I would lightly argue scope insensitivity (and calibration!) are both traditional rationalist topics. Incomplete, yes, but I think they're both well served by quantified intuitions and they'd be part of my ideal rationalist training program.
1I think the Subscribe to Group button will subscribe to the LessWrong community group. In theory I cross post all the things to all the appropriate places. In practice I'd suggest this mailing list or the Discord, because those are the ones I reliably remember.
The time and place is a mistake, I just fixed it. (This is why I usually do the What Where When, as a kind of checksum when I'm cross posting something to a lot of places.) Thanks for pointing that out!
Planned drills, list might get tweaked a bit between now and the meetup date:
- calibration trivia: get asked a question the answer to which is known but not to you ("what's the population of Hyderabad, India?") and put down how confident you are about your answer.
- cambist booking: get asked sets of questions about what you value such that they form a loop, and try to be consistent such that you aren't vulnerable to arbitrage.
- brainstorming: get asked for solutions to a problem, and then keep coming up with more and more and more solutions until you feel like you've actually started to exhaust the search space.
- evidentials: relate some facts you know, explicitly denoting what you think you know and how you think you know it and noticing the distinction between "I have observed this with my own eyes" and "someone I trust told me it was so."
- bayes problems: the mammogram is the classic example but I've got more, we're going to go fast and get used to working the numbers in odds form.
This is an experiment in format. Usually I'd make one of these the focus of a meetup and take our time with each example. This time around I'm going to hit all five repeatedly and at speed.
I'm one of the +4s. I would believe the liars in this post are being filtered out before they get to you Ruby, but they aren't all getting filtered out before they get to me, and they aren't getting filtered out of the general population fast enough that the man on the street won't run into them.
I actually squint a bit at "highly incompetent" in a weird way; the thing they're doing works surprisingly well in my observation. Emphasis on "surprising"! It stinks, I hate it, and also I kinda do think it works sufficiently well in some social environments that it's positive expected value for them.
Some cruxes for me:
- Bayes is the most important part of rationality.
- Maybe you think it's something else? If so, what? (Not a rhetorical question- I have three or four candidates myself.)
- Being able to do the math is important.
- Maybe you think the verbal version is sufficient?
- Maybe you think people don't actually do the math much even when they can, so it doesn't matter?
- This formulation is a meaningful step forward.
- Maybe you think the theorem version is fine and people can rederive the odds version?
- Maybe you think mentioning the odds version partway through the other guides is fine?
- Most rationalists can't write out Bayes Theorem, and can't correctly answer a mammogram problem.
- Maybe you think they can? (I invite you to test this at your local LessWrong meetup and report back! If you want to make a bet about it, I can probably get a lot of LessWrong meetups to test this and report back!)
- Maybe you object to "rationalist" being defined by ~self-identification, such that if someone can't write it out or answer the mammogram problem they don't count as a rationalist?
I am if not alarmed then at least consider it a problem, but haven't felt confusion here for at least a year. I have a pretty good model of how it happens. Someone's doing some searching on the internet, and gets recommended a LessWrong article on an Boston rents, or an AI paper, or hikers going missing. Maybe a friend recommended them a fun essay on miracles or a goofy Harry Potter fanfic. They hang around, read a few more things, comment a bit. Then they see a meetup announcement, and show up, and enjoy conversation. (Very very roughly a third of LessWrong/ACX meetups are socials, with no or minimal readings or workshops.) They go to more meetups, they make more comments on the internet, maybe they make some posts of their own and their posts get upvoted. Maybe they step up and run the meetup when the previous organizer is sick or busy or moves.
At no point did someone give them a math test. I'm basically describing my arc above, and nobody asked me to solve a mammogram problem in that process.
That's how we end up in this world.
As for what the missing thing is: my theory is to change this state of affairs, we'd need two things. We'd need to start actually regularly asking folks questions where they'd need to use it, and we'd need an explanation fast and simple enough that it can survive being taught by non-specialists who are also juggling having snacks out and getting the door for people. I love this not for its intuitiveness, but for rearranging the numbers to a shape people can do easier.
I'd give much higher odds on members of the community being able to gesture at the key ideas of base rates and priors in English sentences! (Not as high as I'd like, but higher, anyway.) But that's not the same as being able to do the calculations. And there's something slippery about describing a piece of math in intuitive sentences then trying to use it as a heuristic without quite being able to actually run the numbers, which is why I'd like to change that.
Again, I suck too! I'm running around doing a dozen things in my day to day life, none of which is remedial math practice. This kind of thing happens a lot actually. Once upon a time I did some basic interviews for some software developers, and watched comp sci grads fail to fizzbuzz correctly.
My hope is that if somehow I can get a tweet or two worth of text that teaches the numbers in a way that can fit in math people already do in their daily life (multiplication between two to four numbers) and add a small battery of exercises that use it, I might be able to package that in a way local organizers not only could use but would spread. Like you say, maybe hoping for just one more Bayes explanation is not the path. To me, this one was a meaningful step simpler and easier.
I guess I'll note as well that I want to raise the sanity waterline. To do that, I can't work with a version that wants above average intelligence. I do genuinely want to figure out how to teach Bayes to fourth graders and then go out and teach some fourth graders. C'mon, don't you want to see what people turn out like if they have access to a better mental toolkit from a young age?
Also,
it's not a particularly complicated piece of math ... even if you don't remember the exact formula, it should be very easy to rederive within a few minutes from first principles once you understand the core idea.
I think you might be having an xkcd feldspar moment.
Please no set notation! The arrow brackets are on thin ice I think.
I meant what I said above, I think there's something really good about having a Bayes explanation that requires no symbols not on a standard keyboard and no math an on-track fourth grader wouldn't know.
(And also, thank you both for improving this! I recognize you two are the ones in the arena at the moment and I wish I was able to help refine this more.)
I joke, but "Thresholding is a Sazen" sure is a sentence I'd call at least 20% correct.
"You see, the problems caused by reading the first essay can be mitigated by the second essay."
"Ah, so the essays will continue until the problems improve."
A question I have been asking since before starting work with Raemon on this, which is only more relevant now:
What rationality skill(s) do you think is most important? Put another way, if we could snap our fingers and teach some rationality skill as broadly as literacy currently gets taught, what skill should we pick?
(I don't think this is Raemon's angle on the project but it is kind of mine.)