Well they are in a sense: you can just do the task yourself. It’s expensive of course, so you can use methods applied to human labelling for ML: _periodically_

Well they are in a sense: you can just do the task yourself. It’s expensive of course, so you can use methods applied to human labelling for ML: _periodically_ injecting tasks with known results and checking how trustworthy the party is, vending the task to multiple parties and aggregating results, blocking parties that make many “mistakes”, etc.