News: 1596092230

AI given a 50-50 chance of correlating a piece of information with a gender.

Gets it wrong so much they turn it off.

Pretty much sums up the state of the field to me.

Unfortunately, gender it is not a binary choice, apparently. Therefore, the chances or correlation should be a 40-40-40 distribution. (The additional probability is for the AI to do the right thing and convict itself of sexism and delete the matrix.)

It may no longer be a binary choice, but if your input is 'woman' and the tool indicates a 90%+ chance of the result being male, then your tool is crap.

" we were not sure if it is worth our time and efforts to make a change in existing biased reality "

It seems obvious that the time and effort to correct this monumental cock-up is going to be well beyond your ability. It also seems to me that equating gender with email is beyond stupid.

At least you were intelligent enough to abandon the project.

Look on the bright side - at least they didn't try and predict race!

Yes, with only one race, "Human", they still would have gotten it wrong!

Take census data and official data about legal given-name changes, if available

Choose bounds (how far back do you want to go? only results from living people? only results from people within a certain state?)

Return percentage results of given-names per sex [1] as specified in the official data collected above.

Return clusters of results as appropriate, e.g. by cultural regions: a given-name may be mostly a girl's name in Japan but a boy's name in Finland.

Now, it is not guessing the sex of the person based on their e-mail address, merely on the submitted names.

[1] Given names are based on the apparent sex of a baby, not its gender, until the person bearing the name actively chooses to change it. Sex & gender are not the same, despite American prudishness.

Sex and gender were the same until recently.

Still are in the real world. Just a few noisy woke folks claiming otherwise.

Not true. 1955. A generation before I was born, at least. 1970's, it was commonly-accepted terminology and knowledge. I wasn't even born for most of that. And gender is no more a choice than your sex, and no more clearly defined.

Plus it makes you sound like a twat.

And what you think is your biological sex has been known to be non-binary for a long time.

Biological sex is very much a combination of a chromosome, where certain things attach to that chromosome, whether body-cells have the receptors for the proteins coded by those genes, whether those body-cells actually act upon those receptors, whether particular hormone levels are enough to trigger certain actions in the body and - in the end - all kinds of other factors.

There's no such thing as XX/XY being the sole determinant of your biological sex. For a start, things can attach to XY instead of XX and vice versa, XYY and XXYY and things like that exist, and those two chromosomes are far, far, far from being the only things that modify your physical sexual characteristics, your genitalia, your fertility or any number of other visible, measurable indicators of your biological sex.

XX males and XY females are perfectly possible.

1-2 percent of the world are biologically intersex. That's the same proportion as diabetics, red-heads or epileptics.

If you know a red-head, or a diabetic, chances are you know someone that doesn't fit the standard school biological sex model. And for every 100 people who "think they're male", there's a chance that at least one of them is genetically / chromosomally / hormonally not - even if they have perfectly ordinary and functional genitalia indicating otherwise.

Now, how do you think that affects the *gender* that people have, if the biological binary determination you were taught in biology is actually so wrong?

It's never been as clear-cut as you were taught in school, same as it's never been true that an atom looks like you've been told it does.

Treating it as some kind of modern hippy fad, where people get to choose who they want to be (oh, what a crime!), is really quite a 1950's way of thinking.

It's time to post [1]this link again.

Watch and learn.

[1] https://www.youtube.com/watch?v=W2y7osVtIX8

good link

Think you might have replied one comment further down the chain than you intended...

On the Internet, it still seems that [1]nobody knows you're a dog .

[1] https://en.wikipedia.org/wiki/On_the_Internet,_nobody_knows_you%27re_a_dog#/media/File:Internet_dog.jpg

Except dogs are NOT ALLOWED, not even Spot.

Is Spotify actually run by a dog? That would explain why I keep getting canine-themed recommendations (Hounds of Love, Dogs of War, Black-eyed Dog, Do the Dog, Hound Dog, Puppy Love, Skinny Puppy, Snoop Dogg, WOLFgang Amadeus Mozart... I could go on...).

The tool is being criticised for being rubbish at accurately identifying gender. This may well be true but does it predict better than random? This might have value to advertisers. It would presumably be capable of improvement but can never be 100%.

The second criticism is that it dares to ignore the current agenda that gender doesn't exist and is a purely social construct. Advertisers don't care, they want to sell to anybody with money and if the tool could identify the individual's personal opinions and self-image they'd be willing to pay for that.

I'm dubious that any data available to an advertiser is of any use at all, to be honest.

When was the last time that you got an ad and thought "Well, that's really relevant to me!"? You get IT ads on IT sites and bikini ads on swimwear sites, and then those follow you through the rest of the Internet because of the magic of Google Ads.

At no point do the advertisers seem to know anything about *me* when they advertise to me. If you're an advertiser and you want to sell, say, tampons... are you really going to go to extraordinary lengths to identify every user's gender based on their username, browsing history, IP address history? I don't think you really are. You're going to advertise on sites that women use, and on searches for feminine products.

It's like the loyalty card thing with supermarkets. When they first came out, they were a vital source of information that the supermarkets couldn't collate themselves. Transactions were just transaction and they had no way to link your shopping this week to next week. Now, the supermarkets themselves admit that it's quite useless as the data is readily available and not very useful. They can link all your purchases together, but they can pretty much do that with credit card data anyway.

And what do they find? People buy more sunhats and salads in the summer. What an insight! And it doesn't help at all to predict, say, a pandemic leaving you with a severe toilet paper shortage.

And every time I get through my Tesco's clubcard statement with discount codes, it's literally for the same things as everyone else's, because they're pushing a particular product, not aiming it at me particularly. They're not trying to make me, a tee-totaller, buy whisky but to sell off the whisky they got cheap. So they don't even customise to whether I drink or not - removing or targeting products based on that - and that's an easy thing to determine from my purchase history (because I've not bought alcohol on it all the time I've had it). My ex and I both had cards and for a while, they were registered at different addresses. The offers to us both were identical. And pretty much things like their stock control and inventory programs tell them everything they need to order or push anyway.

All this data is useless, and I very much doubt anyone's really interested in buying something that guesses based on your username, when Google could let you target to people who have actually visited any of billions of gender-focused websites across the globe, or purchased gendered products.

And even then... I'm not quite sure that you could determine with any accuracy if, say, a couple shared a laptop at home.

All this data is valueless nowadays. It's not used for targeting because it doesn't work and isn't accurate enough and, in the end, makes little difference to the success of an advertising campaign.

Think of everything Google knows about you. Now go to YouTube. Do you get a men's razor ad, or a tampon ad? Neither. You get Grammarly. Because it is relevant to everyone. Go to a review of a men's razor, or even a car website. You'll get razor ads. Go to a beauty salon page or a handbag store, you'll get tampon ads.

The data just isn't valuable and isn't used for those kinds of things. Even Tesco's, Sainsbury's etc. say exactly that. There's a reason that ASDA (Walmart) don't even have such a loyalty scheme. It's seen as a waste of money, because the data is worthless.

"And every time I get through my Tesco's clubcard statement with discount codes, it's literally for the same things as everyone else's, because they're pushing a particular product,"

Interesting. The main supermarket I use here in NL, AH, does a mix of generic and targeted promotions. The odd thing is that the targeted promotions are for items I buy regularly, and would buy even if they were not discounted. So that's costing them money. It would make more sense to send me promotions for different, but related items. Guess I just don't have the kind of brain you need to work in a supermarket marketing dept.

All this data is useless

Not if you can flog it to someone believing otherwise.

It seems there's a long history of people trying to find hidden correlations which reveal an undisclosed truth. Some seem reasonable, some complete nonsense, but I guess you never know unless you put it to the test, can figure out what the correlations are.

"This may well be true but does it predict better than random? This might have value to advertisers."

Let's assume, for the sake of argument, that gender-targeting ads works (I'm not so sure about that). Then if you misidentify a significant proportion of users and serve them the wrong content your advertising is likely to be less effective than when you serve neutral or random ads - a lot of folk might be a bit sensitive about having their gender misidentified :).

Though it can be good for a laugh - in so far as adverts register with me I've noticed that I've been served a lot of ads aimed at plus-sized German-speaking ladies who favour drab-coloured clothing. If that targeting is based on my search history a lot of them must also be interested in loadcells, chainsaws and stain-isolating paint. Inspiration for a novel, anyone?

I regularly use the google feedback X to get rid of ads but I still get the same ads

not interested... not interested... seen it... seen it... seen it... seen it... seen it... seen it... SEEN IT AND STILL NOT INTERESTED!!!

(on my mobile, which I rarely use for more than train times)

To demonstrate how garbage the tool was, someone even entered the name of Genderify’s chief operating officer, Arevik Gasparyan, who is female, for the software to analyze. Unfortunately, it predicted with over 91 per cent confidence that she was, in fact, a bloke.

Some more evidence that it's idiotic to blame power at the top for tech failings.

Didn't think "Oooo... try it on my name!"?

Developers too, don't they try out their own toys?

Is this attitude somehow "AI" specific? The tool has done its "deep learning" so it's job done, no point testing it.

It gives a % chance to two decimal places. How much more non-binary do people want.

Binary is just 0 or 1. By definition it can't have Decimal places.

Binary can be any number you want if it can be written as a fraction of two to the power of n. So, no, not just 0 and 1. That would be a binary digit. Yes, technically it can't have decimals, but can have digits after the radix point, ex 101.110 = 5.75 . Some can't be translated like 1/3 can't be written in base ten, 1/5 can't be written in base two.

So we should all go out correct people that argue against a binary gender classification that what they really mean is that gender is not single digit binary, or single bit, or that they should use words and expression they know the meaning of like "either/or".

I think this plan will go over really well.

Great. Now, can they shut down the ad targeting gender AI as well? I get dresses punted at me several times a month before the content abruptly shifts to more male coded things.

“Think how a trans man might feel if targeted by ads for stereotypically gendered female things, or vice versa," Constanza-Chock said. "Or the harm in opportunity cost of not showing employment ads to people based on misgendered assumptions.”

I don't know where to begin. I mean...come on...seriously?

Okay, first of all, seeing advertisement that is not meant for you is how it has been done since the start. I have not yet felt very offended as a bloke by seeing ads for scented soap. Okay, I can see that it is a little different when that is all of what you get for a while until the algorithms figure out your gender, but nah it happens all the time. And why is gender the only demographic that matters. The wast majority of advertisement I am exposed to are not meant for me. I will never buy a car, I don't need a credit card and I am not sure I will subscribe to "We Over 60" just yet.

Second, the problem is not that you are shown the wrong stereotypical advertisement, but that the advertisement are stereotypical. They are stereotypical because it works, and case and point here where the offence is taken not by that they are but that somebody might be shown the wrong one.

Third, ffs, if a job ad is only shown to one gender then the company is breaking the law in most countries and would regardless have PR disaster on their hand. Is Constanza-Chock here seriously saying that if a company won't show me a job ad for a manager position because it thinks I am a woman the problem is the "misgendered assumption" and not the blatant discrimination?

tfl.gov.uk used to insist on 'title' when signing up for their 'weekend closure' emails (lord/lady/sir/professor/doctor, etc) but now just want first/last/email

Lady Postlethwaite

First world problems, some woke twat screeching "someone" may see the "wrong" advert, ffs.

The article could have been half the length and we could have just all taken the piss of how shit the ML is, without the SJW crap. /rant

So why did they do this ?

The only reason I can think of is that they want to use stereotypes in one form (email addresses etc) to estimate how to use stereotypes in another (the types of advert that is intended to appeal to stereotypical men and women). Seems like it's truncating precision as far as it can by reducing to binary gender, then extrapolating from that to choose an advertising strand.

It difficult to think of a less useful thing to do.

A more useful thing that doesn't pre-assume the results might be to attempt to link the input data with the desired outcome : If prettypolly@gmail.com actually BUYS dresses, that's a useful result that doesn't just throw data away,

Sasha Constanza-Chock's worries about binary gender identifiers seem spurious...

These tools are run without us knowing about how we are being targeted. How would a trans man know they were targeted by ads for "stereotypically gendered female things" (ignoring the insensitivity of stereotyping women) any more than I do? And why would it affect how they "feel"?

If seeing ads that aren't gender suitable make you feel bad then how can anyone tolerate watching commercial tv?

As for not showing employment ads to people on misgendered assumptions... how is that worse than not showing employment ads on correct gender assumptions? Isn't it illegal to exclude on the basis of gender, regardless?

The problem lies with the idea of stereotyping anyone, regardless of gender, sex, sexual orientation, colour, race, age, ... Which is precisely what this pattern recognition system has been developed to do.

If I want someone to do a job, (I am currently seeking to get several tasks done), that is what I need. I would not care if they were a bright green Martian (though I realise the neighbours might worry), but so what, as long as the job was done correctly.

If I am looking to buy something, it might be for me, my wife, one of my daughters, or grandchildren, or even some other relation, what I need are suitable examples of what I want, not some second rate pseudo intelligence pointing me another way. However, if I randomly think I want to buy something but I do not know what or why, then perhaps I should stop and think a bit more about what I am doing.

In short, it was a dumb, stupid pandering to silliness idea in the first place.