It's important to remember that a lot of things are labeled as AI that are not.

For example, I worked in health insurance claims processing. 10 years ago. For 11 years. We used software to help identify claims that were duplicates. It would compare codes, dates, diagnosis codes. A human reviews potential duplicates that weren't exact matches. And matching information and mismatch information would be displayed for the human to review it.

Often it required a phone call to determine if the claim was a correction or if the date was wrong. Or any number of reasons. Sometimes a human could figure it out though.

Now days, such a program is being called "AI". But it was just an program that helped identify possible duplicates, and then displayed the information a human would need to determine if it was or not. The program also could auto deny claims if certain criteria was met that flagged it for being a "almost certainly" duplicate. The program also could auto decide a claim was not a duplicate.

In that case, the duplicate review prompt didn't even come up for a human to see. It was fully automated.

Most claims actually were processed fully with software. Only a small percentage required human intervention/review.

But yeah that was in place before I started there and was still in place when I left 10 years ago.

We didnt call it AI back then. We called it automated claims processing.

If we only count ai programs like LLMs, "machine learning algorithms". That's a completely different thing. And it's kind of shitty.

In comparison to the claims software I used at the insurance company that was tailored and designed by a human at every step.

The distinction needs made when comparing how useful they are. Because it inflates the value of AI when effective software is being lumped in with it.