In the case of the thumbnail the description is head and all the women are white, with the exception of the last two who have mulatto complexions. It is though once the neural nets extracted the primary feature which was probably that of woman with a red scarf, most of the images in its corpus, or those the algos were programmed to pull up which matched the description were those of white women. An interesting point i noted was that when I clicked for Visuall Similar images there seemed to be more dark women, but none as dark as here, which makes you wonder whether the original display gave preference to white women. The mixed race women were the last two at the end.
In this case the search algorithm labels what it sees as lip and goes on to display the same white woman about 8 times as matching the lip with the red scarf.
Something's gone totally wonky here. The same woman you see in the result summary appears in about 25 times in the first 30 results. The only woman near enough in complexion to Hodan appears in the first twenty, and its downhill from then on. There some mixed race looking types and some black people, but few close enough in complexion. Google you ARE racist and colorist. Careful, careful, I have Rashida waiting on the sidelines, and she takes no prisoners.