Data crowdsourcing

Instead of hiring a hundred workers into one building, a company puts jobs in an app and lets thousands of strangers claim them. Each person records a few hours. Added together, it becomes a huge pile of training material.

A robot data platform might list a task like ‘stock a shelf with bottles’ with a set number of slots and an hourly rate. You claim it, record it, and get paid only if the footage passes a quality check.

It scales fast and costs little per hour. The weakness is control: quality varies wildly, much of the material is unusable, and the same easy tasks get recorded far too often.