Apparent racial bias found in Twitter photo AI algorithm

Status
Not open for further replies.

erek

[H]F Junkie
Joined
Dec 19, 2005
Messages
10,875
"“At the end of the day, if I do this extensive experimentation… what if it only serves to embolden apologists and people who are coming up with pseudo intellectual excuses and appropriating the 40:52 ratio as proof of the fact that it’s not racist? What if it further emboldens that argument? That would be exactly contrary to what I aspire to do. That’s my worst fear,” he said.

Twitter chief design officer Dantley Davis said in a tweet this weekend that Twitter should stop cropping images altogether. VentureBeat asked a Twitter spokesperson about potentially getting rid of image cropping in Twitter timelines, ethical questions surrounding the use of saliency algorithms, and what datasets were used to train its saliency algorithm. A spokesperson declined to respond to those questions, but said that Twitter employees are aware people want more control in image cropping and are considering a number of options.

Updated 10:09 a.m. Sept. 21 to include responses from Twitter and Vinay Prabhu."

https://venturebeat.com/2020/09/20/apparent-racial-bias-found-in-twitter-photo-algorithm/
 
I'd believe it, most AI based facial recognition data sets have been shown to be racially skewed just due to the sample sets they use for training containing like 80% white people because that's what there is in the stock photo libraries.
 
  • Like
Reactions: erek
like this
Well AI is only as good as the data we feed it. Given that this is all programmed by a human to begin with, those biases will be there inherently.

I'm neither surprised not offended. Not something to get all worked up over in my mind, just something to fix.
 
See the bias even in micro-processors?

https://twitter.com/IanCutress/status/1308198036783460354

1600736706326.png


1600737522761.png
 
Last edited:
I think they made a short term fix. Now when I tried it shows the bottom image irrespective of what is in top or bottom, lol.
 
  • Like
Reactions: erek
like this
I'd believe it, most AI based facial recognition data sets have been shown to be racially skewed just due to the sample sets they use for training containing like 80% white people because that's what there is in the stock photo libraries.
Or its the same simple problem that has existed for years and years, dark things show less contrast in pictures than light things. Shadows that highlight features are often simply lost, or barely show up, human or otherwise. Ever try to photograph a black cat compared to a white cat? Unless the lighting is great, often the black cat looks like a dark blob with two eyeballs, losing the details of the features because less light is reflected off the surface.

Invisible.PNG
 
Last edited:
I'd believe it, most AI based facial recognition data sets have been shown to be racially skewed just due to the sample sets they use for training containing like 80% white people because that's what there is in the stock photo libraries.
Well, the US is 76% white, so having around 80% in the training is pretty close to reality. So we should skew the training set to make sure it's un-bias? Good ole equal opportunity for AI :). But, what about hispanics and asians, and native americans or (insert 100 other minorities here). AI is going to be something that's going to be very difficult to get right and even when it's "right" it may not agree with how everyone actually thinks due to differences in reality (as right/wrong are made up constructs not based in any reality).
 
If the lawless don't like the cops now.. they are gonna hate Robocop..
 
  • Like
Reactions: erek
like this
Well AI is only as good as the data we feed it. Given that this is all programmed by a human to begin with, those biases will be there inherently.

I'm neither surprised not offended. Not something to get all worked up over in my mind, just something to fix.

I guess you didn’t get the memo. It’s 2020, we’re supposed to get worked up about everything, no matter how minor and/or easily corrected.
 
Or its the same simple problem that has existed for years and years, dark things show less contrast in pictures than light things. Shadows that highlight features are often simply lost, or barely show up, human or otherwise. Ever try to photograph a black cat compared to a white cat? Unless the lighting is great, often the black cat looks like a dark blob with two eyeballs, losing the details of the features because less light is reflected off the surface.

View attachment 281602

Back when he had a show, Arsenio Hall used to joke with his lighting crew about that as they'd really have to crank them up, which would make his pinstripe suits 'twinkle' when he moved. HDR cameras have helped a lot with this in recent years.
 
"“At the end of the day, if I do this extensive experimentation… what if it only serves to embolden apologists and people who are coming up with pseudo intellectual excuses and appropriating the 40:52 ratio as proof of the fact that it’s not racist? What if it further emboldens that argument? That would be exactly contrary to what I aspire to do. That’s my worst fear,” he said.

Twitter chief design officer Dantley Davis said in a tweet this weekend that Twitter should stop cropping images altogether. VentureBeat asked a Twitter spokesperson about potentially getting rid of image cropping in Twitter timelines, ethical questions surrounding the use of saliency algorithms, and what datasets were used to train its saliency algorithm. A spokesperson declined to respond to those questions, but said that Twitter employees are aware people want more control in image cropping and are considering a number of options.

Updated 10:09 a.m. Sept. 21 to include responses from Twitter and Vinay Prabhu."

https://venturebeat.com/2020/09/20/apparent-racial-bias-found-in-twitter-photo-algorithm/
It's the band-aid problem.
 
Status
Not open for further replies.
Back
Top