How to use Human in the loop systems to kick start Data Centric AI?

“Use multiple labellers” was a tip that Andrew mentioned. What capabilities exist to expand our labelling strength and how to reduce bias by using a large enough labeller population?