If you can get people to treat it like a voice-activated commandline (“tea, Earl Grey, hot; Big Mac combo, medium, Sprite”), then sure, it’s probably doable. Th

If you can get people to treat it like a voice-activated commandline (“tea, Earl Grey, hot; Big Mac combo, medium, Sprite”), then sure, it’s probably doable. This is how people mostly use Alexa, Siri, etc. It’s not the greatest experience though, especially when there’s “state” to manipulate. On the other hand, humans conversations assume a lot of background knowledge. “I’ll have a coffee if it’s fresh; orange juice otherwise” needs facts the system might not even have access to. You need to know a bit of pragmatics to figure out that “a cheeseburger with nothing on it” should still have cheese and a bun and exactly how many drinks and fries are being ordered when someone says “Two cheeseburger happy meals and one chicken nugget. I’ll have crispy chicken sandwich too. Fries and milk for the kids; just a Diet Coke for me.” It’s obviously not impossible to figure this stuff out, but it might require emulating more of a teenager than you might naively imagine.