"Applied Rationality Training Regime" #4: Murphyjitsu

 <  |  ^  |  >

January 4 brings me to Training Regime Day 4: Murphyjitsu which we're defining as a kind of intuition-based backtracking, using two mechanisms which, Mark seems to claim, everybody has but few people use effectively. (1) the "Surprise-O-Meter" which is a measure of how braced you are for trouble; would you be extremely surprised, very surprised, somewhat surprised, a little surprised, or not at all surprised by a given failure, such as not sleeping well or having a program crash or having your refrigerator make loud noises? If you were (unconsciously) expecting trouble, then your world-model probably somewhere has information about it. And if so, we can use (2)  "Prehindsight" which, given the thought that X will have gone wrong, guesses that it was likely caused by Y. 

  The Surprise-O-Meter and Prehindsight are obviously connected; in story terms, they're mostly different ways of looking at the same threads within whatever stories are being told at the moment. I wondered for a bit about why they'd even be separate, until I understood that they're looking in different directions. Sarah Som the Surprise-O-Meter is always getting braced for disaster, and the question we ask her is "were you bracing more than usual, or were you surprised?" In other words, we're asking her about the story-processing she was doing before an actual failure was known. A Prehindsight like James Pright is given a specific failure, and then asked work backwards in time, to add that thread to constrain the failure-stories that we might be in. He's not supposed to try to be really precise, and actually he's named for Archbishop James Ussher, who is 440 years old today and who taught us that really careful precise quantitative modelling into the past can lead us horrendously astray.  Sometimes we want to do it, but not as part of a prehindsight.

 (Am I floundering here, just a bit? Maybe so. James would rather think of himself as named for James Bond, the ornithologist whose name was borrowed by Ian Fleming for his fictional spy and who is 120 today, but his eponomy is a different sort of cautionary tale.)

So I tried to set up a Murphyjitsu office in my head, with a vague image for Sarah Som and another for James Pright. And I'm supposed to construct some really small training example of A Plan which I can pre-debug via Murphyjitsu. I think Mark (and presumably CFAR) are thinking of these mechanisms as, well, mechanical, but they are by definition aspects of a human personality and  I'd rather think of them as personalities themselves. 

   IF the plan seems trivially achievable then just do it and return (YAY);

   IF the Som is not surprised by "it failed"

   THEN ask your Pright for a cause

      AND add a subplan to deal with that, or fail and return (BOO);

   IF you didn't return already 

   THEN you still have a problem and partial plan)

   SO choose a way to break your plan down into subplans

 Now, it happens that I didn't sleep well last night, Sarah was somewhat surprised because this isn't normally an issue, and James noted that this was obviously because I'd been thinking about Murphyjitsu and it shouldn't have been a surprise and this is not prehindsight because it already happened. Ooops. Then we went back over the night and I realized that I had been asking for prehindsight on failure predictions for a small task, trying different ways of organizing the 30 posts which I might or might not complete. And there were a bunch of prefailures which didn't surprise Sarah and which James said had to do with this being a Blogger blog so I can't use Javascript the way I would like to, and then that I can't use an approach which depends on all of the posts being in place before I make them accessible. There was then a failure where Sarah was not at all surprised to see that users, including me, thought the whole thing was too visually cluttered, and James said I know I've never been a good visual designer so if I can't visualize it properly when half-asleep then it's probably too complicated. Shrink it! So semi-finally I was just going to have a logical circle of posts with an index page that points at each of them, however many there are, and each page has a simple top-center trio of one-character links labelled 

  "   <  |  ^  |  >  " 

copied manually from each post to the next, with the " < " then set to link to the previous post (or the last one if we're on post#1), while the " > " links to the next post (or the first one if we're on the last), so the user can click around in a circle. (And  " ^ " links up to the index page.)

  And the most useful prehindsight of all, I think, was James saying that doing it that way will have gone wrong because I'm too lazy and sloppy for that much manual editing, and I'm often doing updates when I'm sleepy (okay, maybe that thought occurred to him because it was 2AM, but it's still true). That semi-final method has me manually editing links in three posts (index, #1, #N) after saving a new N+1. If I leave out the idea of clicking around in a circle, leaving out the " < " from post#1 and " > " from the new post, then I've saved almost half of the edits. This will still go wrong, but less often. 

So...I'm not sure that James and Sarah are the right way to approach Murphyjitsu, but I gave it a try and I'll try more later. Meanwhile, I'll see if publishing this post, which already has its two links at the top, and then editing the index post and post#3, will encounter difficulties unforeseen by my Som & my Pright. 

In any case, happy birthday to James and James. And James.