Training Regime Day 22: Murphyjitsu 2

by Mark Xu 2 min read26th Apr 2020No comments

12


Introduction

Society has many pieces of wisdom floating around that are epistemically false, but instrumentally useful.

Individuals are told to "invest, then don't touch the money until you retire". "Invest, then don't touch it" is false because sometimes, some individuals will be able to beat the market; however, if you tell people "Invest, then don't touch it unless you can beat the market", people will be overconfident and move around their money way to much, resulting in them losing money. Instrumentally, you then tell people to never touch their money, because that results in better behavior.

College students are told to "go to class no matter what". As a factual question, there exist students that would benefit from not going to class. However, telling everyone "go to class, unless you think you would benefit from not going" results in many people that would benefit from going to class pleading their own exceptionalism. Instrumentally, you tell people to always go to class, because that results in better behavior.

On a meta level, the pattern I'm trying to describe is common wisdom of the sort "always do X" being factually incorrect but existing to provide pushback against a tendency that the truth would fail to solve. A similar pattern is someone saying "I believe X strongly" when they believe X weakly because they think it'll cause you to update to believing X weakly, which is more correct from their perspective. There's some "truth", but you can't say the truth because the end result is not acting/believing "truth", so you exaggerate to shift the end result closer to "truth".

Remember, the real rules have no exceptions.

On the object level, society has internalized:

No plan survives contact with reality.

On a factual level, this is nonsense. Many plans get executed without hiccups (many run so smoothly we don't even consider them plans). However, people tend to get too attached to their plans, so the corrective wisdom is "all plans fail" so they detach themselves at the proper speed.

The point of an applied rationality training regime is to get good enough at planning that your plans survive contact with reality.

Combative Murphyjitsu

Bog-standard murphyjitsu is an explicit process. Explicit processes are bad because they take time and effort and you have to remember to engage them. We're trying to access an entirely implicit version of muphyjitsu. The problem is, we're trying to access it explicitly, which is already always doomed to failure. (Yet we still try anyway, because failure is a matter of degree.)

Imagine that the Laws of Probability themselves were out to get you. Whenever you try and act in the world, any possible deficiency of your plan that is possible becomes real. Anything that can go wrong, will. But the Laws of Probability are not all-powerful - they can't make extremely low probability events happen, they can only make reasonable failures become more salient. How would you plan in this world? (Murphyjitsu is a martial art, after all.)

There's a mode of pessimism that makes it easier to see failures - too see all the small ways your plan might go wrong that you would normally brush off as insignificant. In the real world, the Laws of Probability aren't actively out to get you, but they sort of are.

As a matter of course, human brains have dedicated agent simulating functions. Pretending someone is Out to Get You is a good way of hardening your plans. Think of it as trying to beat the Laws of Probability at their own game.

Exercise

You have a copy of reality in your brain - use this copy to defeat the enemy. Make a plan so perfect that it survives contact with reality.

12