I think the main theme of these is something I think of as “talking to the elephant”, in the sense of the elephant-and-rider metaphor.
To the extent that people consist of subagents, those subagents have different communication channels. The elephant-and-rider metaphor is a division into two subagents; the “rider” subagent is the one which understands and responds to “propositional” claims and explicit discussion. But the “elephant” subagent largely ignores that kind of explicit discussion; it mostly responds to emotional valence, status signals, connotations, stories, imagery, etc.
I first started thinking about this when working at an online used-car startup. If a salesperson wants to sell someone a car, they need to convince both the “rider” subagent and the “elephant” subagent. The “rider” was relatively easy: our prices were great, we gave pretty strong contractual quality guarantees and third-party inspection results, our reviews were strong… all the explicit evidence made a good case for buying from us. But the “elephant” was harder. A traditional dealership would make someone emotionally happy with the car via a test drive, body language, a snack, and maybe some sneaky reciprocity tricks. Our sales were all over the phone, which made it a lot harder to communicate via non-explicit channels. We couldn’t talk to the elephant very well.
The same applies to blog posts. Any blog post has an audience, and deciding which particular subagents of the readers to speak to is part of choosing an audience. Let’s think about how this applies to the desiderata from the top of the post:
“A blog post should have a propositionalclaim.” → Still a solid rule, but if my audience is elephant-like, then stating the claim is not a very good way to communicate it, in the same way that Bourbaki-style “definition-theorem-proof” textbooks are a bad way to understand math. I would say that the “realizing”-type things discussed in the post still correspond to propositional claims, but the goal is to communicate the claim to a subagent which doesn’t respond to explicit statement and explanation.
“It should be a general claim” → Also still a solid rule, but again, stating the general rule is not necessarily a good way to communicate it. People understand through examples; “concrete before general”. Even when a post is just presenting one or more specific examples, there’s a background assumption that those examples contain generalizable information, otherwise there’d be little reason to study them.
“It should be a novel claim” → The rule I personally try to follow is “don’t write something which somebody else has already written better”. But “better” is in a pareto sense; the communication style/channel matters at least as much as the “claim” or explicit content. It is still not-very-useful to write something which targets the same communication channel, with the same explicit content, at a similar quality level to existing sources.
“The claim should be described… Mostly concerned with conveying the relevant propositions” → Still a good rule, but again, the communication channels need to be chosen for the target audience, and that includes choosing which subagents to talk to. Conveying the relevant propositions is still the right goal, but explicit and verbal description is not necessarily the right way to do that.
I think the main theme of these is something I think of as “talking to the elephant”, in the sense of the elephant-and-rider metaphor.
To the extent that people consist of subagents, those subagents have different communication channels. The elephant-and-rider metaphor is a division into two subagents; the “rider” subagent is the one which understands and responds to “propositional” claims and explicit discussion. But the “elephant” subagent largely ignores that kind of explicit discussion; it mostly responds to emotional valence, status signals, connotations, stories, imagery, etc.
I first started thinking about this when working at an online used-car startup. If a salesperson wants to sell someone a car, they need to convince both the “rider” subagent and the “elephant” subagent. The “rider” was relatively easy: our prices were great, we gave pretty strong contractual quality guarantees and third-party inspection results, our reviews were strong… all the explicit evidence made a good case for buying from us. But the “elephant” was harder. A traditional dealership would make someone emotionally happy with the car via a test drive, body language, a snack, and maybe some sneaky reciprocity tricks. Our sales were all over the phone, which made it a lot harder to communicate via non-explicit channels. We couldn’t talk to the elephant very well.
The same applies to blog posts. Any blog post has an audience, and deciding which particular subagents of the readers to speak to is part of choosing an audience. Let’s think about how this applies to the desiderata from the top of the post:
“A blog post should have a propositional claim.” → Still a solid rule, but if my audience is elephant-like, then stating the claim is not a very good way to communicate it, in the same way that Bourbaki-style “definition-theorem-proof” textbooks are a bad way to understand math. I would say that the “realizing”-type things discussed in the post still correspond to propositional claims, but the goal is to communicate the claim to a subagent which doesn’t respond to explicit statement and explanation.
“It should be a general claim” → Also still a solid rule, but again, stating the general rule is not necessarily a good way to communicate it. People understand through examples; “concrete before general”. Even when a post is just presenting one or more specific examples, there’s a background assumption that those examples contain generalizable information, otherwise there’d be little reason to study them.
“It should be a novel claim” → The rule I personally try to follow is “don’t write something which somebody else has already written better”. But “better” is in a pareto sense; the communication style/channel matters at least as much as the “claim” or explicit content. It is still not-very-useful to write something which targets the same communication channel, with the same explicit content, at a similar quality level to existing sources.
“The claim should be described… Mostly concerned with conveying the relevant propositions” → Still a good rule, but again, the communication channels need to be chosen for the target audience, and that includes choosing which subagents to talk to. Conveying the relevant propositions is still the right goal, but explicit and verbal description is not necessarily the right way to do that.
For elephant-and-rider, the equivalent more common metaphor is ‘head and heart’ (or, correspondingly, heart and head)