it is definitely part of the problem. historically, film was designed for optimal rendering of light skin, with little consideration given to darker skin. that trend has continued with digital sensors and modern image processing. any photographer or videographer that regularly images black skin is aware of this.
yes, it is also a training set problem. training on real-world data will reproduce racial bias, because racial bias exists in the world. racism in, racism out.
but the more direct cause of this kind of thing is the racism of the police using these tools, the states deploying these tools, and the private enterprises making these tools.
> yes, it is also a training set problem. training on real-world data will reproduce racial bias, because racial bias exists in the world. racism in, racism out.
That assumes that all disparities of this kind are due to racism but the fact remains that the disparities in the racial description of arrests matches that of descriptions given by victims when they report crime. Could an entire nation including black people (who are proportionally more likely to be victims that other races) be in on the conspiracy?
Let's focus on trying to avoid innocent people being victimised, either through crime or misapprehension, than making the assumption of ill intent where there is none.
https://www.npr.org/sections/codeswitch/2014/04/16/303721251...
yes, it is also a training set problem. training on real-world data will reproduce racial bias, because racial bias exists in the world. racism in, racism out.
but the more direct cause of this kind of thing is the racism of the police using these tools, the states deploying these tools, and the private enterprises making these tools.