Training Humans, Trevor Paglen and Kate Crawford show an exhibit which looks at how machine learning classifies people, based on the ImageNet dataset. ImageNet was created in 2009 to "map out the entire world of objects."
There are 2833 sub-categories under "person", and some of these are described as "problematic, offensive or bizarre." If you are intrepid enough, you can upload your own photo here.
I had a shot with my author photo, and bizarrely enough, given the theme of the books, it came up with "aviatrix." Even more bizarrely, my husband's photo prompted the label "co-pilot."
In order to dispel my worry that it was all going to be aviation-themed, I uploaded my son's photo.
The one he used for his CV. And the label? "dissimulator, pretender, phoney."
Luckily, the CV (and the photo) got through and he has an apprenticeship. One assumes that the humans were still in control at his place of work (ironically, an aviation engineering company).
But it does all give you pause for thought on who - or what - is sifting through the CVs, and the basis for acceptance or rejection.
Would a roulette wheel perhaps be fairer?
The Three Motivators
1 day ago