As HN: What efforts are currently being made to overcome racial biases in AI?

I've read a lot about racially biassed AIs recently. Eg the depixelizer that turns Obama white:

https://www.theverge.com/21298762/face-depixelizer-ai-machine-learning-tool-pulse-stylegan-obama-bias

Face detection also works better on white faces.

What work is currently being done to overcome this, if any?

Is this a problem with dataset availability? And if so are there any ongoing efforts to create more diverse datasets?

Would creating a way to crowd source datasets work / be useful to anyone - eg if you could submit images/speech/video of yourself and label them?

Thanks

1 comment

[ 4.5 ms ] story [ 16.4 ms ] thread

trash3 5y ago ↗

capitalism, once the models are cheap enough to develop, the models will be made to work on the whole distribution of users. It's the same story for, pharmaceutical research, lots of drugs that work for white people are ineffective or else for black people.