Algorithms aren’t racist. Your skin is just too dark.
Joy Buolamwini
44926

Great article. I appreciated the thoughtful tone, and acknowledgement that most likely no engineer was consciously trying to make sure only white people could be seen well. But at the same time it’s clear that the problem exists, and that it was probably in part due to biased sample sets, especially on the part of the original designers of camera and sensor technology, who tweaked settings to work well on the data they had (most likely a white engineer’s face...) If the overwhelming majority of the customer base was dark skinned when digital camera tech was pioneered, then I am sure that business requirements would have dictated more existing products to be designed better from day 1. Even today, stuff like HDR and IR imaging is usually reserved for higher end web cams which is a shame.

That said, it’s one thing to point out a problem, and another thing altogether to get it fixed. For better or worse, most solutions in the largely capitalist global economy get fixed via market forces. People need to vote with their wallets for solutions that work better on all types of skin. As far as data sets go for AI, I’m hoping that there now exist comprehensive human feature data sets that balance all skin tones and light conditions equally. Making these widely available would go a long way toward making sure algorithms aren’t racist in the future.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.