back to article The infamous AI gaydar study was repeated – and, no, code can't tell if you're straight or not just from your face

The controversial study that examined whether or not machine-learning code could determine a person’s sexual orientation just from their face has been retried – and produced eyebrow-raising results. John Leuner, a master’s student studying information technology at South Africa's University of Pretoria, attempted to reproduce …

  1. Semtex451

    The question is.... WHY?!?!

    On a side note, if I had been made to write this article I would be handing in my notice about now.

    1. Semtex451

      Hello? - Are we not good enough at judging books by covers that we need an algorithm for this?

    2. Wandering Reader

      "The question is.... WHY?!?!"

      If this actually worked, and could be corrected for makeup and tweezers, then it would suggest a genetic link to sexuality, which could be investigated further.

      That's valid science.

      1. Charles 9

        But that would have to challenge existing science that employs twin studies. There are already multiple recorded instances where monozygotic (as genetically identical as possible) twins ended up with different sexual orientations as they grew up, even as they grew up in practically-identical environments in the same household. It's already known that fingerprints are epigenetic (which is why monozygotic twins have different fingerprints); the suggestion is that sexual orientation could also be epigenetic, if not something not set at birth and instead shaped during upbringing.

    3. rcp27

      The question is.... WHY

      If somebody claims computers can be pointed at photographs and determine characteristics about that person better than humans can, I'd appreciate knowing about it. Equally if somebody claims to have some clever new computer system that can determine things from photographs that humans looking at those photographs, I'd appreciate someone double-checking to see if the claims are BS or not. Just because the characteristic in question is a bit clickbaity is a bit of a side-issue.

  2. 45RPM Silver badge

    The wonderful thing about the LGBTQ flag is that it’s a rainbow. A spectrum. I’m not convinced that there’s any such thing as 100% straight or 100% gay (well, perhaps an absolute minority). If you’ve ever ‘jokingly’ said if I were gay I’d definitely sleep with (insert name of same gender person) then you’re on the spectrum.

    So given the wonderful range of sexualities that humanity has on offer, I’d be astonished if any computer program could predict whether someone is ‘gay’ or ‘straight’.

    1. Tigra 07

      RE: 45RPM

      The Kinsey Scale is the best we've got.

    2. Mage

      Re: LGBTQ

      Only some of those in that alphabet soup are actually gay. It's a dishonest catchall especially adding those born Intersex on the end as some do. Also there are various kinds of T.

      Indeed whatever the studies are showing, is it actually gayness or based on structure? Is there even one universal definition of "gay"? Is the accuracy claimed based on self identification on the photo posting only? What is the definition used by photo poster and are they honest?

      I'm sceptical as to how reliable the sources are and the results are. Also what age ranges?

      1. 45RPM Silver badge

        Re: LGBTQ

        Hi Mage, yes. I agree. The alphabet soup is a mess - but that wasn’t the point. The point is the flag, a spectrum. Seen from a certain perspective that’s very inclusive. It covers all possibilities - there’s an infinite number of colours in the spectrum (although not, being pedantic, in the flag - which is merely the representation of a spectrum), and there are an infinite number of sexualities.

        Which, let’s face it, makes homophobia, transphobia, misogyny etc etc all the more silly.

        1. Hollerithevo

          Re: LGBTQ

          The groups in 'diversity' are all 'deviant', in that they deviant away from the default setting. The default setting is white straight young-to-middle-aged able-bodied male. Therefore diversity means beign not all of these or one of these in some way. A woman is a deviant man. A person in a wheelchair or blind deviates from able-bodied. Being over 70 is a deviant age. Being gay is deviant from straight. Being a lesbian combines two, as does being disabled male and gay. And LGBTQ+ is merely a catch-all for 'all those deviant sexualities if you aren't straight and male'. They have, as other commentards have said, very little to do with each other. We who are in that group have accepted the default and define ourselves by it.

          The term Man, as in Mankind, is the default. It was set up by white straight young-to-middle-aged able-bodied males, as they are the dominant group in western socity.

          1. Teiwaz

            Re: LGBTQ

            The term Man, as in Mankind, is the default. It was set up by white straight young-to-middle-aged able-bodied males, as they are the dominant group in western socity.

            That's news to me. I think I'm in that group, but I don't feel dominant. I have a red ass, but that's just ezcema.

            unless you actually meant 'socity', and it's a planned town in Nevada, U.S.A managed by Disney corp.

      2. This post has been deleted by its author

        1. Anonymous Coward
          Anonymous Coward

          Re: LGBTQ

          Technically there isn't even an LGBT community, there are lots of different groups of people who like different people, which is one of the things that makes it all so complicated.

          1. Tigra 07

            Re: Voyna i Mor

            Sounds better than Non-Heterosexual Sexuality Spectrum People though...

            1. cream wobbly

              Re: Voyna i Mor

              If you excluded those showing signs of heterosexuality from the spectrum, you'd end up with a set of zero. That's kind of the point of a spectrum...

        2. Anonymous Coward
          Anonymous Coward

          Re: LGBTQ

          People don't work like that.

    3. Nick Kew

      It's society's obsessions that wants to classify sexuality, and then make all kinds of meaningless claims like lifestyle choices, moral equivalence.

      Plato's Symposium makes more sense to me: he shows many kinds of love, all valid, but he doesn't try to claim an equivalence, nor deterministic categorisation of peoples preferences. Though Alcibiades' speech, where he clearly regards it as an adolescent lad's right to be mentored by an older man, wouldn't go down well today.

      What can you tell from a person's face? Evidently something, but the nature of that something is probably too controversial to be meaningfully researched.

      1. Anonymous Coward
        Anonymous Coward

        RE: Nick Kew

        "What can you tell from a person's face"

        How ugly they are?

    4. TRT Silver badge
  3. Anonymous Coward
    Anonymous Coward

    It might work once someone finds a backdoor in the code.

    1. Efer Brick

      Smashed it

  4. This post has been deleted by its author

    1. Paul Crawford Silver badge

      How? By where your thumb has just been?

      1. Tigra 07

        RE: Paul

        Maybe it's just the people i hang out with then. I see people didn't like me pointing out the effectiveness of my Gaydar, even if it is more effective than the one tested by this AI...

        1. Anonymous Coward
          Anonymous Coward

          Re: RE: Paul

          Maybe you didn't notice quite how many constraints you just applied to your initial claim.

          You don't know whether the ticket inspector on your train was gay.

          1. Tigra 07

            Re: Mycho

            "You don't know whether the ticket inspector on your train was gay"

            Neither does this AI. But i'd wager mannerisms, behaviour, dress sense, even speech and choice of words are all better indicators than a photo. An AI to analyse sexuality from a photograph would only ever be abused by repressive nations. Is there another use for it that isn't abusive or dangerous?

  5. Anonymous Coward
    Anonymous Coward


    This is all part of the general problem of poor reproducibility of social science studies, which is apparently causing a bit of a crisis. Lots of experiments turn out to give very different results when conducted by different groups.

    It would seem social sciences are roughly where the physical sciences were in the 18th century. Phlogiston lasted as long as it did because different workers got different weight gains - or loss - on heating of substances due to factors ranging from impure materials through retorts losing weight on heating to researchers actually not realising they needed to capture the smoke.

    1. imanidiot Silver badge

      Re: Reproducibility

      Social "sciences" are currently way too often not actually good or even half decent science. Too much of it is based on wishful thinking or influenced by the researchers own bias.

    2. Nick Kew

      Re: Reproducibility

      This is all part of the general problem of poor reproducibility of social science studies,

      That's what I thought from the headline.

      You need to read beyond that. To a very large degree they did reproduce the results.

      1. Anonymous Coward
        Anonymous Coward

        Re: Reproducibility

        If you read the whole article you will see that the sets of results were rather variable and they disagreed over what the "AI" was actually detecting. In the physical sciences that would be something of a no-no.

      2. phuzz Silver badge

        Re: Reproducibility

        I guess it would be fair to say, they got similar results, but as a result of different conditions (eg bluring the faces) which lead them to utterly different conclusions.

  6. Anonymous Coward
    Anonymous Coward

    Tinpot dictators

    I think most tinpot dictators already understand how to use pseudoscience to serve their political ends.

  7. Simon Harris

    Haven't we been here before?

    This seems to be the 21st century's answer to the 19th century's phrenology.

    1. Mage
      Big Brother

      Re: phrenology and 19th C.

      Some people still believe in that. And in that other 19th invention, "Flat Earth", which seems to have been a satirical invention. Some Evolutionists claimed Creationists believed it. Actually before circumnavigation most people and the Church never thought about it at all. Both ancients and Christian Era experts actually knew the Earth is round and quite a few had good estimates about the size. Even Galileo vs Church wasn't really about science, he put theology in his treatise. Others had more accurately written about Earth & Sun and had Catholic Patrons, or were Catholic and had no problem. It's not as if they tortured him or burnt him at the stake either. He was put under house arrest. He didn't like leaving the villa anyway.

      Why this digression? Because some arguments are not about what they claim to be, but about politics. Maybe the real issue is that Neural Networks, Machine Learning and "AI" is all a bit rubbish. Some people would like big successes where it's much better than people.

      The more controversial and headline grabbing the better!

  8. Anonymous Coward
    Anonymous Coward

    Deep Learning is amazing!

    Just add garbage and nonsense in one side and it produces a paper on the other side!

    Soon: Bring Back Phrenology! Get one of those caliper-like things and start (again) measuring skulls to see who is a criminal or not.

    (/s just in case)

  9. Oliver Mayes

    And how did they determine if the machine was correct or not, were they using the self-reported sexuality that these people had on their profiles? Are they sure that everyone was being honest on them?

    How did he exclude the spambots using random photos to try and get money out of the real users?

    1. vulture65537

      Did they reveal the picture files from the sites to the CNN or did they re-photograph them? There could be digital features such as resolution or EXIF treatment that can be used to distinguish between the various sites in the sample.

    2. Fading

      That was my first thought....

      Did they cleanse the data and remove photos that were not related in anyway to the profile associated with them?

      Were particular declared orientations more/less photogenic?

      Was there a statistical link between "enhanced" photos and declared orientation?

      Does anyone question their own work anymore?

      1. cream wobbly

        Re: That was my first thought....

        They culled all the training data from dating sites, i.e. each photograph was selected by the subject.

        They culled all the testing data from publicity photographs, i.e. each photograph was selected by the subject.

        So, this is as worthless as any other self-reporting study.

  10. Anonymous Coward
    Anonymous Coward

    Anon for obvious reasons

    I've spent quite a bit of time on Adult FriendFinder for many years since it started about 20 years ago and I've met a lot of people - you need to actually experience this because academics doesn't work. Since AFF is an adult site and you end up enjoying "adult" activities with the other sex (at least initially) when you are sitting there with no clothes on conversations are generally quite honest ... most couples that I meet are very happy to start off with M-F sex but very quickly slide over to M-M, F-F, M-F-M and F-M-F sex.

    Are they "gay" or are they "straight"? Fact is, when you get down to it, pleasure comes first, orientation is secondary - most guys enjoy watching and playing with people involved in F-F sex and virtually every woman I've ever met has been very keen on watching and encouraging M-M sex and helping.

    Of course nobody would ever admit this in a survey.

    1. Anonymous Coward
      Anonymous Coward

      Re: Anon for obvious reasons

      I think you have a risk of self-selection bias in your analysis.

      1. Version 1.0 Silver badge

        Re: Anon for obvious reasons

        While I'm not saying that you are wrong, can you name any study of anything that doesn't have a bias of some kind?

      2. AK565

        Re: Anon for obvious reasons

        Would any self selection bias be non-representative to a greater extent than the atudy's sample which appears to have few controls for inacuracy on tje initial sample set?

  11. Paul Hovnanian Silver badge

    I suspect ...

    ... that if any information regarding a persons sexual orientation could be gleaned from a photograph of their face, it would depend heavily on the context in which that picture was taken. Like who the photographer was, for example.

  12. Robert Grant

    “Moreover, this entire line of thought is premised on the idea that there is value to be gained in working out why 'gay face' classifiers might work – value in further describing, defining and setting out the methodology for any tinpot dictator or bigot with a computer who might want to oppress queer people.”

    Scientist doesn't like attempted reproducing of results of papers?

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like