Its a cylinder that performs canned transactions in response predefined audible commands.

Accurate language processing is a huge technical achievement but let's not elevate this particular use of the technology to more than it is. It's a clapper with more functions. When a device like this can actually understand the commands or queries its given we can call it something more.

Yes, and a fax machine is just a waffle iron with a phone attached.

If I tell it to play music and it plays music, what more is there to understand? What kind of conversation do you want to have?

That's actually a good example though. I can give it a precise incantation to play a particular album, playlist, channel, or artist and it works (mostly). I can't in general tell it to play some "soothing jazz."

You can tell Google Now, Cortana, or Siri to play jazz. Siri recognizes "smooth jazz" (And Google might, but it froze up on me). Curating moods of music is, so far, kind of a niche thing that is mostly left to humans like Pandora's music genome project, or Apple's Music service.

Actually, you can. I just asked mine to "play some soothing jazz", and it loaded up a jazz playlist.

I guess I picked a bad example :-) On the other hand, you do either have to pick from pretty broad categories or have to go with a specific list that you or someone else has curated. But it's a hard problem.