Twitter: Our AI image-cropping algorithm is biased toward White people, women

Twitter said its AI-powered image-cropping algorithm is slightly biased in favor of White people and women after all, and has taken steps to ditch its reliance on the machine-learning code. The algorithm was used to automatically decide which portion of a photo shared on the social network was best for the screen viewing it. …

[CrackedNoggin]

“One of our conclusions is that not everything on Twitter is a good candidate for an algorithm, and in this case, how to crop an image is a decision best made by people.”

Just imagine all the hours wasted as user trieds to figure out where the UI for controlling the cropping was, and all the frustration that caused as it wasn't there. Needlessly.

[Anonymous Coward]

Algorithms within algorithms

As Google pointed out at this year's I/O: For people of color, photography has not always seen us as we want to be seen, even in some of our own Google products. Google says it's making changes to auto white balance adjustments to bring out natural brown tones.

Because almost all images are auto white balanced, which is itself an algorithm, dark portions of images tend to have less definition. Then Twitter's algorithm kicks in and will see more detail in and tend to focus on light skinned people versus dark skinned people.

[JDPower666]

Re: Algorithms within algorithms

Where is google auto white balancing images???

[Anonymous Coward]

Re: Algorithms within algorithms

In the camera software. If you want to see the difference, go to the picture editor and look for white. It's default is 0 and you can move it back and forth to see the effects. Note that this is not just a Google issue, it dates back to film cameras where white balance could be adjusted in the lab.

[a_yank_lurker]

Subject vs Artificial Idiocy

As an serious amateur photographer I sometimes take a photo of a scene with people in it. They may be facing the camera but they are not focus of the scene. Often landscape photographers will have a person in the photo to provide a sense of depth and scale. In some photos, people may provide a setting for the scene but are otherwise not critical.

So how does artificial idiocy know what the subject of the photo is? Without asking the photographer it is just guessing.

[Khaptain]

Re: Subject vs Artificial Idiocy

The algorithms are based upon narcissism; they presume that if you use Twitter in the first place then automatically they know that all you want to see is yourself in the spotlight..

I would pretty much bet that that is not too far from the truth. It would be interesting too see how many of those pics were selfies...

[a_yank_lurker]

Re: Subject vs Artificial Idiocy

Fair point as I despise Twatter and selfies. Both are for narcissists only. I need to think like a Twatter user not a photographer.

[JDPower666]

Re: Subject vs Artificial Idiocy

I think more than anything you need to move the focus away from that chip on your shoulder

[Anonymous Coward]

"which grabbed the attention of the code"rs.

[P. Lee]

Twitter is shocked to realise that AI learns from the data it is given!

In other Twitter news... GIGO

[ComputerSays_noAbsolutelyNo]

"It’s possible that the problematic crops of women’s chests were caused by logos or images on their clothes, which grabbed the attention of the code."

... women's chests have the tendency of grabbing attention.

[tiggity]

just don't crop

Surely if someone posts an image they want a view to see the whole image (albeit reduced from original size) as "key" part of image may not be "standout", may be off in a corner, or if its a landscape shot then it's all equally relevant