I'm not familiar with the process of this particular company but in general that's unlikely to have that much impact. While labelling itself might have an effec

I'm not familiar with the process of this particular company but in general that's unlikely to have that much impact. While labelling itself might have an effect, lots of biases happen earlier, namely in the selection of samples to be labelled in the first place. If you think about bias in ML models that we've seen in recent years for example: if you let people label 980 images of old white people and 20 diverse ones that's still a heavy bias, even if they spot and label the 2% outliers correctly.