Part of the problem with AI is that it requires significant skill to understand where AI goes wrong.

As a basic example, get a language model like ChatGPT to edit writing. It can go very wrong, removing the wrong words, changing the tone, and making mistakes that an unlearned person does not understand.

This sets up a dangerous scenario where, to diagnose the results, you need to already have a deep understanding. This is in contrast to non-AI language checkers that are simpler to understand.

Moreover as you can imagine the danger is that the people who are making decisions about hiring and restructuring may not understand this issue.

And AI is not always the best solution. One of my tasks at my job is to respond to website reviews. There is a button I can push that will generate an AI review. I’ve tested it. It works… but it’s not personal. My responses directly address things they say, especially if they have issues. Their responses are things like, “thanks for your five-star review! We really appreciate it, blah blah blah.” Like a full paragraph of boilerplate bullshit that never feels like the review is addressed.

You would think responding to reviews properly would be a very basic function an AI could do as well as a human, but at present, no way.