Google dev apologizes after Photos app tags black people as


Recommended Posts

I am white, I have a similar flesh tone to a pig but I wouldn't get mis-identified as one.

Perhaps that is the bigger problem here, not that it picked similar colours but that it mixed a human with an animal.

well human are still animals, sooo :D

Link to comment
Share on other sites

This kind of tech is very new, and like some of the comments mentioned, not fully understood. Errors are prone to happen, if it's possible, it will happen, and nothing can be planned perfectly because unexpected things happen. Apologize and move on, it's not like the code is racist or anything.

Link to comment
Share on other sites

Judging by these pictures, I can honestly believe why the algorithm tags these as gorillas.

Link to comment
Share on other sites

how would one interpet this?

 

 

also, i see it as a programming/coding mistake I guess, but it just comes  to show, like neofuse says, it will happen, just for the sake of it happening :/

(How I interpret it) There is a lot more to that verse before and after it and it goes on and on basically saying, nothing is new even when it's new to you it has been done before and the novelty will always wear off. You always want something new and then that thing becomes old and you want another new one and it gets old etc, etc.

  • Like 1
Link to comment
Share on other sites

Judging by these pictures, I can honestly believe why the algorithm tags these as gorillas.

I disagree. It's really not hard.

 

If face is black and covered entirely in hair = gorilla

If face is black and not covered entirely in hair = human

 

Gorillas have rather distinctive noses and they tend to be pictured against plant filled backdrops, factors that can help avoid mistakes in recognition.

 

Sure it was an honest mistake but tech companies need to be aware of the potential problems with such a technology. It's not like they weren't aware of this potential issue, as Flickr had the same issue a few months ago. Was it racist? Obviously not, but companies can't afford PR disasters like this when rolling out immature technology.

Link to comment
Share on other sites

Why should a technical glitch be associated with racism?

 

I remember something similar happening in college.

 

One of our assignments was to design a neural network based security system that could diferentiate between strangers and a group of authorized people. Took us about 8 weeks of trial and error trainining sessions to teach it to recognize a human face AND separate strangers from authorized people to an acceptable degree.

 

And it worked most of the time, then we showed it several photos of dogs faces and it matched some of them with people :laugh:

 

Nobody really understands how these things actually work, even training them is mostly a trial and error process with random chances of success.

Link to comment
Share on other sites

I disagree. It's really not hard.

 

If face is black and covered entirely in hair = gorilla

If face is black and not covered entirely in hair = human

 

Gorillas have rather distinctive noses and they tend to be pictured against plant filled backdrops, factors that can help avoid mistakes in recognition.

 

Sure it was an honest mistake but tech companies need to be aware of the potential problems with such a technology. It's not like they weren't aware of this potential issue, as Flickr had the same issue a few months ago. Was it racist? Obviously not, but companies can't afford PR disasters like this when rolling out immature technology.

Gorillas aren't 100% hair.. and she kind of looks like a gorilla.. I mean I know no matter how I say it it will come off as being racist.. because I am white.. but my thought when I saw her face... was she kinda looks like a gorilla...

Link to comment
Share on other sites

All technology has limitations, and most modern software has some bugs. This isn't racism it's just bad coding, but I'm sure the usual media outlets will make a mountain out of this molehill.

Link to comment
Share on other sites

Gorillas aren't 100% hair.. and she kind of looks like a gorilla.. I mean I know no matter how I say it it will come off as being racist.. because I am white.. but my thought when I saw her face... was she kinda looks like a gorilla...

I agree with what you are saying... eg. whatever you say will come out as racist; but:

CIoZEu7VAAAFi-X.png

 

Doesn't look much like this:

 

Gorillas_7.31.2012_Our_closest_cousins_H

gorilla_thumb.jpg?itok=Ad2-ATF0

 

Link to comment
Share on other sites

Gorillas aren't 100% hair.. and she kind of looks like a gorilla.. I mean I know no matter how I say it it will come off as being racist.. because I am white.. but my thought when I saw her face... was she kinda looks like a gorilla...

 

I agree. Many people get compared to dogs and there is never any problem.

  • Like 1
Link to comment
Share on other sites

Google's filter doesn't take evolution into account at present. They're working on that. We all looked like apes 1,000,000,000,000,000,000 years ago. Yes really, the Earth is that old. /s

Link to comment
Share on other sites

I disagree. It's really not hard.

 

If face is black and covered entirely in hair = gorilla

If face is black and not covered entirely in hair = human

 

Gorillas have rather distinctive noses and they tend to be pictured against plant filled backdrops, factors that can help avoid mistakes in recognition.

 

Sure it was an honest mistake but tech companies need to be aware of the potential problems with such a technology. It's not like they weren't aware of this potential issue, as Flickr had the same issue a few months ago. Was it racist? Obviously not, but companies can't afford PR disasters like this when rolling out immature technology.

planet_of_the_apes.jpg?itok=JF-tkrbo

 

michael-jackson-12.jpg

 

Explain.

 

edit: no, i'm not trying to be a racist; algoritms and neural matrix are way complexes and unfortunately, mistakes can happen.

Link to comment
Share on other sites

planet_of_the_apes.jpg?itok=JF-tkrbo

 

michael-jackson-12.jpg

 

Explain.

 

edit: no, i'm not trying to be a racist; algoritms and neural matrix are way complexes and unfortunately, mistakes can happen.

Wait, are you calling the king of pop a gorilla? :angry::rofl:

  • Like 1
Link to comment
Share on other sites

Gorillas aren't 100% hair.. and she kind of looks like a gorilla.. I mean I know no matter how I say it it will come off as being racist.. because I am white.. but my thought when I saw her face... was she kinda looks like a gorilla...

 

I was thinking it, but I wasn't going to say it.

Link to comment
Share on other sites

Explain.

 

edit: no, i'm not trying to be a racist; algoritms and neural matrix are way complexes and unfortunately, mistakes can happen.

Are you seriously comparing a fictional ape from a movie to Michael Jackson (who in that picture was white and had had a lot of plastic surgery)? :huh: If you're trying to be offensive you certainly succeeded.

  • Like 1
Link to comment
Share on other sites

Are you seriously comparing a fictional ape from a movie to Michael Jackson (who in that picture was white and had had a lot of plastic surgery)? :huh: If you're trying to be offensive you certainly succeeded.

 

No, i'm showing that an ape isn't just "all facial hair" and, in that case, a really well know human could be mistaken the same way that couple was mistaken.

 

Automatically identifying photos and tagging them is incredible difficult, even with hi resolution images, so the difficulty increases exponentially when the photos are in bad shape or low quality.

Link to comment
Share on other sites

I am white, I have a similar flesh tone to a pig but I wouldn't get mis-identified as one.

Perhaps that is the bigger problem here, not that it picked similar colours but that it mixed a human with an animal.

 

Do humans (regardless of skin tone) look more like a pig or a gorilla?

Link to comment
Share on other sites

how would one interpet this?

 

 

also, i see it as a programming/coding mistake I guess, but it just comes  to show, like neofuse says, it will happen, just for the sake of it happening :/

I interpret it as everthing has already been made, we just use what is made to make other things. So nothings new, only the way one would use it.

 

 

As for this post, it's an algorythm that compares faces of all species. If it matches up with gorilla images it's going to see gorilla images. Algorythms dont build based on emotion, it's not trying to hurt feelings, it just computes. I can see how this isnt intentional from the programmer, the algorythm just needs to get better, as they may not have tested it against people with this complextion or facial structure.

Link to comment
Share on other sites

No, i'm showing that an ape isn't just "all facial hair" and, in that case, a really well know human could be mistaken the same way that couple was mistaken.

You did nothing of the sort. You provided a picture of an ape

Link to comment
Share on other sites

I'm not buyin the lame excuses being made for Google and algorithms they came up with.

 

Was Google intentionally trying to be hurtful, probably not.  But they hurt the feelings of many rather it was an accident or not.

 

Being associated with 'gorillas' is a bit familiar to most African Americans, and very much hurtful.

 

I'm pretty sure a company like Google has more than enough $$$ and resources to make sure that there was proper QA being done, and situations like this never making out of test environment and into the wild.

 

Its not like Black people are new to the planet Earth, Google should immediately recognize that all people come in all shades of black, white, Hispanic, Asian, Etc.  But they probably made sure all other races were nailed down in swifter fashion. I could be wrong though.

 

Edit: I must admit this is good in Goog's to get ahead of this and apologize. It seemed sincere for the most part.

But still hits a little close to home.  

Link to comment
Share on other sites

The problem is they are using neural networks to identify things, and how exactly neural networks work are still a bit of a mystery.  I remember studying these things in Artificial Intelligence class back in the early 90s when they were new.  The neural matrix would associate things they didn't expect.  Example: Teach the matrix to learn to recognize a tank, so you show it 500 pictures of tanks.  Then you ask it, is this picture a tank?  Sometimes it wouldn't work.  They realized that the neural matrix was associating tanks with a sunny sky, and it couldn't tell spot one when it was raining in the picture.  And it seems after 20 years they can't don't know exactly what the matrix is learning.

 

That's not exactly true anymore. They do have ways to visualize what a convolutional neural network (which is most certainly what they are using) is learning. There is a lot of research going on over at Google to do just that. This article is evidence of that: http://www.iflscience.com/technology/artificial-intelligence-dreams

 

 I think the most likely problem is that there are thousands of categories they have trained their network to recognize... and they haven't checked what every single category is actually learning.

 

A similar story to your tanks recognition story... researchers looking at what their network was learning when training to recognize boats found out that certain parts of the network were being used to recognize water. Show it a picture of a boat out of water, and it will have a bit more difficulty recognizing it. If so, add more pictures to your training data of boats out of water. Similarly, a network trained to recognize dumbbells might think a muscular arm must be attached to it.

 

FBqjiv5.png

Link to comment
Share on other sites

  • 3 weeks later...

I'm not buyin the lame excuses being made for Google and algorithms they came up with.

 

Was Google intentionally trying to be hurtful, probably not.  But they hurt the feelings of many rather it was an accident or not.

 

Being associated with 'gorillas' is a bit familiar to most African Americans, and very much hurtful.

 

I'm pretty sure a company like Google has more than enough $$$ and resources to make sure that there was proper QA being done, and situations like this never making out of test environment and into the wild.

 

Its not like Black people are new to the planet Earth, Google should immediately recognize that all people come in all shades of black, white, Hispanic, Asian, Etc.  But they probably made sure all other races were nailed down in swifter fashion. I could be wrong though.

 

Edit: I must admit this is good in Goog's to get ahead of this and apologize. It seemed sincere for the most part.

But still hits a little close to home.  

 

You have absolutely no idea of how facial recognition works, do you?  It's quite possible for the algorithms to correctly ID a subject millions of times, and then suddenly, thanks to some random conditions just as lighting or contrast, get it wrong. That pic has a pretty dark skinned woman and a very bright background, that's a huge amount of contrast difference to deal with and the algorithm just got it wrong.  It happens. Computers aren't perfect.

Link to comment
Share on other sites

This topic is now closed to further replies.