There is enormous variability in how hard a tool is to use correctly, how likely it is to go wrong, and how severe the consequences are. AI has a wide range on all those variables because its use cases vary so widely compared to a hammer.

The use case here is police facial recognition. Not hitting nails. The parent wasn't saying "AI is a liability" with no context.

This is a common kind of "data or nothing" fallacy. Data doesn't reliably capture evidence for the thesis "TSA agents and aviation workers are burning out and ICE is going to make it worse". The part that data is good for hasn't happened yet over a long enough timeline to reflect properly.

If the argument is "deadly accidents are up over the past decade", then yes, of course, we must point to data.

If the argument is, "the aviation industry might be on the verge of a steep decline in availability and/or safety due to recent political/financial problems", then what do you mean "show the data"? That doesn't make sense. It's a concern based on observation, which is fine if it's not presented as a fact.

And if it turns out that a specific accident is due to said forces - what, we don't address those forces, because "data"?

This trope is being worn to the point of absurdity. Yes, people don't like things. All throughout history. Sometimes reasonably, sometimes unreasonably.

X is not Y. It's X.