Why not just ask the AI what she meant? LIME - Local Interpretable Model-Agnostic Explanations comes to mind. As far as I do understand the classification was

Why not just ask the AI what she meant? LIME - Local Interpretable Model-Agnostic Explanations comes to mind. As far as I do understand the classification was the tagging 'primate' on a video of people with different color of skin and different facial features. The interpretation, that our Lady AI meant harm to the people with dark skin color with tagging them as primates is a loading of the viewer. By the way,´a great new opportunity for cyber lawyers and machine inquisitors.