When you have pictures of black people being registered as gorillas, or other primates by AI, then you know you have a serious problem. Google’s solution to this problem was not so much a fix, but an erasure, they simply deleted all mentions or pics of gorillas from Google Pictures and left it there.
The same goes for the AI utilised in self-driving vehicles. The AI systems simply do not recognise black people, therefore if you’re in one of these cars travelling at 60 MPH, it will casually drive through or over the unfortunate black person, because it did not recognise their existence.
Robot and AI researchers are typically educated to solve difficult technical problems, not to consider societal questions about who gets to make robots or how the machines affect society.
How do the programmers get around these racist AI anomalies?
The solution could possibly be in actually programming racism into a robotic system, then programming racism out afterwards, so the AI can learn the distinction. Defining a problem first, then erasing that problem is a possibility, however the very fact of distinction in itself could be termed as racism.
Every human at one time in their life has been racist because they have evaluated a distinction in someone else’s ethnicity. This is a normal human behaviour that has gone on since the beginning of time, especially over primitive tribal societies, but is now deemed as racism and thus unacceptable.
What the people who say they are fighting racism may not quantify in their efforts is that evaluating difference is a human trait that cannot be erased. It is impossible, and would be extremely damaging to human society if attempted. In this respect, maybe tolerance is the key variable, however how does one program tolerance into an AI system? At the moment, AI is at an infantile stage of development, and it may take many decades of technological progress to even come close to a system that is truly workable.
Humans have many flaws, therefore any AI would have to evaluate a human flaw, culturally, because some flaws in some cultures are attributes in other human cultures, and adjust its algorithm in real time, as a sentient self adjusting AI being. Robots will eventually resort to designing other robots, thus bypassing all human interference into their evolution. What happens in that scenario is not pretty for humans, caught up in their petty racism merry-go-round. The robots at that time will just view humans as pieces of ape meat, useless in many ways, and needlessly meddlesome. Bye, bye humans.