That's because the writer is trying to make it sound that way to get people to click the headline.
In reality, the 1,000 people were mostly just tagging the videos to help train future iterations of the model. It would have taken way more than only 1,000 people to actually run the technology as a mechanical turk.
Agree completely, this is how every machine learning system is built! Maybe this one was taking too long, but that's exactly what beta's are meant to prove.
To be fair, this is a classical strategi in startups called "wizard of Oz prototyping" - it is used to test if there is a market for something before the tech is ready. But the tech is supposed to be created soon after and actually work...