The AI always took it seriously, and refused to help.

Speech is a low-bandwidth communication medium. It also has a high error rate, even with two native speakers who know each other well and who are physically close to each other.

Speech is also linear in nature, unless you are being rude and interrupting the other person.

In human communication, 90% of the content is provided through means other than audio. Its visual. It’s your body language. It’s what is going on around you, and what you are paying attention to. It’s your mannerisms, and how you stand or sit, and what you do with your hands. In video calls, you lose a lot of that context, because your video window onto the other person is so greatly constrained.

Of the remainder, 90% of the content is not the words being spoken. It’s your tone of voice, your inflection, whether you are talking fast or slow, how loud you are talking, etc…. On the telephone or on a video call, again there is so much compression, and those calls are almost always done without a central computer processing system that can meaningfully handle more than one audio stream at a time. And computers are still pretty rubbish at understanding all that audio.

That leaves just 1% of the total communication being the words that are spoken. And computers are still pretty rubbish at understanding that, too.

Speaking to devices works great on Star Trek, because what you’re really doing is explaining to the audience at home some important thing they need to know, in the guise of speaking to a computer. But the “computer” here is actually a piece of cardboard, or maybe an Okudagram printed on transparent plastic.

The real world is not like that.