human have anything to do with whether or not an algorithm could have success in any field?
[0]https://en.m.wikipedia.org/wiki/Blink_(book)
[1]https://en.m.wikipedia.org/wiki/Silvan_Tomkins
[2]http://gladwell.com/blink/the-mysteries-of-mind-reading/
Meaning the author really doesn't understand the concept of unbalanced classes