Written by 
Prompted by this post from Mario Guzman, I was finally about to put into words what bothers me about Siri – and what I hope Apple’s upcoming A.I.-focused WWDC will improve.
We’re over a decade into the industry’s voice assistant experiment, and given the same input, the output doesn’t feel reliably deterministic. Voice is an interface that is not stable or discoverable.
Every morning, I say, “Hey, Siri. Turn on the office lights.”
This morning, Siri responded, “Sure. What would you like them set to?”
Did I misspeak? Did background noise throw off the voice recognition? Did a random number generator deep inside the HomePod choose the wrong code path? Was my Hue bridge momentarily offline, so Siri fell back to controlling my thermostat?
I don’t know. And there’s no way to find out.
Everything about Siri (and competing assistants) is unknowable.
And even though the lights will turn on nine out of ten times, or even ninety-nine out of one hundred times, I still feel a little tension in my voice each time. My requests feel more like guesses than instructions.
Every interaction feels like finding a light switch in a dark room.