I've re-read the excerpt, and realize now that I had initially misunderstood the meaning of the sentence "Accuracy remained high (69%) even when controlling for age, gender, and ethnicity."
I realize now that the author means that efforts were undertaken to ensure that the A.I. program compensated for facial features tied to age, gender, or ethnicity.
After all, it would prove little if the program simply determined the melanin content of the individual subjects and decided, on that basis, that a dark-complexioned person was a "liberal."
Rather, efforts were made to filter out the effects of age, gender, and ethnicity, so that characteristic facial features associated with, e.g., Blacks would not be used to automatically sort that person into the "liberal" category.
Indeed, it would be far more interesting to learn about the non-age-related, non-gender-related, and non-ethnicity-related facial features indicative of conservative or liberal leanings.
A firm jaw - and hence conservative?
A dazed expression - and therefore liberal?
Do Blue-Dog Democrats have, e.g., a shifty expression?
Do Antifa supporters have "lifeless eyes, black eyes, like a doll's eyes"?
THAT would indeed be very interesting to discover!
Regards,
Yes, learning what factors other than those assumed by conventional wisdom -age, race, etc-to indicate particular political leanings one way or another would be interesting. It would also be interesting to learn the accuracy rate.