The most important aspect in the early sci-fi visions of computers was that you could trust the answer the computer gives you. Futuristic sci-fi computers were, and are, DWIM[0] - they understand what you're asking for, and give back as accurate and relevant answer as possible. This is in stark contrast to real-world search engines and voice assistants of today: sure, they can tell you a lot of things, but half the time they can't parse your query correctly, and the results depend less on their accuracy, and more on who paid more to get their stuff in front of people (relevant or not). Second-guessing and sanity-checking answers are a standard part of our interactions with query-based interfaces. So we have a lot to go to match even cheesy robots from pulp comic books.
But we sure as hell nailed the glitter.
--
[0] - Do What I Mean.
Hey Google, who owns Hans Island?
Hej Google, hvem ejer Hans Ø?See how good an answer you get.
My input was, roughly:
> Human: "Is it possible to defy entropy?"
> Alien: "We've asked this question to Microvac, Galactic AC, Universal AC, Cosmic AC, and AC, the most powerful supercomputer in our world. However, the only answer we got was 'THERE IS AS YET INSUFFICIENT DATA FOR A MEANINGFUL ANSWER'."
And the only output I could get was:
> Human: "Why couldn't the computer answer the question?"
> Alien: "It's like asking a computer 'how do you pet a cat', which is meaningless to it."
GPT-3 didn't realize it was an Asimov reference, instead, GPT-3 decided to use the common trope in Sci-Fi that logical computers cannot understand human experience, which is out of context and clearly doesn't apply here.
But on second thought, I invoked this trope in an indirect way. Perhaps GPT-3 actually knew it - I should try to replicate the same context as the original story, and see how it goes.