Is that a result of a skewed training set or are people really hard to tell apart from gorillas if there are no obvious tells like large difference in brightness of different areas of the face?
Skewed training set, very few dark skinned human faces.
Facebook apology as AI labels black men 'primates'
11–20 of 291 posts
Re: Facebook apology as AI labels black men 'primates'
#12Is that a result of a skewed training set or are people really hard to tell apart from gorillas if there are no obvious tells like large difference in brightness of different areas of the face?
Gorillas have black and human-like faces. I think they're just quite similar so AI can more easily confuse them.
Re: Facebook apology as AI labels black men 'primates'
#13Is that a result of a skewed training set or are people really hard to tell apart from gorillas if there are no obvious tells like large difference in brightness of different areas of the face?
the biggest tell is probably that, unless the person is in a zoo, I haven't seen many gorillas at starbucks which says a lot about the state of our alleged human outperforming AI
I don't know. I think I would not have a hard time picking out a gorilla at Starbucks.
And I'd like to see a gorilla in any pose that's really hard (for a human) to differentiate from a person.
The truth is: the recognition algorithm is not very sophisticated after all.
Re: Facebook apology as AI labels black men 'primates'
#14Is that a result of a skewed training set or are people really hard to tell apart from gorillas if there are no obvious tells like large difference in brightness of different areas of the face?
Deep learning, for all its recent glories, still suffers from relatively crude, slow-converging training algorithms compared to other areas of ML and statistics.
Maybe to your typical SGD-type algorithm, working off a dataset filled with mostly light skin toned people, skin tone just looks like a real solid first-order way to distinguish humans and primates, and picking up the black people / primate distinction seems much more marginal and second-order, in terms of impact on the cost function.
If most of the people in the dataset were black, I predict you wouldn't see this.
Re: Facebook apology as AI labels black men 'primates'
#15This happened to both Google Photos and Flickr too. Which makes it an inexcusable mistake to make in 2021 - how are you not testing for this?
Google Photos in 2015: https://www.wired.com/story/when-it-comes-to-gorillas-google...
Flickr in 2015: https://www.independent.co.uk/life-style/gadgets-and-tech/ne...
Re: Facebook apology as AI labels black men 'primates'
#16Maybe the algo or the training set or something else was racist, maybe it wasn't. But if you code something that labels people slurs, you've messed something up. Like, you need to be 99.999999% sure you're not throwing out slurs or your whole project is failing spectacularly. And then you have to apologize to the 0.0000001% , which is still probably like 10 people if half the planet uses your site. How do you get there? I don't know. I guess it'd help if you could be 99.999999% sure you weren't looking at a human face before using another label. Like, bias towards humans in a big big way. Heck, the pre-test probability that your algo is looking at a person is probably much higher than the one from your training set if you're facebook. Or maybe you drop primates from your training set. I guess in that case you'll misidentify some primates as people-- which is kind of the flipside of the same problem technically but oh so much more acceptable.
Re: Facebook apology as AI labels black men 'primates'
#17Is that a result of a skewed training set or are people really hard to tell apart from gorillas if there are no obvious tells like large difference in brightness of different areas of the face?
Gorillas have black and human-like faces. I think they're just quite similar so AI can more easily confuse them.
Human-like really depends on your interpretation. That's a generous reading of what's going on. If you google Gorilla faces, I don't think you would be confused.
The AI is not that smart and these examples show it.
Re: Facebook apology as AI labels black men 'primates'
#18Earlier quoted context omitted.
Gorillas have black and human-like faces. I think they're just quite similar so AI can more easily confuse them.
Human-like really depends on your interpretation. That's a generous reading of what's going on. If you google Gorilla faces, I don't think you would be confused. The AI is not that smart and these examples show it.
Us humans are super good at distinguishing faces. So what's obviously different to us might not be so clear to an AI or another species.
Re: Facebook apology as AI labels black men 'primates'
#19The AI is very honest and innocent, it doesn't know what political correctness is. I've heard stories of parents whose kids would also mislabel a black human as a gorilla.
Re: Facebook apology as AI labels black men 'primates'
#20I mean... It's not wrong, we're all primates.
That's not how language works. We're also all animals but not animals at the same time. Context and culture affect the meaning of words.
Parent's point is not irrelevant though, did the AI label only people of one specific racial background as primates?
The article doesn't really provide any details as it's only a few lines long.
The word "primate" in itself is not a slur but it can be used as a slur, as you said, depending on the context, however not much context was given in the article.