Feb 7, 2025
Say we take predictive coding seriously and accept that the chief goal of a human is minimizing prediction error. Does this imply anything useful about how we should live our lives? Can we use concepts from these beautiful concepts from the intersection of neuroscience and deep learning to write pithy new self-help books? I think we can.
(If you’re unclear on what predictive coding is, read Scott’s post about it first, I promise it’s excellent).
(which you can also rephrase as “doing things to maintain homeostatis/equillibrium - not too cold, not too hot, not too hungry, not too full, etc”). I think this line of thought lets us derive some good life advice, some of it obvious and some of it less so.
Your body’s chief objective is maintaining homeostasis - being not too cold, but not too hot, not too hungry, but not too full, keeping blood pressure not too low but not too high, and so on, staying in the narrow range of parameter combinations which are conductive to life. In predictive coding we could phrase this as a) having a prediction that we’re in this ideal state and b) trying to minimize prediction error (deviations from this ideal state). In a sense, you are just a very complicated thermostat.
I think some interesting advice comes from taking this literally: trying to be a good thermostat, consciously helping your body stay in its sweet spot.
TODO
Or what things are “not right’ at a given moment, or in what ways it’s “too hot”. You probably want to consciously practice the process of iteratively feeling out what temperature, trying something to get closer to perfect homeostatis, repeating. You can do this at multiple levels.
For example: in what ways is your body uncomfortable at this moment? How could you nudge your posture in the direction of being more comfortable? Given some particular feedback from your body (a specific kind of soreness or discomfort or urge to move) what can you do to relieve it? What does it feel like when you’re moving in the right direction, when you successfully get more comfortable or stretch out a sore muscle? See here for more along this line of thought.
Or you could do something similar with food, trying to feel into how given food makes you feel, getting good at listening to what your body wants.
or: keeping things in your cache is expensive. do TODOs quickly
Or at another, more unobvious level: in what ways is your life direction misaligned with what you feel it should be? In what ways are you not living up to your expectations? What things do you keep doing that some part of you feels are bad?
The problem here is that the thermostat/predictive coding thing is too low level to give actionable advice here. What it tells you is almost trivially true - sure, I am deep down a prediction error minimization machine, but that’s not very helpful way to think when picking a career (will a doctor or a programmer experience less prediction error? I don’t know!) So naturally you want some kind of higher level heuristics - i.e. boring old life advice which it’s also not always obvious how to follow, like “follow your passion” or “get a good career”.
If the thermostat framing still offers some advice in these more nontrivial cases, I think it’s something like: take seriously the project of figuring out where you are in life and what you want. Contemplate this deeply and without prejudice: your world model / set point is probably a hacky mess and includes parts which are silly (like a stuck prediction that you’re an astronaut) or which you find distasteful (repressed dark sexual urges). Try to untangle this mess through some kind of intentional meditation practice, either adjusting your beliefs or taking actions in the world to change your life. Taking a step in the direction of perfect homeostatis will involve some things getting better and some getting worse. This is normal; have faith in the power of gradient descent through high-dimensional spaces.
An important thing to understand is that in these more non-trivial cases, you really have two levers to pull: you can change your beliefs to match reality, or change the world to make your beliefs come true. The important thing is the alignment between expectation and reality.
More speculatively, I think this is a good argument for following some kind of Holy Book, be it the Bible or the Bhagavad Gita or How to Win Friends and Influence People or whatever. People have the urge to take advice from books piecewise, trying to take the good and leave the bad. But if you adopt the whole thing, that’s likely a nicely internally consistent set of beliefs, and by adopting it somewhat axiomatically you simplify the problem of deciding whether to change beliefs or to change behavior - you usually change behavior. Maybe even having a mediocre book as your Holy Book is better than nothing for achieving this alignment.
This is much harder than you think.
First, because changing beliefs is generally difficult, and completely refactoring all your beliefs is tremendously difficult, to the extent that becoming Good within your current understanding of it is actually easier.
Second, because your beliefs are tied up with other beliefs which are hard-coded. If you don’t eat for a while you will get hungry, good luck believing you’re not. Similarly, your desire for social approval and physical touch and sex are also difficult to manually override. But your complex beliefs are “grounded” in these hard-coded beliefs! You want a good career because you think then people will respect you, which will give you social validation, which you want because it’s associated with opportunities for sex, which is associated with physiological/chemical processes which you are basically wired to want. In reality this chain is some complicated chain of uninterpretable black-box bayesian inferences, but the result is the same: it’s difficult to wirehead yourself into being deeply content with your life.
really, this idea could be rephrased as ‘contemplate what the Good is, and follow it’.