Causality principles

How to deal with incoming signals

Recall one of my previous recurring tropes - the distribution of X’s and O’s in a population. My previous piece cast the X’s in a population as the rational ones and the O’s the irrational zeros. Here I will take it one step further.

I will posit that each X in a population, due to their unique rationality offers everyone else a certain useful signal. Think about the various authors on Substack, fiction writers or even fashion models. Now we all can be X’s or O’s at differing times but ultimately each of us chooses one or the other to be most of the time. Think of the X’s like signal generating machines and the O’s like the followers of those signals.

Now expand the metaphor even further into the field of econometrics and machine learning. Let each X not be a rational thinking human with all our faults and randomness, but instead a steady signal generating variable. Call this the x-variable.

In the field of causal econometrics, what we require ultimately in this field is to establish a causal relationship between the x-variable and the output. The aim is to find the x-variables that affect the output the most or have the closest relationship with. The metaphor can be brought into life when we make our daily choices - as rational human beings we aim to identify the factors that have the closest connection and affect us most deeply as human beings.

Spinoza articulates this the best. By understanding the true causes (the "common properties" of things) through reason, we move from being controlled by external passions to being the active, free cause of our own states, leading to lasting joy and freedom. This is one aspect of the red pill- to expose ourselves to the true causes of things which would move us and spur us to action the most rather than what we mistakenly think we want to be the cause. And in it also lies a form of power. We all want these X’s in our lives (and for some of us also to be X’s to others).

The principles of causality can help us extricate the most useful signals from the noise. Haven’t you had the annoying feeling that something that appeared like a useful signal to you at first glance is in fact a confounded variable in disguise? That X was correlated with another variable or the error term and that made the signal appear bigger than it actually was. Causality wants us to make sure variables are exogenous or in other words, not have entanglements with other variables or X’s to derive its true causal impact on your output. The problem can get so bad that one (myself included) can end up ignoring all X signals, as it is so difficult to separate the wheat from the chaff.

My current directive is to ignore all signals or at least reject them in good humour until a proper channel has been established that can establish a true causal link between incoming signals. Once the channel has been established, the X’s will reveal themselves to you in a calm and profound way.

(Note: Causality ultimately is a powerful but one-dimensional tool. It is highly clunky and inflexible and requires much more computational power. Machine learning techniques are the new advancement that allows more flexible model fitting techniques that is many times less frustrating than purely relying on the principles of causality. More on machine learning advancements later.)