Looking Outwards 11-Racial Bias in AI

The article AI & Creativity: Addressing Racial Bias in Computer Graphics discusses a topic brought up by Theodore Kim at SIGGRAPH 2021. Kim asserts that algorithms used for human computer generated images have encoded racial biases that reflect in the final outcome, and are being propagated in more forms of CG AI – a dangerous seed in spreading racial bias within society. Specifically, Kim talks about the subsurface light transport feature that CG artists use to give a realistic glow to their characters. While it reflects well in white figures, it produces inaccurate results when translated to darker skinned ones, as they have an observed shine to their skin surface. The same can be applied with the variety in hair textures, and the accuracy with portraying different hairs/defaulting to blonde hair. The issue with this is that algorithms are programmed to more accurately depict lighter skinned individuals compared to darker ones, creating a bias for CG artists to choose the first when doing these projects. Kim’s call to action is for these artists to be more conscious of their encoded racial biases, and make the effort to develop diversity with their own works. It was especially interesting when the comparison was brought up in regards to fine arts, and how artists had to “learn” to paint darker skinned people as there was a lack of exposure during their education.

https://nettricegaskins.medium.com/ai-creativity-addressing-racial-bias-in-computer-graphics-f5fc0c255e7