Faception is phrenology for the 21st century!


What? You never heard of Faception? It’s amazing!

Faception is first-to-technology and first-to-market with proprietary computer vision and machine learning technology for profiling people and revealing their personality based only on their facial image.

Faception can analyze faces from video streams (recorded and live), cameras, or online/offline databases, encode the faces in proprietary image descriptors and match an individual with various personality traits and types with a high level of accuracy. We develop proprietary classifiers, each describing a certain personality type or trait such as an Extrovert, a person with High IQ, Professional Poker Player or a Terrorist. Ultimately, we can score facial images on a set of classifiers and provide our clients with a better understanding of their customers, the people in front of them or in front of their cameras.

In short, they claim to be able to take a picture of you, and in a fraction of a second, in the absence of any other information, be able to determine your personality. It’s done with computers, so you know you can trust it.

But, you know, we have these meat computers in our heads that have been trained through millions of years of evolution and throughout our development to be extraordinarily sensitive to faces. We can respond to an eyebrow that lifts a fraction of a millimeter, a lip that curls just so, a pupil that dilates by a hairs-width. We are really, really good at reading emotion and incredibly subtle social cues from facial expressions. There are autistic individuals who have more difficulty than most at picking up on those cues, but I suspect they’re far, far better at it than any computer program.

But even with this amazing sensitivity, do you really think you could look at a snapshot and accurately judge whether someone is a poker player or a terrorist? Don’t you suspect you’d get very different perspectives on the person’s personality if the snapshot were taken at a party, vs. taken at a funeral? Have you ever discovered that your first impression of someone, on the basis of their appearance, proves to be totally wrong once you get to know them better? Just on the face of it, this is an absurd claim they are making.

We haven’t even gotten to the actual “science” that they claim supports their thesis.

The face can be used to predict a person’s personality and behavior.

This claim relies on a combination of two known research observations

OK. They’ve got two justifications. Here’s the first. The second isn’t going to get any better.

1. According to Social and Life Science research personalities are affected by genes.​

In fact genes play a greater role in determining key personality traits like social skills and learning ability than the way we are brought up by our parents, researchers claimed.

Researchers from Edinburgh University studied more than 800 sets of identical and non-identical twins to learn whether genetics or upbringing has a greater effect on how successful people are in life. Writing in the Journal of Personality, the researchers found that identical twins were twice as likely as non-identical twins to share the same personality traits, suggesting that their DNA was having the greatest impact.

There’s an obvious problem with this argument: identical twins don’t have identical personalities. I’ve known a few identical twins, and it’s true that it may be hard to tell them apart visually, but try talking to them. They’re different people. Shocking, I know. Also, I think every twin has heard the joking question about which one is the evil twin, which the good twin (and is tired of it). How does software that only examines the most superficial aspects of a person fail to recognize this fundamental issue?

And, oh Jesu, they bring up that nonsense about measure the relative roles of genetics or upbringing. It doesn’t work that way! Both are interdependent. Genes will be expressed differently depending on the environment. Anyone who brings up nature/nurture as if they are competing hypotheses should be kicked to the curb and ignored.

But now watch the magnificent flying leap they have to make to argue that personality and facial features are linked.

2. Our face is a reflection of our DNA​

Researchers have identified five of the genes that shape a person’s face. Researchers previously knew that genetics played a large role in determining face shape, since identical twins share DNA. However, little was known about exactly which genes are involved. Three genes were thought to have roles in the arrangement of facial features, and the new research confirmed their involvement.

There are more than 3 or 5 genes involved in face development. Facial structure, like all complex traits, is almost certainly omnigenic. In point of fact, though, claiming that only a small number of genes affect face shape effectively contradicts their own argument, because that implies that these genes could do their job in morphology while another, completely different set of genes work to generate personality. Do these guys even understand what they’re talking about?

Working on mice, researchers have identified thousands of small regions of DNA that influence the way facial features develop. The researchers said that although the work was carried out on animals, the human face was likely to develop in the same way. In fact, It’s already possible to make some inferences about the appearance of crime suspects from their DNA alone.

Let it be stipulated that genes play a role in morphology, and that individuals who are genetically identical will develop very similar facial features. This is not a point of contention. The question is whether there is a correlation between physiognomy and personality. Do people with the same shape of nose have similar political stances, for instance? Is there a connection between bushy eyebrows and charitable giving? Is ear shape linked to gambling skill? These are the questions they need to answer.

They claim on their “classifers” page that this is the face of a pedophile, for instance. I think that looks like about half the white dudebros who work in tech. Are you comfortable with that suggestion? Because I’m not.

It does make me wonder if they’ve run their own photos through the machine, and whether the software has been optimized to weed out all the non-flattering descriptors. I wouldn’t be surprised if the first pass of the alpha version classified everyone as “asshole”, and they worked assiduously so it would instead classify them as “genius”.

But where is the evidence that face and personality are linked in a predictable way? Here we go.

And indeed, researchers were able to demonstrate that “Internal facial features are signals of personality and health. While these type of affirmations are quite recent, in Chinese history, there have been people that have studied the “mapping of the face” for thousands of years.

Also, the epigenetics phenomena has recently demonstrated, in academia researches and March 31th 2017 Science magazine.

Vague reference, no specific citation.

Vague reference to Chinese tradition.

Handwavy mention of epigenetics buzzword.

This is bad stuff. It’s just more hocus-pocus, like astrology or phrenology or iridology or tea-leaf reading or Myers-Briggs tests or any of a thousand gimmicks that have been used to make false claims throughout history. This Faception boondoggle is going to be no more effective at detecting terrorists than it would be at ferretting out witches.

Worse, it just genetic determinism coupled to racism as an excuse. Cesare Lombroso lives! It does make one wonder if his pickled head doesn’t actually dictate racist pseudoscience to bad writers and software authors.

Comments

  1. blf says

    From almost two years ago, ‘Facial-profiling’ could be dangerously inaccurate and biased, experts warn (Oct-2016; Business Insider edits in {curly braces}):

    [… T]here’s no evidence that face-based personality predictions are more than a tiny bit accurate. And there are many reasons to fear it might promote bias and other dangerous effects.

    There’s a long history of trying to predict personality from faces. Called physiognomy, it’s been largely refuted.

    “Impressions from faces are very, very low quality evidence,” said Alexander Todorov, a Princeton psychologist who specializes in facial perception. “You’re going to make important decisions, whether someone’s competent or fits the job, based on their appearance? That’s a ridiculous notion.”

    […]

    “Current psychological research shows that by adulthood, personality is mostly influenced by the environment,” wrote Daniel Preotiuc-Pietro, a post-doctoral researcher at the University of Pennsylvania who has worked on predicting personality from profile images, in an email. “While it is potentially possible to predict personality from a photo, this is at best slightly better than chance in the case of humans. I seriously doubt the 80% accuracy for personality {claimed by Faception}.”

    Faception is also on uncharted territory when it comes to predicting behavior, rather than personality, from facial traits. It’s one thing to spot an extrovert and another to spot a terrorist.

    […] CEO Shai Gilboa also tells us it has reached over 90% accuracy on some classifiers. It’s not clear what these claims really mean, however, and it’s impossible to verify them.

    […]

    “[… W]hat are the costs when you make an error?” said Todorov. “It’s stigmatizing and it has a lot of ramifications.”

    In fact, false positives are inevitable with classifiers looking at things like terrorists.

    “Even the most accurate model aimed at a rare outcome will produce a great majority of false positives — extremely rare ‘true positives,’ like being a terrorist, will be hidden among thousands or hundreds of thousands of ‘false positives,'” said Kosinski.

    If you don’t think that’s a problem, imagine how you’d feel if word leaked that you were at at high risk of being a terrorist or pedophile.

    Another risk is that algorithms might fail to disassociate ethnicity, gender, and other factors. Then a model supposedly looking at facial traits could really be a cover for even more controversial kinds of discrimination.

    “The possibility of encoding biases is very real,” wrote Preotiuc-Pietro. […]

  2. SpaceDog says

    I dunno, I’m 90% sure it’s a scam/hoax/’art project’/joke (in that order). Although the people behind it have legit looking histories and connections at first glance.

  3. zenlike says

    Their scientific advisor seems to actually be the Dean of the Faculty of Biology at a seemingly legit Israeli university. How embarrassing for them…

  4. blf says

    I’m 90% sure it’s a scam/hoax/’art project’/joke (in that order).

    How was that accuracy achieved? Use Faception on photos of the people who claim to be associated with it? What does that “90%” even mean? Nine-out-of-every-ten such photos “profiled” were said to be those of “scammers”?

  5. says

    No, it’s not a scam. Marcus has posted about several different people who are doing the ‘facial recognition’ crap, and it’s all based on those sets (poker player, terrorist, etc.) Sam Harris is in favour of this nonsense.

  6. cartomancer says

    Funnily enough people don’t tend to ask which of us is the evil twin. I spend most of my days chanting at impressionable children in Latin and he spends his evenings beating people up for fun.

    However, it is possible to determine, with 100% accuracy, which of us had a nasty accident with a bin ten years ago, just by looking at the skin on our noses…

  7. blf says

    Funnily enough people don’t tend to ask which of us is the evil twin. I spend most of my days chanting […] in Latin and he spends his evenings beating people up […].

    Makes sense. Latin is enough to turn anyone into an Edward Hyde type.

  8. DonDueed says

    Here’s an experiment I’d like to try. Give them a properly-exposed picture of a fair-skinned, fair-haired person. Then give them one of the same person, several stops underexposed, with hair dyed black.

    Any takers that the algorithm’s output would be identical?

  9. garnetstar says

    This is utter nonsense.
    The morphology of one’s face, which is genetically determined, has far less meaning than the particular expression it’s frozen in when the still photo is taken. As PZ said, incredibly subtle involuntary contractions of sets of muscles, etc., result in conveying emotions and “personality”. So, you’d have to get photos of that person in every conceivable situation, to get more than one expression.

    And, as also said, people are already excellent at this. There was once a quiz, posted, I believe, in the NT Times, in which 50 photos of people’s eye areas only were shown (I mean, a narrow horizontal slice of the eyes and skin around and just above them.) You were then given multiple choices of which emotion that person was feeling, just from that small portion of their face that could be seen (and they used photos of the same models in many pictures, to make it more difficult.)

    I got 48 of the 50 correct. No computer could beat that, because even I cannot tell you how I knew, which tiny features of their faces I was looking at to decide. So how to program a computer to do it? Just hire me.

  10. blf says

    richardelguru, Laughs, thanks, don’t recall ever encountering that before…

  11. Rich Woods says

    We develop proprietary classifiers, each describing a certain personality type or trait such as an Extrovert, a person with High IQ, Professional Poker Player or a Terrorist.

    I’m going to take a wild guess at one of these proprietary classifiers and suggest that having brown skin makes it more likely a person will be classified by the Magic AI-Ball as a terrorist.

  12. says

    Here’s a stumper… Let’s say that there is truth in this, that you can tell all of this about a person based on their Real Actual DNA Facemash. Let’s say that this solution can dig deep into that, and detect all of the people in a crowd that are dangerous monsters. Now we have a security cam photograph taken moments before a terrorist attack. We run our handy profile-o-matic on the picture, only it turns out that this happens to be single moment in many real lives without context. That frowny dude, with the slightly dilated pupils and flaring nostrils! He’s our guy! Except maybe he is a guy who just found out his mother got beat up by someone, and he’s angry for whole other issues. That lady, there! She has the sloping eye socket of a psychopathic monster and a look of disdain that’s obvious! She’s gotta be the one! Except maybe she’s one of the large number of people who are psychopaths that simply don’t go around blowing things up, and at the same time she hates people that talk with their mouth full, and just walked by someone doing EXACTLY THAT.

    Sadly, as a software developer, I think I can reverse-engineer the algorithm:

    if(person..getSkinTone().equals(“white”))
    {
    if(person.inGroup()=false) person.setLikelihood(.05); // catch all loner white people
    if(person.hasMoney()) person.setLikelihood(0)
    }
    else
    {
    if(person.getSkinTone().equals(“yellow”))
    {
    // do nothing
    }
    else
    {
    person.setLikelihood(98.5);
    }
    }

  13. chrislawson says

    I presume then that if a terrorist wants to slip past this test, all they need to do is take up poker.

  14. kurt1 says

    @16 nah it’s probably not an expert system. They use fancy machine learning to first get descriptors from the image input and then map them to some output. There is this great anecdote from the first wave of machine learning where the army wanted to automatically classify photos to determine if there was a tank or a truck behind some trees. The classifier worked! except that it only worked because the lighting was different, the training set consisted of bright photos of trucks and dim photos of tanks. Works the same way here as you say, if the complexion is fairly white everything is fine, if it’s brown it’s a terrorist and even more dark => criminal.

    I wonder where they get the labled data from…

  15. unclefrogy says

    they will sell you their system software for $350 mill and provide cameras and system integrators for an additional $250 mill a year service contract
    what corporation or government entity should be without it.
    uncle frogy

  16. blf says

    Are they saying that all pedophiles wear glasses?

    It’s all made-up, fantasy, to be $old to teh gullible. If they get more sales by asserting green-skinned one-eyed unicycling-riding cyclops means “probable shoplifter”, then wear dark glasses and use roller skates.

  17. says

    The description for “professional poker player” under “Classifiers” was interesting:

    Endowed with a high concentration ability, perseverance and patience. Goal-oriented, analytical, with a dry sense of humor. Silent, devoid of emotion and emotional expression, strict and sharp minded, with high critical perception.

    You mean like Daniel Negreanu, who is well known for how talkative he tends to be? Or Phil Laak, who isn’t exactly Mr. Silent either? Tony G.? It could sort of fit someone like John Juanda, but he isn’t exactly known as humorous. Or Phil Ivey.

    Of course the categories are arbitrary in the first place. What is the dividing line between a poker pro and an amateur? Some Cuban exiles would be upset if you called Orlando Bosch a terrorist, despite the fact he took part in acts like the 1976 bombing of Cubana de Aviacion Flight 455, which killed 73 people. They have “High IQ” as a category, despite the fact that anyone who falls under their other categories might have a high IQ.

  18. komarov says

    Our face is a reflection of our DNA​

    As are our forearms, which is why we at the Lovely Biometrics Corporation are hard at work developing forearm-based identification and classification systems to improve your security. Important classifiers include a propensity to wear jewelry on one’s arms and an affinity for tattoos.

    We are also engaged in various non-profit and charitable projects. For example we are currently petitioning governments all over the world to include biometric forearm photographs in their passports to improve overall security and prevent identity theft.

  19. Bruce says

    If there are only two reasonable options at each gene site and if there are only 5 genes that affect the face, since 2^5=32, there are only 32 possible types of personality. That’s almost 3 times as good as astrology. Too bad thrice nothing is still nothing.

  20. chrislawson says

    Logical error #261:

    “Twice as likely to share personality traits” still doesn’t mean that genetics is the dominant factor. Essentially any trait with even the faintest hint of heritability should be twice as strongly associated in identical twins compared to non-identical twins if it is inherited in a Mendelian fashion. This says nothing about how strong the heritability is.

    And indeed, looking at the research they themselves quote, “the lowest heritability estimates were for traits in the environment, reproduction and social values domains” at around 30% (https://www.nature.com/articles/ng.3285) and in a smaller study by the same department, for “personality traits these [genetic] effects are smaller, with 34-48% of the variance being explained by genetic differences” (http://dx.doi.org/10.1038/s41380-017-0005-1), which means that most personality differences are not genetically determined even if you accept the findings uncritically.

    Of particular note: around a third of traits did NOT follow this Mendelian doubling-of-association trend, including conduct disorder, which would seem to be one of the traits that might be of interest to a behavioural profiler. As the researchers say, for these traits “gene-mapping studies may yield disappointing results.”

    Logical error #492:

    Since the purported method does not rely on genetic testing, it’s completely irrelevant how heritable the traits are. It’s nothing more than a crappy marketing ploy to appeal to idiotic genetic determinists.

    Logical error #722:

    I looked up Science magazine’s 31 July 2017 issue — the paper they seem to be referring to is http://science.sciencemag.org/content/355/6332/eaal2913, which shows that epigenetic processes help maintain plasticity of structures in mice until late in craniofacial development. Nothing to do with personality traits. Again, this is crappy marketing designed to appeal to people who have heard that epigenetics is the new buzzword even though it has absolutely nothing to do with the product.

  21. says

    To follow up my previous post I should note that the first 3 poker pros I mentioned are white. John Juanda is Chinese Indonesian, while Phil Ivey is African American. And poker pro isn’t a term limited to men either. Something tells me their professional poker player “Classifier” was assembled around a few white male poker players, as having to include data from non-white and female poker players wouldn’t produce a simple, easy to describe result.

  22. kurt1 says

    Whats even worse is, that even if you had the complete genome of people sequenced, it would still be impossible to correctly classify for these targets.

  23. jrkrideau says

    Do these guys even understand what they’re talking about?
    No.

    Daniel Preotiuc-PietroI seriously doubt the 80% accuracy for personality {claimed by Faception}.”
    I think he meant that these guys are rather incompetent idiots who probably have no real idea of what they are trying to do but determined to scam fools.

    @ SpaceDog
    You can be 99% sure it is a scam. But see note to Caine below.

    @ Caine
    Just because other fools are doing the same thing does not make it less of a scam. Admittedly, it could just be that we have more total idiots around than I was expecting and it is not an intentional scam but it is still a scam. There are some wonderfully crazed “medical” treatments out there so these guys might really believe this in the personality testing area.

  24. says

    Physiognomy updated for the modern era. I wonder whether they’re actually using an AI, or whether they’re just using humans to send out a selection of randomised “cold reading” texts to all respondents once the money goes through. (Pseduo-AI is the term being used – presumably it sounds better than “lying” or “con artistry”).

  25. SpaceDog says

    @jrkrideau

    Agreed. I felt to me like a classic fake startup — flashy website, team with history and credentials, genuine value if it works, science that sounds plausible to a layperson, buzzwords. Attract a lot of investment and then disappear with the money — the only thing it’s missing is a crypto-currency tie-in and an ICO. Of course they could believe in what they’re doing in which case they either fail spectacularly (and the investors lose their money) or they sell the system to a bunch of equally gullible people …

    The list of classifies was what made me think it’s a joke, “Pedophile”, “Terrorist”, “Pro Poker Player”, “Brand Promoter” — they just seemed so mismatched. But if those are classifiers based off previous ‘research’ then maybe they’ve just lumped a bunch of tech together into a plausible looking venture capital honey trap …

  26. zetopan says

    Since little toes are also a function of the owner’s DNA … This reminds me of an idiot that I once had as a PhD “physics” teacher in high school. He was certain that graphology was valid and his graphologist had told him to watch out for me after he showed his graphologist a sample of my writing. Both of them were raving morons, with the “physics” teacher also being a religious fanatic who had the actual goal of swaying his students away from science. Many decades later we have a large number of people in the USA who apparently completely agreed with that specific direction.

  27. wcorvi says

    I’m quite certain that their ‘algorithm’ is ‘proprietary’ because it is a random number generator. I was at the county fair many years ago, and a guy was doing ‘computer personality profiles’. It was a CARD SORTER, not a computer. He’d put a deck of punch cards into the hopper, run the sorter, pull out a stack, and hand it to the unsuspecting rube. I called him on it, and he insisted it was a computer. It used cards, after all.

  28. lucifersbike says

    I have been married to an identical twin for 34 years. I can tell the difference between my wife and her sister – my wife is the one I’m married to, and her sister is the one I’m not married to. :)

  29. Petal to the Medal says

    And police departments start using this technology to identify possible suspects in 3…2…1…

  30. garnetstar says

    @Charly #34

    Thanks, that’s the one! Yes, isn’t it interesting that people are so good at that? I, and I’ll bet you as well, never knew I was getting so much information out of so little. I never even knew I was scanning for it.

    Can you tell what you’re looking at, what particular features you’re assesing, to make your judgments? I can’t.

    So, let’s see some computer beat that.

  31. says

    Petal to the Metal@#36:
    And police departments start using this technology to identify possible suspects in 3…2…1…

    England and China are both already doing it. It’s horribly inaccurate, as expected, but gives an excuse to stop and search. Accuracy is not the objective. In China it is being used to identify and track non-Han. You can imagine how Mississippi would use such a system.

    The FBI has a facial database of 25+million faces (they arranged to get drivers license data from all the states) and it is very poorly regulated. [stderr]

  32. jrkrideau says

    @ 31 SpaceDog

    he list of classifies was what made me think it’s a joke, “Pedophile”, “Terrorist”, “Pro Poker Player”, “Brand Promoter” — they just seemed so mismatched.

    Yes, these are so mismatched (f##ing silly? ) that it is clear that either they are completely clueless or scammers. I think I am leaning towards scammers but as I said to Caine, they may just be absolute fools. Every once in a while one sees something like this. I can remember a “personality test” that warned about men wearing red ties.

    Duh, at the time a red tie was worn by almost every serious male Liberal politician in Canada. Unless you were a very committed Conservative, I am not sure the warning made sense.

    It is a bit like a non-physicist setting out bravely to show where Eisenstein got it wrong in the General Theory.

  33. jrkrideau says

    @ 38 Marcus

    I was under the impression that the Chinese and, at least the British police, were using facial recognition software to identity people, not that they were using it for analyzing personality traits. This thing seems to be intended to replace Freudian imagination with a machine. Should be an interesting race to the bottom.

    I do agree that ithe facial recognition seems amazingly inaccurate. That trial in South Wales was worthy of a Monty Python sketch.

  34. says

    Just curious, but is there somewhere that individuals can send a picture of their face to get an analysis, sort of like where you pay a few bucks to have your DNA analyzed? If it didn’t cost much, I’d love to see what kind of horse manure they came up with for me.

  35. blf says

    There is an article in the Grauniad about Michal Kosinski, who is connected with Faception (also see the two-year old article linked-to in @2). Kosinski is also connected with SCL / Cambridge Analytica, and is known to have met Russian prime minister Dmitry Medvedev & others (presumably members of the cabinet, albeit the article suggests some of them may have been operatives (spooks)), I was shocked it was so easy: ​meet the professor who says facial recognition ​​can tell if you’re gay. Whilst that meeting is a side-story, he, among other things, latter tries to get the Grauniad to suppress mentioning it.

  36. kemist, Dark Lord of the Sith says

    Garnetstar @11 :

    I don’t have to program for it. I’ll build a neural network and give it training sets. It will program itself.

    Neural network classifiers can be tweeked somewhat faster when you have a little bit of knowledge about what you’re looking for, but you don’t necessarily need it. My company is starting to dabble in AI for some industrial computer vision problems that cannot be tackled with conventional vision algorithms. Within one summer, one intern already had something that performed slightly better than a human.

    Doesn’t mean these jokers have any leg to stand on, though. Training a neural network to classify something based on a correlation that does not exists / is meaningless is pointless.

  37. says

    Several points:

    (1) Does one’s personality change permanently in the days immediately following a Botox injection?

    (2) Does one’s personality change permanently if photographed using professional-level makeup and/or prosthetics?

    (3) Are these turkeys relying upon still, posed photos in which the subjects know they’re being photographed… or snapshots of natural behavior taken without the subjects’ knowledge?

    (4) If the police/FBI rely upon driver’s license photos, do they control for the states that require one to remove glasses before the photo is taken (often resulting in squinting and other abnormal face-muscle tensing under the scourge of government-contract-cheap fluorescent lighting)? Or a driver’s license photo taken a few days after major surgery when both fluid retention and sudden weight change may be issues? More to the point, how about controlling for the learned behavior of interrogators, which greatly affects even “resting” facial-muscle orientation?

  38. jrkrideau says

    @ 44 Jaws
    Does one’s personality change permanently in the days immediately following a Botox injection?
    Of course. Based on totally unreliable rumours that I am just now inventing, people change from mild mannered vegans to aggressive carnivores within days.

  39. says

    jrkrideau@#40:
    I was under the impression that the Chinese and, at least the British police, were using facial recognition software to identity people, not that they were using it for analyzing personality traits.

    Yes, that’s correct. It is theoretically there to identify people – but with an accuracy rate that is less than 5%, it may as well be trying to analyze their personality, too.

    Sorry if I was unclear. One of the problems with this sort of system is that the databases are all server-side so there is no chance at all of telling what kind of testing is being done on the data-sets. Some goober could add Faception and – other than it being a bit more goofily inaccurate – nobody’d know. This kind of crap is going to get rolled out with no oversight at all. Oddly, I’m glad that the money is going to be wasted because the alternative would mean that it worked.

  40. says

    Jaws@#44:
    (4) If the police/FBI rely upon driver’s license photos, do they control for the states that require one to remove glasses before the photo is taken (often resulting in squinting and other abnormal face-muscle tensing under the scourge of government-contract-cheap fluorescent lighting)?

    They don’t care if it works as long as it can tell non-white people from white people, and thereby provide an excuse for an expanded search. None of this is about producing an actual working system – it’s all about providing pretext. It’s about catching Uighurs and giving them a hard time, in China. And it’s going to be about stopping people who look like Sam Harris at the border of the US. He’s going to get a chance to see what he asked for, in action, within the next decade.

  41. blf says

    Perhaps only loosely-connected, some police forces / goon squads use what are known as “super-recognizers” — individuals with an astonishingly ability to identify people from only a glimpse. As I noted at Mano Singham’s blog here at FtB some years ago, How good are you at recognizing faces?:

    Over at Thoughts of Crys [also at FtB] I excerpted-from and linked-to an article from last year [2015] about the London Met (police force) squad of super-recognizers, and the best performer in that squad.

    As numerous commentators at the time noted, whilst the ability to match faces from only glimpses is astonishing, perhaps the most remarkable ability of these talented people is the ability to remember the names.

    There is a slightly more recent article in The New Yorker, The Detectives Who Never Forget a Face (August-2016), which include these observations, which seems very on-point with respect to the delusional Faception claims:

    One quirk of facial recognition is that, from infancy, we tend to be better at recognizing faces of the ethnicity that we are most frequently exposed to: white people are generally better at recognizing white faces, black people tend to be better at recognizing black faces.

    […]

    The human brain is often less reliable than digital algorithms, but it remains superior at facial recognition. This might no longer seem true in an era in which Facebook prompts users to tag friends in photographs. With its database of two hundred and fifty billion user photos, Facebook has developed a facial-recognition algorithm that is more reliable than those developed for the FBI. But Facebook’s software has conspicuous advantages: people usually post photographs of friends, and this helps Facebook narrow the range of possible matches. The image quality also tends to be high: you’re unlikely to post a grainy shot taken from a bad angle while a person is moving or in shadow.

    “It’s bullshit,” [head of the London Met’s super-recognisers unit, Detective Chief Inspector] Mick Neville said when I asked him about automated facial recognition. “Fantasyland.” At the airport, when a scanner compares your face with your passport photo, Neville explained, “The lighting’s perfect, the angle’s perfect.” By contrast, the average human can recognize a family member from behind. “No computer will ever be able to do that.”

  42. says

    They claim on their “classifers” page that this is the face of a pedophile

    They’re training their networks wrong, as that is clearly Carlos the Jackal.

Leave a Reply