Artificial Intelligence Has a ‘Sea of Dudes’ Problem
bloomberg.com
bloomberg.com
Now it is true they trend male. Particularly true if they are not white.
That just pointing this out, yields this knee-jerk reaction yours is very sad.
There is a an opportunity here to make a global society where computers can serve all of humanity in specific needs of everyone instead of just their creators. Let's not squander it.
Swap that out with anything else and you can immediately see the problem. How does "Fashion has a sea of chicks problem" sound?
They're trying to be edgy. It's not like they said it's a sausage fest.
I find it funny that feminists and SJWs who find something wrong in IT, don't want to do their own version of it and would rather complain about it on social networks and blogs instead.
It is like they look for problems, but never solutions. That is why there is a sea of dudes, instead of complaining the sea of dudes look for solutions to the problems.
There's nothing prejudicial about noting a deficiency or proposing a more fully rounded environment.
I hadn't even noticed, probably because I'm not looking
Maybe there are not data sets made by woman cause woman don't take their time to contribute to them? But of course that's men fault... white men fault as usual.
To make a very blunt argument: If you were to write an AI to reason about menstruation, and let a group of women train it then you would get a particularity different outcome than if a group of only men did it.
This argument extends to many other topics on different levels of subtlety.
... there are of course complex reasons why "women don't take their time to contribute", and to just dismiss that as "This PC thing" is a little short-sighted.
You refer to actually running the training phase of machine learning algorithms? You think if you are a woman you get a different result if you run a Linux command?
> "I do absolutely believe that gender has an effect on the types of questions that we ask,"
but I am a bit at loss, if you thought...
> You think if you are a woman you get a different result if you run a Linux command?
...I was actually this stupid why bother replying at all ?
Hitler used to talk a lot about "Jewish physics" vs "Aryan physics" as if there's a difference.
I'm just surprised to see extremist talk like that on HN.
> I'm just surprised to see extremist talk like that on HN.
Except you are the extremist here. If your take-away from me saying that women and men have different experiences about life on this planet is that it's just like what HITLER! said, your argument lacks nuance.
Anyway, yes - Women and men would train an AI differently. To the point that things become "closed-minded, inside-the-box, not-very-good" though? That's what the article claims...right here:
> "From a machine learning perspective, if you don't think about gender inclusiveness, then oftentimes the inferences that get made are biased towards the majority group—in this case, affluent white males,"..."If un-diverse stuff goes in, then closed-minded, inside-the-box, not-very-good results come out,"
(Sounds pretty bitter. Combined with the "sea of dudes" label, I think we can forgive anyone that laughs at this lady without resorting to name-calling.) But where's the actual evidence for this? Because I don't believe it at all. Men can have a feminine side and through empathy can absolutely think like women. There are men who believe they should have been women...and we all start out as females.
(EDIT: Lastly, that you practically responded to every negative opinion here makes it seem like you are thin skinned.)
We need to stop being ok with racism and sexism against anyone. White males included.
I don't see what's so hard about that. And really, as a dude, to be offended because she said "sea of dudes" to mean that the field, which will probably impact the global society in a huge way, is dominated by men, and not as you imply diminish their contribution, is pretty thin skinned.
Sexism or racism are societally useful concepts if you include the power dynamics. Without them all that's left is that someone wasn't very nice to you, which really as a grown up you should be able to deal with.
The power dynamic here is that the software of future is created by a majority of white men, who, overall, were the first and last group who were and remain globally in charge.
So to say "uff so many men!" in an office 75% full of white men, is completely different than saying "uff so many black women!" in that same office. Not necessarily because of the numerical, but because of the powerbalance.
To point that out, does in no way mean that each individual man in that office is a terribly racist sexist! It just means that there are biases in the whole setup.
In the same vein it's not "sexist or racist to use the data provided by the participants",but if you're not careful and you let the biases in your reading of the data, that were shaped by your experience as a human on this planet, affect the way your AI is trained (and it's almost impossible to be unbiased in that sense), then the outcome of what you're trying to do might well be.