2. I'm not sure what part of the article you're referring to here, because as far as I understand it, the author is speaking about political and epistemological and every other possible difference, framed as axes in a worldview space.
I don't think anyone is internally consistent all the time (yes, including me). Whatever principle or set of principles someone thinks they have, there's always some conceivable situation where we could line them all up so there should be an obvious "right" conclusion, but the person will still instinctively react against it and feel it's wrong.
This is why we won't ever be able to, for example, train up AI surrogates to go have all our arguments for us, on a societal level, faster and in more depth than we ever could since we have to worry about things like continuing to eat, and let them make decisions about agreements that should govern our actions. The agents would run into our self-contradictions all the time, and have to point them out to us to try to get clarity, and then we'd get so frustrated with their pointing this out all the time that we'd turn them off and abandon the whole endeavor.
The thought experiment wasn't supposed to be 'plausible'. It's purpose was to introduce the concept of worldviews residing in a high dimensional space, which could perhaps be approximated. It was also actually a quote from an earlier article, with more context.
> I don't think anyone is internally consistent all the time
I agree. Only a moron or a God could accomplish such a feat. However, this doesn't invalidate the thought experiment - all you need to do is add a time axis.
"On this particular morning, the Judge had dreamed that she sat in a beautiful parlor while doves fed grain to her kittens. Her general amiability is notably up, and she will deliver far less lenient verdicts all morning because of her contrarian tendencies. Then, a bone in her smoked haddock sandwich triggers repressed childhood memories, leaving her 31.64% more disaffected and apathetic." - The thought experiment could encompass full knowledge of all of this, from moment to moment.
> This is why we won't ever be able to, for example, train up AI surrogates to go have all our arguments for us, on a societal level ... and let them make decisions about agreements that should govern our actions
Is that why? Or is it because by the time AI is that advanced the likelihood of our being able to control it is virtually nil, thanks to a breakneck developmental arms race...
Besides which, being frustrated with clarity is something of a contradiction in itself, lol. I imagine a hyper-intelligent being could teach us to be better people in a way we'd actually like, if they were into that sort of thing. The Culture books explore this quite well.
Some AI models like those from OpenAI or Anthropic are biased to the left. Anyone can test that. So actually people thought asking machines build arguments for them and trained them accordingly.