Is her only argument that AI relies on third world labor to label training data? She's ignoring that most Big Tech companies already have more than enough of their own customer's data, or can use any of the many standard datasets.