Rationality Exercises Prize of September 2019 ($1,000)

I’m giv­ing out $1,000 of prize money for the best ex­er­cises sub­mit­ted in the next two weeks on a topic of in­ter­est to the LW com­mu­nity. I’m plan­ning to dis­tribute $1,000, with $500 of that go to the first place.

To sub­mit some ex­er­cises, leave a com­ment here link­ing to your ex­er­cise(s) by mid­night at the end of Fri­day 20th Septem­ber PDT (San Fran­cisco time), and I’ll an­nounce the win­ners by the Fri­day two weeks later (give me the time to try a bunch out). You’re wel­come to post them as a LW post, on your short­form feed, or pri­vately link them to me in a PM if you want, though I’ll be pub­lish­ing all the en­tries that win a prize.

Why ex­er­cises?

I’d like to be to prac­tice us­ing ideas, and to know whether I ac­tu­ally un­der­stand them.

I want to con­cretely prac­tice the art of ra­tio­nal­ity (and other arts), but I don’t have many nat­u­ral af­for­dances to do that. If peo­ple added ex­er­cises to their posts, I think that I’d do them. I do sab­bath-like re­cov­ery days, and I have a weekly ses­sion with Ja­cob Lager­ros where we work on prob­lems in the HPMOR-recom­mended book Think­ing Physics, both of which I find ex­ceed­ingly valuable. I’d love to spend more time play­ing with other ideas peo­ple put for­ward on LW.

I also think it’s sur­pris­ingly com­mon for me and a friend to achieve a dou­ble illu­sion of trans­parency where we’re both us­ing a con­cept or phrase in con­ver­sa­tion, but ac­tu­ally have a very differ­ent refer­ent in mind. I think small tests and checks can zoom in sur­pris­ingly quickly on mis­com­mu­ni­ca­tion.

So I’m run­ning the prize to get some ex­er­cises, for me and for oth­ers on LW who want to try them out.

I could talk more about why ex­er­cises are valuable, but a lot of my think­ing here is down­stream of read­ing the book Think­ing Physics, so I’d rather just let its au­thor, Lewis Car­roll Ep­stein, speak in­stead. (This is from the open­ing of the book, all for­mat­ting is origi­nal.)

The best way to use this book is NOT to sim­ply read it or study it, but to read a ques­tion and STOP. Even close the book. Even put it away and THINK about the ques­tion. Only af­ter you have formed a rea­soned opinion should you read the solu­tion. Why tor­ture your­self think­ing? Why jog? Why do push-ups?
If you are given a ham­mer with which to drive nails at the age of three you may think to your­self, “OK, nice.” But if you are given a hard rock with which to drive nails at the age of three, and at the age of four you are given a ham­mer, you think to your­self, “What a mar­vel­lous in­ven­tion!” You see, you can’t re­ally ap­pre­ci­ate the solu­tion un­til you first ap­pre­ci­ate the prob­lem.
What are the prob­lem of physics? How to calcu­late things? Yes—but much more. The most im­por­tant prob­lem in physics is per­cep­tion, how to con­jure men­tal images, how to sep­a­rate the non-es­sen­tials from the es­sen­tials and get to the hear of a prob­lem, HOW TO ASK YOURSELF QUESTION. Very of­ten these ques­tions have lit­tle to do with calcu­la­tions and have sim­ple yes or no an­swers: Does a heavy ob­ject dropped at the same time and from the same height as a light ob­ject strike the earth first? Does the ob­served speed of a mov­ing ob­ject de­pend on the ob­server’s speed? Does a par­ti­cle ex­ist or not? Does a fringe pat­tern ex­ist or not? Th­ese qual­i­ta­tive ques­tions are the most vi­tal ques­tions in physics.
You must guard against let­ting the quan­ti­ta­tive su­per­struc­ture of physics ob­scure its qual­i­ta­tive foun­da­tion. It has been said by more than one wise old physi­cist that you re­ally un­der­stand a prob­lem when you can in­tu­itively guess the an­swer be­fore you do the calcu­la­tion. How can you do that? By de­vel­op­ing your phys­i­cal in­tu­ition. How can you do THAT? The same way you de­velop your phys­i­cal body—by ex­er­cis­ing it.
Let this book, then, be your guide to men­tal pushups. Think care­fully about the ques­tions and their an­swers be­fore you read the an­swers offered by the au­thor. You will find many an­swers don’t turn out as you first ex­pect. Does this mean you have no sense for physics? Not at all. Most ques­tions were de­liber­ately cho­sen to illus­trate those as­pects of physics which seem con­trary to ca­sual sur­mise. Re­vis­ing ideas, even in the pri­vacy of your own mind, is not painless work. But in do­ing so you will re­visit some of the prob­lems that haunted the minds of Archimedes, Gal­ileo, New­ton, Maxwell, and Ein­stein. The physic you cover here in hours took them cen­turies to mas­ter. Your hours of think­ing will be a re­ward­ing ex­pe­rience. En­joy!

What does this look like?

Here are great ex­er­cises that have been on LessWrong in the past.

In my primer on Com­mon Knowl­edge, I opened with three ex­am­ples and asked what they had in com­mon. Then, to­wards the end of the post, I ex­plained my an­swer in de­tail. I could’ve triv­ially taken those ex­am­ples out from the start, in­cluded all the the­ory, and then asked the reader to ap­ply the the­ory to those three as ex­er­cises, be­fore ex­plain­ing my an­swers. There’s a du­al­ity be­tween ex­am­ples and ex­er­cises, where they can of­ten be turned into each other.

But this isn’t the only or pri­mary type of ex­er­cise, and you can see many other types of ex­er­cise in the pre­vi­ous sec­tion that don’t fit this pat­tern.

What am I look­ing for in par­tic­u­lar?

While I’m open to most pos­si­ble sub­jects, let me add one op­er­a­tional con­straint: it should be an ex­er­cise that more than 10% of LessWrong com­menters can un­der­stand af­ter read­ing up to one-to-three posts you’ve speci­fied, or af­ter hav­ing done your prior ex­er­cises. As a rule I’m gen­er­ally not look­ing for a highly niche tech­ni­cal prob­lems. (Though it’s fine to as­sume peo­ple have read any cu­rated LW se­quence.)

I asked Oli for his thought on what makes a good ex­er­cise, and he said this:

I think a good tar­get is uni­ver­sity prob­lem sets, in par­tic­u­lar for tech­ni­cal de­grees. I’ve found that al­most all of my learn­ing in uni­ver­sity came from grap­pling with the prob­lem sets, and think that I would want many more prob­lem sets I can work through in my study of both ra­tio­nal­ity and AI Align­ment. I also had non-tech­ni­cal classes with ex­cel­lent es­say prompts that didn’t have as clear “cor­rect” an­swers, but that nev­er­the­less helped me deeply un­der­stand one topic or an­other. I think both tech­ni­cal prob­lem sets and good es­say prompts would be great sub­mis­sions for this prize, though I’d en­courage pro­vid­ing at least sug­gested solu­tions (prob­a­bly best posted be­hind spoiler tags).

(What are spoiler tags? Hover over the text of this post to read the black box be­low.)

This is a spoiler tag! To add this to your post or com­ment, see the in­struc­tions in the FAQ that’s ac­cessible from the front­page on the left-menu.

(Also see this com­ment sec­tion for ex­am­ples of lots of peo­ple us­ing it to cover their solu­tions to ex­er­cises. Also to see the level of de­mand for ex­er­cises on LW.)

I’m in­ter­ested in ex­er­cises that help teach any key idea that I can’t already buy a great text­book for, al­though if your ex­er­cises are bet­ter than those in most text­books, then I’m open to it too.

I think tech­ni­cal al­ign­ment ex­er­cises will be es­pe­cially hard to do well, be­cause many peo­ple don’t un­der­stand much of the work be­ing done in al­ign­ment, and the parts that are easy to make ex­er­cises for of­ten aren’t very valuable or cen­tral.

I think that it’s of­ten eas­ier to build ex­er­cises for very ex­plicit, leg­ible con­cepts (e.g. things that look more like math), and while that’s re­ally valuable, I’m also re­ally ex­cited about ex­er­cises for other ideas too.

Ex­am­ples of things I think could have exercises

Definitely ex­er­cises for any cu­rated post or cu­rated se­quence on LessWrong. I’ve taken a look through our cu­rated posts, here are a few I think could re­ally benefit from great ex­er­cises (though the tractabil­ity varies a lot on these).

Cu­rated Sequences

Here are ex­am­ples from Cu­rated Se­quences.

Cu­rated Posts

Robin Han­son has masses of brilli­ant ideas that I’ve not got the time to mine. Some that comes to mind are his more re­cent posts on au­to­matic norms that I think could be­come some re­ally great ex­er­cises.

Some of Nick Bostrom’s ideas would be ex­cel­lent too, like the unilat­er­al­ist’s curse, or the vuln­er­a­ble world hy­poth­e­sis, or the Hail Mary ap­proach to the Value Speci­fi­ca­tion Prob­lem.

If you leave a pub­lic com­ment de­scribing what sort of ex­er­cises you might want to try cre­at­ing, I will try to re­ply with my best guess on whether it can be a good fit for this prize.

I’ll re­peat: To sub­mit ex­er­cises, leave a com­ment here link­ing to your ex­er­cise(s) by mid­night at the end of Fri­day 20th Septem­ber PDT (San Fran­cisco time), and I’ll an­nounce the win­ners by the Fri­day two weeks later (give me the time to try a bunch out). You’re wel­come to post them as a LW post, on your short­form feed, or pri­vately link them to me in a PM if you want, though I’ll be pub­lish­ing all the en­tries that win a prize.

I look for­ward to try­ing out your ex­er­cises.

Added: Prizewin­ners an­nounced in this com­ment be­low.