Google Apologises For Photos App's Racist Blunder

Mark Wilson writes: Google has issued an apology after the automatic tagging feature of its Photos apps labeled a black couple as "gorillas". This is not the first time an algorithm has been found to have caused racial upset. Earlier in the year Flickr came under fire after its system tagged images of concentration camps as sports venues and black people as apes. The company was criticized on social networks after a New York software developer questioned the efficacy of Google's algorithm. Accused of racism, Google said that it was "appalled" by what had happened, branding it as "100% not OK".

Comments

I know how this is going to be fixed...

[CajunArson]

Anything that's politically incorrect will be blacklisted from being labeled as a result. No more gorrillas or any other of a million and one potentially offensive labels!

Although gorillas might be labeled as people, which would actually make some SJWs happy.

Re:I know how this is going to be fixed...

[TsuruchiBrian]

Their algorithm incorrectly identified one member of the Homininae family (gorillas) with another member (humans) sharing at least 95% of the same DNA. Yeah they are fucking idiots. How well does your algorithm work?

Re:I know how this is going to be fixed...

[Hevel-Varik]

No your comment is sad. Go and view the tagged pictures. And tell me that there is no similarity. Your lying or have so deeply fooled yourself that you have difficulty thinking un-good thoughts.

Amazing, modern man believes man descended from apes and when a computer algorithm mistakes a black colored and ape shaped ape descendant for, you know, an ape, it's because the photo alg guys are a bunch of frat boys.

It's sad the world is bending over backwards affecting this total bullshit 'how COULD this happen!!!!,' though really it isn't the world. It's media and blog bubble who are encouraged, nay, validated! by the corporate "This. Is. 100% not. ok" but show the pic and explain the story off-line and see how normal mankind really still is.

alogrithms aren't racist

[Anonymous Coward]

Alogorithms aren't racist, and teaching a computer to visually recognize objects is hard. Move along.

Re:alogrithms aren't racist

[jellomizer]

I followed the link and looked at the photos. I could see how it would make that mistake.
1. The Color balance was off: What we call black people are actually just a richer brown. the color balance gave their color more of a real Black/Gray color, the natural color of a Gorilla.

2. The Angle of the shot. The tilted Angle makes it appeared that they are not upright but slouching in.

3. They were making unnatural facial features for humans. They were making funny faces at the camera.

4. The dark hue of the gentleman who was behind shirt, combined with the ladies hair style, makes it seem the body with much broader shoulder.

I expect the combination of a lot of factors created the wrong choice. But computer decision making, while getting good, isn't perfect, but it is often better then not having it because then it wouldn't be possible to catalog the millions of images. We need to accept that computers make mistakes and there should be a way to fix them when they are found.

Many of our derogatory comments come from the fact that we find similarities with something else, so it come to reason that a computer may make an actual mistake that will reinforce such derogatory meaning.

Re:alogrithms aren't racist

[Fwipp]

But the people writing the algorithm and choosing the input data *can* be racist. And even in the absence of malice, you can create racist outcomes.

If your training set has many photos of white people and few photos of black people, it's not going to be great at recognizing black people. If it doesn't know what black people look like, it's bound to misclassify them more often than white people.

Anecdotally, I noticed that the Microsoft "how old are you" site a while back recognized me (a white person) in every picture, but only detected my (black) partner in about a third of the pictures I fed it. In one instance, in a screenshot of a video chat, it recognized my little 100x100 picture in the bottom right, but failed to detect my partner's face in the center.

Your real-world performance can only be as good as your test/training data.

Re:alogrithms aren't racist

[Whiteox]

OTOH If it was a pic of a gorilla but labelled 'Black Afro-American' then you would have the same issue.

Re:alogrithms aren't racist

[chipschap]

I find it hard to believe that there was racism intended in any way, shape, or form. It is unfortunate that this took place but Google certainly took care of the problem in short order, as is right.

There are too many of the LBTO (looking to be offended) crowd these days. Come on, there are plenty of real problems with racism, there's no need to label inadvertent and unintentional things.

Re:alogrithms aren't racist

[johanw]

The algorithm wasn't that far off. I'm sure the gorilla will come over it.

Re:alogrithms aren't racist

[lq_x_pl]

It isn't a racist outcome. It is the outcome of a flawed algorithm. Might even be able to argue that wider testing (and improvement) is needed for the image sensors for computer-attached video equipment. If I my own photo albums for "seal" or "dog" I get pictures of my kids in both. I don't believe the algorithm is impugning the humanity of my offspring, I just think it is far-from-perfect. The outcomes of my search aren't hateful. The outcomes of the picture labels in this story aren't racist.

Re:alogrithms aren't racist

[Copid]

The developers building vision algorithms don't typically create their own datasets. They purchase archives of images, and a lot of these problems stem from how many samples of each type are in those archives. The Google team likely has a giant database of human faces that it works with, and the ethnic frequencies are probably either the result of choices made by whatever origanization compiled it (and for whatever reason they compiled it) or the ethnic breakdown of the userbase of some app they used to grab the data. It's extremely unlikely that either of those will produce the same number of samples of every ethnic type.

It's also one thing if this was a program just designed to distinguish between different people. But it looks like it's trying to recognize objects of all sorts and distinguish between people and just about everything else. That's a hard problem, and the only response to this sort o thing is to take a regular failure case and feed it back into the training data so you can hit the next regular failure case. Hopefully it will be less coincidentally embarrassing, but it will definitely be there. Perhaps confusing bald men with balloons or something like that.

But I also think people underestimate how much skin color affects machine vision problems. I spent years in the biometrics industry and one consistent fact is that people with darker skin just don't provide as much easy-to-recognize detail as people with lighter skin. There will be more misclassifications as long as the image is taken using the visible spectrum. To a computer extracting features, dark skinned people and gorillas are both human-ish face shapes with a particular color range and somewhat indistinct geometry due to weak contrast and shadows. Distingushing between those two sets just isn't as easy as distinguishing between fair-skinned blondes and gorillas. You can make that decision just by looking at the color histograms and not even bothering with geometry.

Re:alogrithms aren't racist

[ScentCone]

It isn't a racist outcome. It is the outcome of a flawed algorithm.

You're not paying attention. These days, outcomes that have nothing to do with intention, purpose, or simple transparent standards, but which happen to lean statistically towards results not in perfect balance with skin color as a function of population (though, only in one direction) ... the process must be considered racist. The whole "disparate impact" line of thinking is based on this. If you apply a standard (say, physical strength or attention to detail or quick problem solving, whatever) to people applying to work as, say, firefighters ... if (REGARDLESS of the mix of people who apply) you get more white people getting the jobs, then the standards must surely be racist, even if nobody can point to a single feature of those standards that can be identified as such. Outcomes now retro-actively re-invent the character of whoever sets a standard, and finds them to be a racist. Never mind that holding some particular group, based on their skin color, to some LOWER standard is actually racist, and incredibly condescending. But too bad: outcomes dictate racist-ness now, not policies, actions, purpose, motivation, or objective standards.

So, yeah. The algorithm, without having a single "racist" feature to it, can still be considered racist. Because that pleases the Big SJW industry.

It's the same thinking that says black people aren't smart enough to get a free photo ID from their state, and so laws requiring people to prove who they are when they're casting votes for the people who will govern all of us are, of course, labeled as racist by SJW's sitting in their Outrage Seminar meetings. It's hard to believe things have come that far, but they have.

Re:alogrithms aren't racist

[TsuruchiBrian]

We should throw in some white colored apes that google can mistake for white people, and then it will be fair, and everyone can stop being outraged.

It's an algorithm

[guruevi]

It's impressive that it can even recognize and classify things as such. Great apes and humans share about 99% of our DNA, any 'alien' entity would classify us amongst the apes.

The fact that black people are black and thus have a closer resemblance to the generally 'darker' great apes is not racist because an algorithm that is not programmed to have biases cannot be racist. It's just peoples interpretation of the facts that makes things 'racist'. Superficially, black people and apes look mathematically more alike than white people and apes. If the thing was trained on albino apes (which do exist), white people would be considered apes AND NOBODY WOULD THINK IT WAS RACIST.

Re: It's an algorithm

[imgod2u]

One could point out that there are fewer instances of white males being miscategorized. I suspect this has less to do with any actual racism and more to do with the fact that the people who developed the algorithm are likely predominantly white males and they tend to first test the algorithm on their own collection of photos or those in their circle.

This is an argument for a more diverse workforce...

Re: It's an algorithm

[Anonymous Coward]

It's an argument for more QA testing

Re: It's an algorithm

[buybuydandavis]

One could point out that there are fewer instances of white males being miscategorized. I suspect this has less to do with any actual racism and more to do with the fact that the people who developed the algorithm are likely predominantly white males and they tend to first test the algorithm on their own collection of photos or those in their circle.

This is an argument for a more diverse workforce...

Yeah, because I bet that's how Google develops their image recognition algorithms - white guys walk around taking pictures of themselves.

As is more likely the case, there are few pictures of Albino gorillas in their machine learning corpus (racist against Albino gorillas!), and hence less data for white folks to more easily match gorillas based on macro level color characteristics.

Re: It's an algorithm

[sexconker]

Testing, let alone QA testing or MORE testing, is NOT the Google way.

In over news

[oobayly]

Google's algorithm also identified photos of some sunburnt Essex chavettes on the beach as Yorkshire pigs. A Google spokesperson said "no apology if necessary - it's an accurate assessment".

"Software" has no opinions of race.

[pla]

Software doesn't hate black people. Software doesn't dislike Islam. Software doesn't think kids these days need to pull their damned pants up and stop playing that crap music too loudly.

Apologizing for a program miscategorizing an image it has never seen before as somehow "racist" makes about as much sense as GE apologizing because my toaster looks like a frowny-face from just the right angle.

Yes, Virginia, we've taken this shit too far.

Re:"Software" has no opinions of race.

[turp182]

Actually, apologizing makes sense in this case. It's not about being at fault (they are not), but common courtesy.

Early in our marriage, my wife taught me to say "I'm sorry" when those around me were hurt or bothered, to be a nice person. Prior to that I only said it if I was at fault.

So if someone spills soda on their shirt, then "I'm sorry". Same for Google, it is the decent thing to do.

Regarding photo identification, Google should have images of the Confederate flag show up in the category "Racist"...

 

that's right

[Anonymous Coward]

And as Richard Dawkins has said, we ARE apes - all of us humans.

Re:Accepting Responsibility

[bluefoxlucid]

...NO IT ISN'T, YOU ASSFACE!

Let's see, we'll do this completely-innocent thing, which is hard, but helps society. Suddenly, hard thing does some harmless,amusing, not-entirely-predicted thing, and people whine about it. OMG, LET'S LEGITIMIZE THEIR STUPIDITY AS A VALID OPINION!

No, you're admitting fault here for something that is NOT YOUR FAULT. You're admitting bad behavior and bad decisions for something that was good behavior and good decision-making, but produced a bad outcome.

THIS IS WHY WE HAVE SHIT SCHOOL SYSTEMS!!! If we have 60% success rate and improve the school system by broad, visible measures to give a better education and improve to an 85% success rate, 15% OF PEOPLE WILL CRY THAT OUR NEW EDUCATION SYSTEM FUCKED OVER THEIR KIDS! Someone will point to all the failures, create a collage, and claim we're totally incompetent!

The appropriate response to bitchwhining about this non-issue is to tell people to stop fucking whining.

Oh the outrage a non sentient can be racist

[Crashmarik]

What next a hairy fat guy in a pool gets tagged as a walrus ? His girlfriend a whale ?

Oh the horror the machine was mean.

In related news...

[wbr1]

Google announce that it will change the tagging of Caucasians from 'cracker' to 'saltine'.

Re:that's right

[magarity]

"Ape" is a general language term that can be used by different people to specify quite different groups of creatures; it's more correct to say we're all hominids.

Re:Accepting Responsibility

[DigiShaman]

It's part of the cult of PC (Political Correctness) to carry out self-flagellation as a method of atonement.

Re: lol

[Anonymous Coward]

The truth is that there are a ton of entitled idiots who believe they have the right to be offended, and gigaton of idiots who chose to oblige. That's why we don't have nice things anymore.

Re:Casper is Concerned

[eth1]

So, do really pale "white" people get mis-labeled as ghosts? Inquiring minds are somewhat concerned because they are rather pale....

One of the articles I saw about this mentioned that in the past, light-skinned people had been identified as dogs and seals. Strangely, there was no outrage about that.

Gorillas aren't so bad

[Theovon]

I'm not going to downplay the feeling of insult that the black couple experienced. There is a long history of racism against blacks, referring to them as apes and other things, with the intent of putting them down. In *this* case, it was an accident of a flawed algorithm, but there's some history here that makes that a hot button. For the sake of repairing the effects of racism of the past, we should be careful about how we use racial slurs, even accidentally.

That all being said, we're learning more and more about gorillas and other higher apes and how intelligent they are. We're closely related. To an alien from another planet, they may look at humans and other apes and not perceive much difference. To compare humans (in general) to apes (in general) isn't all that unreasonable. And some day, when all this racism shit is behind us, mistakes like what happened here might be merely amusing.

Re:Casper is Concerned

[war4peace]

Searched for "dog" in my Google Photos. 6 photos came up, all of my kids or kid and wife. I don't care. It's an algorithm.
Searched for "seal" in my Google Photos. Only one came up, and it's of my elder kid. I don't care. It's an algorithm.
People who feel "offended" by an algorithm are batshit crazy.

Re: lol

[Vermonter]

You have a right to be offended, just as I have a right to not care.

Prediction

[sootman]

They tweak the algorithm a bit. A week from now, a gorilla in a photo is tagged as 'black person'. Hilarity ensues.

Re:Accepting Responsibility

[bluefoxlucid]

This is kind of like being hit in the arm by a baseball as you are walking by your neighbor's yard.

It's kind of like being hit in the arm by a baseball THAT YOU IMAGINED, BUT WHICH DOESN'T ACTUALLY EXIST, as you are walking by your neighbor's yard.

Re:that's right

[faway]

Richard Dawkins is a biologist. he would never say something so stupid. we are all hominids, and we are certainly not apes.

Re:that's right

[Ramze]

No, "Ape" is a very specific term used to specify members of Hominoidea. It is unfortunate many are ignorant of the meaning of the term and use it improperly to include monkeys.

https://en.wikipedia.org/wiki/...

Humans are apes - specifically, great apes. (aka Hominidae aka "hominids"). "Hominids" simply means human-like. It used to mean only humans, then it included other extinct human-like creatures and now it generally includes all hominidae. While "hominid" (or alternatively "great ape") is a more specific term, it is certainly NOT a more correct term, merely the Family of the SuperFamily.

One could say that humans are mammals and it would be no less correct. Humans are animals, chordates, mammals, primates, apes, and also great apes.

It's unfortunate that the Google facial recognition software was not aware that humans don't like being reminded that they are indeed very closely related to other great apes and could easily be confused with gorillas by a non-human intelligence. Our indignance at the notion we're apes that look a lot like gorillas is rather silly -- like zebras being offended at being miscategorized as ordinary horses.

Granted, I understand the racist implication that those flagged erroneously as gorillas are somehow less human than others. Thankfully, the computer isn't racist. It merely wasn't sophisticated enough to discern the difference given the input, the algorithm, and its training.

I'm impressed it figured out the object in the photo was a living thing and got the kingdom, phylum, class, order, superfamily, family and sub-family correct. If it had chosen chimp or bonobo, it would have been even closer.

Heck, check out this comparison of a gorilla baby and a human baby -- no one would have blinked an eye if the software said the gorilla was a human baby.
http://intentblog.com/wp-conte...

Another cute gorilla baby -- a bit older:
http://www.ctvnews.ca/polopoly...

Mod parent down

[SpaceXXX]

This person is using "blacklisted" word. Mod it down.

Re:Accepting Responsibility

[ScentCone]

It's called an "apology" - did you skip that day in kindergarten?

When the apology is a completely over-wrought bit of silly nonsense rendered in response to gleeful press releases from the Big SJW industry (who desperately NEED there to be events like this, whipped hugely out of proportion, in order to have things to get sound angry about), then it's not an apology. It's a forced sacrifice on the alter of Political Correctness gone (ever more) insane. There's nothing to apologize for here, because nobody at Google sat down to create a racist process or racist results. People who can't mentally untangle the difference between intent and coincidence should just shut up ... except, they're all media darlings now, because it's fashionable to be completely irrational on that front, now.

If Google tagged me as "albino ape" or "yeti" or "Stay-Pufft Marshmallow Man" I'd think it was hilarious. Those manufacturing faux offense at this bit of completely benign nonsense are the real racists. They are the ones who are saying that black people aren't smart enough to understand the situation. As usual, the racist SJW condescension is the most actually offensive thing in the room.

Re:that's right

[Barefoot Monkey]

Richard Dawkins is a biologist. he would never say something so stupid.

I'm curious what you feel is stupid about that straightforward statement. Regardless, Richard Dawkins did, in fact, say exactly that.
Gaps in the Mind, by Richard Dawkins
"We admit that we are like apes, but we seldom realise that we are apes."
"In truth, not only are we apes, we are African apes. The category 'African apes', if you don't arbitrarily exclude humans, is a natural one"
"'Great apes', too, is a natural category only so long as it includes humans. We are great apes."

I did a search for the words "dawkins" and "ape" and the first result was a video of Dawkins saying that he is an ape. I challenge you to find any living biologist that claims otherwise.

we are all hominids, and we are certainly not apes.

Gorillas are hominids, and all hominids are apes. Humans are apes and hominids, just like gorillas.