To me, the use of cheap labor for training suggests the level of importance AI companies place on avoiding social/cultural biases in the model in two ways. 1.

To me, the use of cheap labor for training suggests the level of importance AI companies place on avoiding social/cultural biases in the model in two ways. 1. Companies standards would be much higher if classification was done by engineers paid engineering salaries. And engineers would be directly accountable for model biases instead of bias being excused on the basis of how sausage gets made. 2. Outsourcing classification explicitly means classification is not valuable as a core competency; not an area in which expertise is important; and not an activity deserving of the benefits of direct employment such as financial rewards correlated to business success (e.g. bonuses, raises, options, etc.).