This is exactly the problem. An AI classifier without sufficient context is simply a very efficient discrimination machine. We must make sure that AI systems have enough context and protection from human bias before large scale deployment.