2017-12-04 RRG Notes
- When doing inadequacy analysis, we need to draw a distinction between wrong guesses and false cynicism
- If people truly don't care about the outcome of a bad equilibrium, they won't make extraordinary efforts to break out of that equilibrium
- Even inadequate systems only have a finite amount of failure
- Sloppy cynicism will usually be wrong, because it will explain too much - i.e. the system isn't as broken as a completely cynical view would have it be
- Seeing inadequacy everywhere is the same as seeing inadequacy nowhere
- The point of learning inadequacy analysis is to give yourself permission to try novel strategies, while having an understanding of whether those strategies are likely to work
- Break blind trust in institutions
- But also learn when institutions are likely to be right
- Three Step process:
- Realize that there are exploitable strategies that haven't already been used
- Calibrate until you're not always seeing exploitability or inexploitability
- Fine tune against reality
- Eliezer finds out that medical competence is high-variance
- You can't blindly trust your doctor
- You can't blindly trust yourself either
- When we think about inadequacy, we're deciding whether we trust society to be more or less competent than we are
- The modest viewpoint turns competence into a social status game
- But our beliefs shouldn't depend on what sort of person we are - our beliefs should depend on what the world is actually like
- Modest people end up believing that they live in an inexploitable world because they're trying to avoid acting like an arrogant person
- The true alternative to modest epistemology isn't an immodest epistemology where you decide you know better than society on all questions, the true answer is to decide for yourself on a question-by-question basis
- Trust, but verify
- Realize that a system as a whole will often perform worse than any individual within the system due to misaligned incentives and communication overhead
- It takes far less effort to identify a correct expert than it does to become a correct expert
- Not easy, but possible for amateurs to do
- There is no shortage of contrarians whose ideas are better than the mainstream
- It's possible to know things that the average authority doesn't know by learning from the best, rather than the mediocre
- When looking for exploitability, pick your battles
- Even when you've found a problem that you think is exploitable, you might be wrong
- Don't go after the first inadequacy you see
- Coming up with a brand new model is something that you'll be lucky to do once or twice in a lifetime
- Coming up with a new synthesis of pre-existing ideas is something that you'll see once or twice a year
- Picking sides between experts when you can follow their arguments is something that you ought to be able to do quite frequently
- Fortunately, for most day-to-day decisions, the latter is quite sufficient
- To improve everyday thinking about inadequacy
- Update hard every time you come across new data
- Don't worry so much about overcorrecting, because you'll often quickly receive additional data that will prompt you to correct back
- Bet real money on everything - the sting of losing money helps you learn
- Being a "fox" shouldn't preclude you from having overarching theoretical frameworks
- You have to be ready to say that those frameworks were in error and update
- Having a theory doesn't lock you into being insensitive to evidence
- The ideology of empiricism is harmful when it blocks you from making hypotheses
- Being cognizant of the outside view is useful, but you have to make sure that you're truly comparing comparable things
- If you have a new product, it may or may not be applicable to consider related products
- Your market and customers may or may not be exactly the same as the market you're using as a baseline
- In truly novel situations, the outside view usually fails
- Moreover, in many ambiguous situations, it's not actually clear which "outside view" is correct - a new product or situation is often comparable to multiple reference classes
- In many cases, the outside view can't compete with a good model
- Most evident in the sciences
- Physicists build mathematical models of particles - don't say that particle x is like particle y, therefore it should behave the same
- You need both the ability to make theories and the ability to abandon those theories when they're proven wrong by evidence
