[some initial menu nav omit]
  AI: I can look up your order. What is your phone number?
  Me: XXX-XXX-XXXX
  AI: I'm sorry, I don't have an order under that number.

  AI: Would you like to look up another order?
  Me: Can I look up by name?
  AI: Yes, I can look up by number. What is the number?

  Me: Beetlejuice.
  AI: I'm sorry, I didn't understand that.
  Me: Beetlejuice.
  AI: I'm sorry, I didn't understand that.
  Me: Beetlejuice.
  AI: I'm having trouble understanding you. Let me get a human on the line.

My need isn't much more than a basic decision tree, either; that level of "conversation" is sufficient / GPT isn't necessary. But … the decision tree, once it has parsed a human sentence into a correct action … has to actually be able to correctly take that action. (And, e.g., not lie.)

Of course, one I get a human on the line, I come to find out the health provider — whose AI I was having to first grovel to — … did have the order, and had just effectively perma-ignored it, without communicating that. They had blackholed my order "because" the health insurer had declined the service … despite the requested service being allegedly available for free to all Americans. The larger pattern here is that this is the sort of garbage tier service that it seems every industry in the economy is rife with these days, and is what generates probably 105% of my support calls.

And I am tired of every support call service telling me "we are experiencing above average call volume". It is mathematically impossible for the call volume to always be above average. Also, your call volume is "above average" as a result of playing stupid games, and then winning stupid prizes, like the above. There were two alternate universes: 1. the insurer did their job -> problem solved, no support call. 2. the insurer failed to do their job -> you contact the customer & relay the problem -> no support call. But, again, we go for garbage-tier-economy answer of "just ignore the problem?"

"Guaranteed two day shipping" took 8 days. The tracking number tells me the "package is out for delivery, delivery by 5pm" … at 8pm. Mass transit board tells me the 3:00pm train is delayed by 20 minutes. At 4:00pm. The list goes on and on. If you want to resolve anything, talking to a conversational AI is not going to cut it.

As another hilarious example, I was talking to a Google Home. I gave it one query, and the device answered in a feminine voice. I give it the next query, … and it responds with a masculine voice? I ask it what happened there, and of course it can't parse that, and for as much "intelligence" as what people ascribe to these things, it's lost by the third exchange. (The owners had some explanation. That's not the point.)

I haven't often fantasized about having a tool like this when trying to explain abstract or arcane concepts. How amazing would that be.