Deriving techniques on the fly

Origi­nal post: http://​​bear­​​de­riv­ing-tech­niques-on-the-fly/​​

Last year Lach­lan Can­non came back from a CFAR re­union and com­mented that in­stead of just hav­ing the CFAR skills we need the deriva­tive skills. The skills that say, “I need a tech­nique for this prob­lem” and let you de­rive a tech­nique, sys­tem, strat­egy, plan, idea for solv­ing the prob­lem on the spot.

By anal­ogy to an old clas­sic,

Give a man a fish and he will eat for a day. Teach a man to fish and he never go hun­gry again.

This con­cept always felt off to me un­til I met Anna. An amer­i­can who used to live in Alaska where they have enough fish in a river that any time you go fish­ing you catch a fish, and a big enough one to eat. In con­trast, I had been fish­ing sev­eral times when I was lit­tle (in Aus­tralia) and never caught things, or only caught fish that were too small to feed one per­son, let alone many peo­ple.

Silly fish­ing mi­s­un­der­stand­ings aside I think the old clas­sic speaks to some­thing in­ter­est­ing but misses a point. to that effect I want to add some­thing.

Teach a man to de­rive the skill of fish­ing when he needs it. and he will never stop grow­ing.

We need to go more meta than that? I am afraid it’s tur­tles all the way down.


To help you de­rive you need to start by notic­ing when there is a need. There are two parts to notic­ing:

  1. triggers

  2. introspection

  3. What next

But be­fore I fail to do it jus­tice, agen­ty­duck has writ­ten about this. Art of notic­ing, What it’s like to no­tice things, How to train notic­ing.

The Art Of Notic­ing goes like this:

  1. An­swer the ques­tion, “What’s my first pos­si­ble clue that I’m about to en­counter the prob­lem?” If your prob­lem is “I don’t re­spond pro­duc­tively to be­ing con­fused,” then the first sign a cru­cial mo­ment is com­ing might be “a fleet­ing twinge of sur­prise”. What­ever that feels like in real time from the in­side of your mind, that’s your trig­ger.

  2. When­ever you no­tice your trig­ger, make a pre­cise phys­i­cal ges­ture. Snap your fingers, tap your foot, touch your pinky finger with your thumb—what­ever feels com­fortable. Do it ev­ery time you no­tice that fleet­ing twinge of sur­prise.

How To Train Noticing

  1. I guess. I re­mem­ber or imag­ine a few spe­cific in­stances of en­coun­ter­ing weak con­trary ev­i­dence (such as when I thought my friend wasn’t at­tracted to me, but when I made eye con­tact with him across the room at a party he smiled widely). On the ba­sis of those simu­la­tions, I make a pre­dic­tion about what it will feel like, in terms of im­me­di­ate sub­jec­tive ex­pe­rience, to en­counter weak con­trary ev­i­dence in the fu­ture. The pre­dic­tion is a ten­ta­tive trig­ger. For me, this would be “I feel a sort of match­ing up with one of my be­liefs, there’s a bit of dis­so­nance, a tiny bit of fear, and maybe a small im­pulse to di­rect my at­ten­tion away from these sen­sa­tions and away from thoughts about the ob­ser­va­tion caus­ing all of this”.
  2. I test my guess. I keep a search go­ing on in the back­ground for any­thing in the neigh­bor­hood of the ex­pe­rience I pre­dicted. Odds are good I’ll miss sev­eral in­stances of weak con­trary ev­i­dence, but as soon as I re­al­ize I’ve en­coun­tered one, I go into re­flec­tive at­ten­tion so I’m aware of as many de­tails of my im­me­di­ate sub­jec­tive ex­pe­rience as pos­si­ble. I pay at­ten­tion to what’s go­ing on in my mind right now, and also what’s still loop­ing in my very short-term mem­ory of a few mo­ments be­fore I no­ticed. Then I com­pare those re­sults to my pre­dic­tion, not­ing any­thing I got wrong, and I feed that in­for­ma­tion into a new pre­dic­tion for next time. (I might have got­ten some­thing wrong that caused the trig­ger to go off at the wrong time, which prob­a­bly means I need to nar­row my pre­dic­tion.) The new pre­dic­tion is the new trig­ger.
  3. I re­peat the test un­til my trig­ger seems to be ac­cu­rate and pre­cise. Now I’ve got a good trig­ger to match a good ac­tion.

Deriva­tions (as above) are a “what next” ac­tion.

My deriva­tions come from ask­ing my­self that ques­tion or other similar ques­tions, then at­tempt­ing to an­swer them:

  • What should I do next?

  • How do I solve this prob­lem?

  • Why don’t other peo­ple have this prob­lem?

  • Can I make this prob­lem go away?

  • How do I de­sign a sys­tem to make this not mat­ter any more?

(you may no­tice this is stim­u­lat­ing in­tro­spec­tion—this is what it is)


The post that led me to post on deriva­tions is this post on How to pre­sent a prob­lem hope­fully to be pub­lished to­mor­row.

This post took ~1 hour to write.

Cross posted to lesswrong