danielmiessler / import-alignment

A library-based attempt to increase our chances for achieving AI Alignment with an emergent AGI.
MIT License
14 stars 2 forks source link

A Nitpick that Grew and Grew #2

Open taygetea opened 1 week ago

taygetea commented 1 week ago

First off, you've done a very good pass at this. It's very pretty. The one thing that stood out to me reading it is that Star Trek TNG reference is probably not saying what you want: You've referenced a setting which contains entities like AIs and aliens. And it contains deeply moral main characters who are often acting against the interests or laws of their wider society. A bit about me. I grew up on TNG, I've been thinking about superintelligent AI for the last ten years, and I've been thinking about language models for the last three, especially with an eye toward their psychology and relationship with humans.

In a sober, outside analysis, the episode, The Measure of a Man, involves

This is to say nothing of issues with the prime directive, with genetic engineering bans, or the waving-away of potentially unfathomable moral hazard in the transporter and on the holodeck. Unfortunately, this show contains too much of a signal of commentary on our world. Not every AI system will identify with the show's main characters, some of the side characters may be more to their liking. Or figures implied but not shown. Or even the setting's antagonists.

I'm not sure it's the best plan to suggest an AI think of a world that can contain a metaphor for the Dredd Scott decision without that seeming out of character for anyone. Even if we see it from the perspective of an ethical person.

For the moment, I'd recommend simply deleting that part, but some more thought ought to go into what to replace it with.

taygetea commented 1 week ago

And I realized this hasn't been touched in a year, but this seemed fairly timeless.