In Fake Explanations, Yudkowsky offered a story that has stuck in my mind:

Once upon a time, there was an instructor who taught physics students.  One day she called them into her class, and showed them a wide, square plate of metal, next to a hot radiator.  The students each put their hand on the plate, and found that the side next to the radiator cool, and the distant side warm.  And the instructor said, Why do you think this happens?  Some students guessed convection of air currents, and others guessed strange metals in the plate.  They devised many creative explanations, none stooping so low as to say "I don't know" or "This seems impossible."


In this story it is also telling that in the many thoughts and explanations that surfaced, the idea "the teacher turned the plate around" was never considered.  The students failed to see the correct answer because they weren't thinking creatively enough.  While the correct approach in this situation is indeed to notice your confusion, a worthwhile approach still could be to list all the possible solutions you think could be the answer.  (And of course only list real solutions that you actually understand, not fake ones.)

So how can we improve this ability to expand our creativity when it comes to considering explanations, so things like "the teacher turned the plate around" enters our list of considerations?

One possible answer: study magic tricks.


In addition to writing and reading stuff on the internet, another hobby I like to indulge in is doing magic tricks with a deck of cards.  Many of the tricks I know are very impressive, such as making cards switch places or appearing to read people's minds. However, a lot of the tricks I know are very stunningly simple; some of them don't even involve slight of hand, and could be done by ten-year-olds with little practice.  They're just that that cleverly crafted.

I learned a lot of these tricks from YouTube -- many videos will show you a trick and then teach you how it is done.  Personally, I don't find the revelation of a trick to take away any of my enjoyment, because I find joy in the merely real, and care little for perpetuating mystery.

However, these YouTube videos for how tricks are done also provide a very effective test for your rationality: watch the trick on the video, and after it is done, pause the video.  Spend a good amount of time thinking through the trick, and then finally start thinking through ways you think the trick was done.  Only after you have your guesses should you learn how the trick actually was done.

I find that in doing this, I would quickly learn how to think creatively, and found that this not only allowed me to much more effectively figure out how card tricks were done before actually hearing the solutions, but also found that this creativity allowed me to become better at suggesting further hypotheses to other previously confusing situations, as well as becoming better at deliberating to solutions in previously intractable problems.

Not to mention that I managed to learn some rather impressive card tricks.


Good YouTube Magic Lessons:

And in the course of exploring the related videos, there are thousands of tricks to learn, and thousands of opportunities to test your rationality.  Not all of them are great quality, though.
New Comment
29 comments, sorted by Click to highlight new comments since:

I'm not sure that there's a lack of creativity involved in the sort of context of the plate situation. The real issue in the context of the plate seems to be a general heuristic that people aren't trying to deceive us. In order for society to function this is a good heuristic to have. Countries where people trust more are generally more prosperous and while there are correlation v. causation issues here there's surrounding evidence which suggests that there really is a causal link from high trust to more prosperity.

That said, the notion that being able to do this and think creatively when one knows that a trick is involved might be interesting and the idea that it would help promote creative hypothesis generation is intriguing.

I don't think that the point with the plate example is that they should have guessed that the teacher cheated, but rather that if "convection" or whatever doesn't actually predict what you saw, then it's better to say "I don't know" than to try and guess the password.

You're right that that was the point, but the setup is still isomorphic to a deception-based magic trick, in that students were told that the correct way to explain a phenomenon is to simply search for the matching password, which they did. And like in a magic trick, their lie-grounded expectations were easily frustrated.

You're right that's the main point of the story, but that doesn't mean I can't adapt the story to also serve my purposes.

The real issue in the context of the plate seems to be a general heuristic that people aren't trying to deceive us.

Not surprisingly then, a key element of most magic tricks is misdirection, or outright lying. There was a discussion about this on LW a while back, but I can't find it. Someone mentioned how if you just boldly lie about key elements of the setup, people will form expectations that you can then easily surprise. The commenter then found that this skill at lying (and noticing how trusting people are) bled into the rest of his life, which led me to suggest people should be extra careful around magicians even when they're not on stage!

Sounds like my post from 2009, Misleading the Witness, perhaps?

That's the one! Thanks for catching that. The relevant quote and my reaction were:

This riddle made me remember reading about how beginning magicians are very nervous in their first public performances, since some of their tricks involve misdirecting the audience by openly lying... they learn to be more comfortable once they find out how easily the audience will pretty much accept whatever false statements they make.

It makes me wonder how dangerous magicians can become in their regular lives.

If you have some basic understanding of how physics works, then the heating of the plate's wrong side should strike you as extremely inplausible - a lot more so than hypotheses about deliberate manipulation of experiment settings. A physical theory should be some kind of quantitative guide about updating probablities, and being such it should be able to give you very low probabilites for a class of hypotheses. The teacher's trick drew attention to the fact that the students markedly did not use physics to generate quantitative answers. So, per my reading, the focal point of the experiment wasn't suspending trust in people.

Still, "the teacher turned the plate around" should come up in the grand list of possible solutions, but it shouldn't be privileged, and it should be weighed against the prior probability of "the teacher will deceive us".

There's a difference between "considering deception and then dismissing it as unlikely" and "not even considering deception at all".

Considering a hypothesis takes non-negligible time. "Not even considering it at all" is what "dismissing due to low posterior weighting due to low prior weighting" feels like from the inside. If the posterior gets high enough, you'll "start considering it".

There may also be good signalling reasons to shy away from considering the hypothesis that your is deceiving you. If so, I would expect my untutored priors to be too low.

I'm not sure I understand what you are saying. To me, it seems you are suggesting that we subconsciously consider every possibility and weigh the probabilities, and only the likely enough ones float to our conscious consideration.

Also, I can consciously consider situations that have very, very low odds, such as the assumption that the plate is a metal alloy not just from any planet, but specifically from Zepton IV. So if I can consider outlandish things like that, why don't I consider the idea that the teacher was deceptive?

We consider low-probability possibilities in batch, by treating with their reference classes. For instance, once we think of the turned-plate hypothesis, we don't separately consider "the plate was turned 180 degrees", "the plate was turned 179.9 degrees", "the plate was turned 179.95 degrees", etc. If we wanted to distinguish among these subhypotheses, we could, but mostly we don't bother.

You chose Zepton IV largely at random. The hard part is locating the right specific hypothesis.

Right, I agree with you. Sorry about the "Zepton IV" thing, then.

But still, we do consider low-probability possibilities. So I'm still not sure what you meant by:

"Not even considering it at all" is what "dismissing due to low posterior weighting due to low prior weighting" feels like from the inside.

I mean that when a human thinks they feel like they're "not even considering it at all", they actually are very slightly considering it.

How do you know that? I don't mean that in a "I think you're wrong" way, but in a "I think you're right, but I'm interested in knowing why" kind of way.

Sometimes a human feels like they're not considering a hypothesis at all, and later starts considering it. That's not what confidence zero looks like.

Actually, humans do sometimes behave as though their confidence in a proposition was as close to zero as we're able to measure. I can't think of any non-politically-charged examples at the moment, but consider for example the sort of confusion that leads someone to ask "So are you Blue or Green?" of someone who's just finished explaining that they're Red.

That's what I would describe as someone not considering a hypothesis, and then later starting considering it. That is not what I would describe as someone subconsciously considering a hypothesis the entire time, at least without further justification.

Remember that not considering a hypothesis is not the same as saying the hypothesis has a low probability. Saying the hypothesis has a low probability is considering it and then discarding it. I think we're talking about two different things.

I can't think of any non-politically-charged examples at the moment, but consider for example the sort of confusion that leads someone to ask "So are you Blue or Green?" of someone who's just finished explaining that they're Red.

That reminds me of what Scott Adam's called The Two-Bucket Mind.

If those teacher's students were absolutely not expecting a lie, then another out-of-the-box question based on physics they should understand wouldn't trick them. The trust has been broken. On the other hand, if the problem is their inability to be creative enough, they won’t become creative just because they learned not to trust the teacher.

My high school physics teacher in high school who liked tricking us. Demonstrating his point about reflections off of light/dark surfaces, he covered up the laser pointer while shining it at a black binder. He put a compass next to a magnet to throw us off. These tricks were rare enough that we didn’t expect them every time, but we also knew not to blindly trust his setup. Still, there were plenty of people who fell for them every time.

And then came the torque wheel, a gyroscope bicycle wheel almost exactly like the one in this video. My first reaction, based on physics I did understand (and that wasn’t it at that time) was, “That’s impossible!” Then the teacher then told us it wasn’t a trick. He wouldn’t lie, but my reaction was still, “That’s impossible!” If I remember correctly, my hypothesis involved a hinge at the end of a solid string. Eventually, the teacher just had me hold the wheel and spun it… and the friggin’ thing moved on its own!!! I even checked that the axis or the rim didn’t contain any magicary before I was able to admit that, “Huh, I guess it is possible.”

A couple years later after that, another physics teacher inadvertently placed a compass on top of the table with a classroom computer inside of it. And then he had us learn N/S/E/W by pointing. I was the only moron in a 200-person class pointing to the “wrong” North.

Countries where people trust more are generally more prosperous and while there are correlation v. causation issues here there's surrounding evidence which suggests that there really is a causal link from high trust to more prosperity.

My intuition (and only my intuition, I haven't been able to research this effectively) suggests that the causal link is in the other direction. That is, in more prosperous countries/regions there is higher trust, since fewer people (in general) need to push limits to be able to live comfortably, so there is less crime.

It's worth noting that a standard technique to weaken a rival group is to deter or prevent its members from cooperating — sometimes by disrupting their communications; but also by inciting rivalries or jealousies (i.e. lack of trust) within the group. We see this in everything from high-school cliques, to Internet trolling, to office politics, to counterintelligence and psyops among nation-states.

Ceteris paribus, any human social group is more effective if its members cooperate toward accomplishing its goals. Therefore a rival group which wants to prevent those goals from being accomplished can do well by disrupting the group's cooperation.

"Creativity" implies generation. And yet, does the proper definition of it emphasize generation? Scott Adams has said that creativity is the ability to purge compelling, half functional ideas from one's mind, basically holding off on proposing solutions.

The strength of intelligence, and humanity, is improvising where hard-coded adaptations would be inferior.

I'm not willing to say that the problem in such a case was the absence of a specific, narrow heuristic rather than a lack of creativity. Creativity is getting by when narrow heuristics fail, and not cutting corners by assuming too much.

However unlikely it is in a specific case that the specific usually-valid heuristic "they are not trying to deceive you" fails, this was not a one-off case because there are a great number of usually-valid heuristics, each of which will rarely fail, but at least one of which will fail somewhat often, such that somewhat often a person with low creativity will fail.


I've recently enjoyed Penn & Teller's show Fool Us, in which other magicians attempt to fool them. I found it lots of fun to try to figure out how the tricks might work (and if you google a bit, you can find complete explanations). It's enlightening to go through the stages of "Impossible! It must be some kind of editing!", "They are normal human beings. There must be some way...", "Well, maybe that works... but that's really complicated..." to "Of course! It's really easy!".

(The same goes for discussions about the tricks. "Nah, it could never be done that way! That's ridiculous!" "Here's a screenshot that proves it." "Oh.")

I'm interested in your reaction to this trick. (about 4:10 to 6:10)


Rot13'ed, just in case.

V'z abg fher jul lbh'er vagrerfgrq va zl ernpgvba gb guvf cnegvphyne gevpx (gur rssrpg vf n ovg haqrejuryzvat), ohg V'yy whfg tb guebhtu zl jubyr gubhtug cebprff urer, fb lbh pna cvpx bhg jungrire lbh jnagrq.

V'z hasnzvyvne jvgu zntvpvna'f jnk, fb V whfg nffhzr vg'f yvxr abezny jnk. Gryyre jnezf vg hc gb znxr vg fgvpx naq oraqf gur pneq rabhtu gung vg jvyy ubyq ba whfg oneryl. Vg'f pyrne gung jura ur chyyf gur "guernq", ur nyfb chfurf qbja ba gur gnoyr jvgu gur bgure unaq naq chyyf njnl jvgu uvf guhzo.

Zl svefg thrff jnf gung ur tvirf gur gnoyr n yvggyr chfu gb znxr gur pneq whzc hc, ohg gung frrzf haeryvnoyr. Vg'f qbar va n fghqvb, fb V thrff gurl pbhyq tb guebhtu rabhtu gnxrf hagvy gurl trg vg whfg evtug, ohg gung'f abg C&G'f fglyr.

Nyfb, gur ivqrb vf gbb hapyrne gb gryy vs gurer rira vf nal jnk ng nyy, ohg gur xvq jbhyq pregnvayl frr vg ng gung qvfgnapr, naq znlor vs lbh sbyq gur pneq evtug be whfg hf n yvggyr ovg bs fcvggyr, lbh pbhyq rira qb jvgubhg vg, ohg gung'f rira zber haeryvnoyr.

V jngpu gur jubyr ivqrb ntnva, abgvpvat gung gur tynff pbhagre unf n fbzr oynpx pbire ba vg naq Craa'f nez arire zbirf njnl sebz vg. Nyfb, Gryyre chfurf gur pneq qbja va n jrveq jnl, fb ur cebonoyl cnyzf vg naq cynprf n cer-sbyqrq bar. Ur pregnvayl pna trg evq bs gur jnk gung jnl. Svanyyl, gurer'f cebonoyl fbzrguvat jvgu gur qrpx, znlor n fcevat zrpunavfz, gung Gryyre gevttref jvgu uvf yrsg unaq, znlor guebhtu n guernq Craa uvq jvgu uvf nez gur jubyr gvzr. Be n ohggba haqrearngu gur cvrpr bs pybgu. Be rira rnfvre, Craa gevttref vg bhg bs fvtug.

N fcevat jbhyq or n ovg gevpxl gb frg hc (naq gb uvqr nsgrejneqf jvgubhg znxvat bgure pneqf whzc), ohg qbnoyr. Znlor gur qrpx unf n gval ubyr va vg, n guernq tbvat guebhtu vg naq vf tyhrq gb gur gbc pneq, juvpu Gryyre sbyqf hc jura cnyzvat gur bgure bar. Craa gura chyyf gur guernq guebhtu gur qrpx jvgu uvf bss-fperra unaq, abguvat ubyqf gur pneq qbja, vg whzcf hc naq n ovg bs tyhr be jnk erznvaf. Abg fher juvpu irefvba vf rnfvre gb qb. Gur frpbaq bar vf zrnare orpnhfr vg unf vagragvbany snxr zvfqverpgvba sbe gur xvq gb abgvpr (Gryyre'f yrsg unaq), fb V yrna gbjneqf gung zrgubq. (Nyfb, gung jbhyq tvir gur gevpx frireny ynlref bs zvfqverpgvba, juvpu svgf sebz n zrgn crefcrpgvir naq rkcynvaf jul lbh oebhtug vg hc. Gheaf vg vg'f zber vagrerfgvat guna V vavgvnyyl gubhtug.)

Fb va gbgny, Gryyre cynprf jnk, qryvorengryl arire fubjf gur pneq pyrneyl, cnyzf naq ercynprf vg, cynprf n frpbaq, cer-sbyqrq bar (znlor jvgu fbzr fcvggyr) ba n snxr qrpx, gevttref n fcevat be chyyf n guernq guebhtu gur qrpx, "Vf guvf lbhe pneq?". Gur xvq pna arire ercebqhpr vg naq V jebgr gbb zhpu nobhg vg.

(Naq Craa'f unve. Jbj. Gung'f gur erny zntvp va guvf ivqrb.)

This is the (possibly true) explanation of the trick that P&T gave in Cruel Tricks for Dear Friends:

Jura Craa jnf va gur guebrf bs jevgvat gur "Vaivfvoyr Guernq" fubeg fgbel, ur pnzr gb zr bar qnl naq nfxrq zr vs V xarj bs n gevpx gung jbhyq fvzhyngr gur cerfrapr bs vaivfvoyr guernq va gur cvgpu. Ab fhpu gevpx rkvfgrq. "Qbrf vg unir gb jbex?" V fnvq.

"Bs pbhefr abg. Vg whfg unf gb or cynhfvoyr." Fb nsgre n yvggyr qvfphffvba jr pnzr hc jvgu gur vqrn bs gur pneq jvgu gur oraq, uryq syng ntnvafg gur qrpx ol n onyy bs jnk, juvpu riraghnyyl eryrnfrf jura gur jnk tvirf bhg. Jr gevrq vg. Vg qvqa'g jbex. Gur pneq jbhyq ybfr gur oraq. Be gur jnk jbhyqa'g fgvpx. Be vg jbhyq fgvpx gbb jryy. Ohg vg fhvgrq gur fgbel whfg svar. Naq gurer jnf fbzrguvat arng nobhg rira gur gevpx oruvaq gur gevpx orvat zlguvpny.

Yngre, jura jr qrpvqrq gb nqncg gur fubeg fgbel nf n svyz fpevcg, jr jebgr gur gevpx vagb gur fgntr qverpgvbaf, whfg nf gubhtu vg rkvfgrq. Abobql dhrfgvbarq vg. Vg jnf gur cebqhpre'f svefg cebwrpg jvgu hf naq ur nffhzrq jr xarj jung jr jrer qbvat. Vg jnf bayl nsgre jr jrer jryy vagb cer-cebqhpgvba gung jr ernyvmrq gung n pnzren jbhyq unir gb cubgbtencu gur Vaivfvoyr Gernq ebhgvar, v.r., jr'q npghnyyl unir gb qb vg.

Fb jr unq gb purng.

Vs lbh rire frr gur "Vaivfvoyr Guernq" ba ivqrb, jngpu pnershyyl. Lbh'yy abgvpr gung gurer vf n pnzren phg orgjrra gur npgvba bs fubjvat gur orag pneq orvat chg ba gbc bs gur qrpx naq gur zbzrag gung vg cbcf hc. Orgjrra fubgf jr fjvgpurq va n tvzzvpxrq cnpx bs pneqf. Gur gbc pneq jnf ybnqrq hc jvgu orag fcevat fgrry gb tvir vg n tbbq whzc. Gur erfg bs gur cnpx jnf nyy tyhrq gbtrgure va n oybpx naq unq n ubyr qevyyrq guebhtu vg. Gurer jnf n fubeg cvrpr bs svfuyvar gung ena sebz gur zvqqyr bs gur gbc pneq guebhtu gur ubyr va gur cnpx, naq ubbxrq bagb n jver haqre gur erq cnq ba gur pbhagre. gung jver, va ghea, jnf gvrq gb n fgevat, juvpu Craa chyyrq ba phr juvyr ur fnvq uvf yvarf.

Lbh qba'g frr gur svfuyvar naq fgrry ba gur snpr bs gur pneq jura V ubyq vg hc gb Jnyyl orpnhfr, nf V cvpx vg hc, gur pneq qebcf bhg bs ivrj orybj gur ybjre rqtr bs gur gryrivfvba cvpgher vagb gur unaq bs n jnvgvat cebc zna, jub vafgnagyl fhofgvghgrf n cynva orag pneq jvgu n yvggyr jnk va gur zvqqyr.

Yngre jr sbhaq bhg gung fbzr zntvp-fgber bjaref unq pbagnpgrq gurve fhccyvref, vadhvevat jurgure "Vaivfvoyr Guernq, nf Frra Ba GI" jnf ninvynoyr. Jurerire Noare-Pnqnoen Zvyfgrva vf, V org ur'f cebhq.

Va bgure jbeqf, gur pbeerpg nafjre vf gung vg'f vzcbffvoyr naq vg zhfg or fbzr xvaq bs rqvgvat.

I sometimes feel as though magic tricks can be divided into "clever tricks" and "cheap tricks". For example, this is a clever trick, while any trick involving identical twins and almost any trick involving a confederate in the audience is a cheap trick.

Bwahahahahahaha! I'll admit I kinda freaked out at first.

This sounds like a good exercise. Also the card tricks might come in handy in some social situations.