They’re simply wrong. Do you think AI was writing papers in 1994, as this chart shows? No — this is just a measure of papers about AI, in the field of AI, but written by humans. A couple of commenters here have linked an article about how a researcher used GPT-3 to write a paper, but that is unrelated to this measure of scientific papers in the fields of AI and machine learning. GPT-3 is, in general, not reliable enough to write scientific papers, and, anyway, it was only created in 2020, so it wouldn’t explain how this chart tracks AI papers in the period from 1994-2020.