SilasBarta comments on The Social Coprocessor Model - Less Wrong
You are viewing a comment permalink. View the original post to see all comments and the full post content.
You are viewing a comment permalink. View the original post to see all comments and the full post content.
Comments (570)
I think the distinction can be helpfully represented in terms of my levels of understanding.
NTs understand social interaction at Level 1, the level at which one is capable of outputting the right (winning) results, even if that is due to an inscrutable black-box model contained inside oneself (which is the case here).
But there are higher levels to reach than that, and it is not an academic distinction. To advance to Level 2, you must not only produce the right results, but also be able to "plug in" your understanding of the social interaction domain to various other domains, and make inferences between them. And most NTs cannot do this: they get the results "for free", even (and perhaps especially) if they cannot derive these results as implications of other domains (or vice versa).
Roko's point, in turn, can be rephrased as saying that HFASes try to build up a Level 2 understanding directly, checking for cross-domain consistency before they adopt any rules; and that this is because their hardware doesn't feed them the correct black-box output, as happens in NTs. Further, inferences that HFASes make come from applying a more general-purpose "reasoning engine" to social interaction; to NTs, the inferences just look dumb, even if they can't explain why.
Someone with sufficiently advanced understanding will be able to connect the NT black-box model to useful models for everything else, explaining the basis for NT conventions. This can grow from an NT mind or a HFAS one, but they will take different paths.
In any case, there's a higher level of understanding to reach, even if a specific threshold suffices for some purpose.