Gender Test

[andrewducker]

I'm curious:
Go here and spend 30 seconds choosing some photos. Then come back and fill in the poll...
[Poll #1471192]

Comments

[feorag.livejournal.com]

It crashed after the first picture!

[andrewducker]

How odd. Working fine for me in FF on Windows. Seems odd that it'd be doing something so complex that a Mac would have problems.

[meihua.livejournal.com]

Hurrah for reinforcing the gender binary. Extra hurrah for:

- Men like construction
- Women like babies
- Men like pictures of women (because all men are straight)
- Women prefer pictures of people, men prefer pictures of things

What utter tosh.

[cartesiandaemon.livejournal.com]

Male -> 86% female. I would have *thought* it would work fairly well even if it was quite simple, but apparently not.

[andrewducker]

It presents a bunch of pictures. It then makes a guess at what gender you are. It then alters itself depending on whether it was right or not. It's a very simple bit of AI code. If it reinforces anything, it's what its users tell it...

[snowking.livejournal.com]

"there is a 50% chance that you are male".

Um. Riiight.

[slemslempike.livejournal.com]

(via friendsfriends)

It suggested 56% male for me. This is possibly because I don't like children very much.

[andrewducker]

How much more conclusive do you want?

[anton-p-nym.livejournal.com]

The quiz was 84% certain that I had outside plumbing, which was correct.

-- Steve's surprised at the number of misses the poll's reporting on males, though given some of the selections of photos maybe he shouldn't be.

[meihua.livejournal.com]

... except I would bet you money that it has the following assumptions coded into it:

1. There are only two genders. Male and female. Well, that one's obvious from the name, but still. It's not accurate. Even if you just go with chromosones, there are more combinations than XX and XY.

2. It won't do any internal clustering within the categories of "male" and "female" for, example, "men who like pictures of people not objects". Which reinforces the mars/venus understanding of gender. As long as there are overall traits towards babies/pictures of women, then those traits drag gender classification towards the two poles.

There's a

[recycled-sales.livejournal.com]

53% chance that I'm female

:(


oh and in the test

[andrewducker]

I'm willing to bet that the system has _no idea_ what's in the pictures. It just knows that picture 37 is chosen by 3% more men than women, while picture 92 is chosen by 2.5% more women than men.

And yes, things are more complex than that - but for (last I checked) less than 1% of the population. Not something that they need to take into account for a mass market bit of fun any more than they need to test on a browser with that market share.

[mooism.livejournal.com]

Male → 96% female.

[autopope.livejournal.com]

It worked for me (different FF plugin configuration, I guess).

Said it was 92% certain I was female.

[headinclouds.livejournal.com]

Seems to think there is a 64% chance I'm male.

[meihua.livejournal.com]

Sorry, yes, that's what I meant. Clustering algorithmns don't know about babies, for example - they just let this kind of application work in situations where the data is XOR'd and you can't draw a simple line.

> And yes, things are more complex than that - but for (last I checked) less than 1% of the population.

... whose daily lives consist of being told all the time that they don't exist. Which times should they object, do you think, and which shouldn't they?

[dougs.livejournal.com]

In IE8 on Vista (for me) it crashes consistently after the first pair of pictures. In Firefox, it crashes after the first pair of pictures half the time, before the first pair of pictures almost half the time, and in the few remaining cases it only shows me one of the two pictures, and then crashes.

[andrewducker]

I knew it!

[andrewducker]

You may feel free to object to anything you like at any point you like. And I'll feel free to respond to it depending on my mood at the time :-> If it had been something more than a silly quiz then I'd have been more sympathetic I suspect - we can find out next time...

I don't expect these things to work well - I was merely curious to see how well it worked on the sample set of my friends list, who are notoriously odd. I was expecting to be more accurate with males than females, which isn't being demonstrated so far.

[andrewducker]

Oh - one of the reasons these things interest me is that they get me right _a lot_ - and I'm fascinated to know why I'm more stereotypical on some axes than I clearly am on others...

[meihua.livejournal.com]

> You may feel free to object to anything you like at any point you like. And I'll feel free to respond to it depending on my mood at the time.

This really winds me up. Yes, you can walk away and enjoy being cisgendered and not have to deal with this shit any time you like. You're really, really fortunate to not have to deal with it. I know you don't mean it this way, but don't rub my face in it, please.

[cybik.livejournal.com]

I did it five times because I don't want to start packing was interested. Three times it thought female (70%+ each time), the other two times it though male (about 65%).

I think the times it thought I was male there were more pictures of children, which I didn't click.

[andrewducker]

I'm not trying to rub your face in anything. I'm just saying that if you open a conversation in my journal then I will engage with it in the manner which I feel is correct. I can't see how it can work any other way - that's how dialogue works surely?

[meihua.livejournal.com]

I just told you that the way you expressed yourself made me feel as if you were rubbing my face in your cis privilege. Do you want to recognise that I feel that way or not?

I already said, "I know you don't mean it this way." Is that the point?

[andrewducker]

I do recognise that - but I don't understand why, or what way I could engage that would make you happier.