In the lucrative world of artificial intelligence, data are king. Model developers like OpenAI have inked multi-million dollar deals with companies such as The Associated Press to get access to hard-to-find data with which to train their AI models. And startups like Amsterdam-based Adaptive have landed eight-figure sums of capital just with the promise of developing another form of training data. Adaptive claims to be working on a product to help AI developers get high-quality feedback on models from humans.

But there’s also a dark underbelly in the training-data industry—thousands of low-paid workers around the world labeling data, like images of animals, to train models on. LLM developers have come under fire for their harsh treatment of these people, which has prompted some of the workers to seek mental-health counseling.

Tech veteran Ahmed Rashad wants to change that. He has started Sapien, a data-labeling startup which he hopes will make the job of labeling less grueling for people. Rashad, a veteran of Amazon, Oracle and McKinsey, knows something about this industry: he spent three years as the head of operations and growth at Scale AI, which provides contractors to AI startups such as OpenAI to produce training data for their models.