This is great! One subtle advantage of the list-of-koans format is that it provides a natural prompt for the reader to think up their own as an exercise.
Irene wants to believe the claim “G is an H.” H is a fuzzy category and the word “H” can be used in many ways depending on context. Irene finds she can make the sentence “G is an H” more probable by deliberately choosing a sufficiently broad definition of “H”, but only at the cost of making the word less useful.
Jessica posts something to a social-media platform that has “Like”s, but not downvotes. She doesn’t know whether no one saw her post, or if lots of people saw it but they all hated it.
10 is vague, and lacks examples. (Is it the Sorites paradox?)
11 is great. (Though it does raise the question—if you can only see upvotes minus downvotes, how do you know whether a score of 1 indicates no one cared, or everyone cared and were split both ways?)
That’s fair. For a more concrete example, see the immortal Scott Alexander’s recent post “Against Lie Inflation” (itself a reply to discussion with Jessica Taylor on her Less Wrong post “The AI Timelines Scam”). Alexander argues:
The word “lie” is useful because some statements are lies and others aren’t. [...] The rebranding of lying is basically a parasitic process, exploiting the trust we have in a functioning piece of language until it’s lost all meaning[.]
I read Alexander as making essentially the same point as “10.” in the grandparent, with G = “honest reports of unconsciously biased beliefs (about AI timelines)” and H = “lying”.
This is great! One subtle advantage of the list-of-koans format is that it provides a natural prompt for the reader to think up their own as an exercise.
Irene wants to believe the claim “G is an H.” H is a fuzzy category and the word “H” can be used in many ways depending on context. Irene finds she can make the sentence “G is an H” more probable by deliberately choosing a sufficiently broad definition of “H”, but only at the cost of making the word less useful.
Jessica posts something to a social-media platform that has “Like”s, but not downvotes. She doesn’t know whether no one saw her post, or if lots of people saw it but they all hated it.
(I moved the meta thread that used to be below this comment to its own post here)
10 is vague, and lacks examples. (Is it the Sorites paradox?)
11 is great. (Though it does raise the question—if you can only see upvotes minus downvotes, how do you know whether a score of 1 indicates no one cared, or everyone cared and were split both ways?)
That’s fair. For a more concrete example, see the immortal Scott Alexander’s recent post “Against Lie Inflation” (itself a reply to discussion with Jessica Taylor on her Less Wrong post “The AI Timelines Scam”). Alexander argues:
I read Alexander as making essentially the same point as “10.” in the grandparent, with G = “honest reports of unconsciously biased beliefs (about AI timelines)” and H = “lying”.
Note that it’s a central example if you’re doing agent-based modeling, as Michael points out.