Why would a veteran factory operator help you build the AI that might replace them?

Just read the article about how veteran factory operators have knowledge that can't be captured in any dataset. they can hear a machine failing before any sensor picks it up, stuff like that. I work with manufacturers on AI implementation and honestly the article is spot on, but I think it's missing the harder part of the problem. Everyone in the comments is jumping to how do you capture that tacit knowledge with better instrumentation, labeling loops, operator-in-the-loop design, etc. All valid.