# Orthogonality

post by lsusr · 2020-05-21T03:17:36.365Z · LW · GW · 3 commentsTL;DR: I used to think the best way to get really good at skill was to specialize by investing lots of time into . I was wrong. Investing lots of time into works only as a first-order approximation. Once becomes large, investing in some other produces greater real-world performance than continued investment in .

I like to think of intelligence as a vector where each is a skill level in a different skill. I think of general intelligence as the Euclidean norm .

I use the Euclidean norm instead of the straight sum because **generality of experience equals generality of transference**. Suppose you are exposed to a novel situation requiring skill . You have no experience at so you must borrow from your most similar skill. The wider a variety of skills you have, more similar your most similar skill will be to .

The best way to increase your general intelligence is to invest time into your weakest skill . **If your invested time for your strongest skill is already high then investments in can also increase the real world performance of your strongest skill faster than investments in .**

Suppose you want to increase , your real world performance at . . Investing time into always results in increasing . But eventually you will hit diminishing returns. For every there exists a such that if then .

Here's where things get interesting. "All non-trivial abstractions, to some degree, are leaky" and a system is only as secure as its weakest link; cracking a system tends to happen on an overlooked layer of abstraction [LW · GW]. All real world applications of skill are non-trivial abstractions. Therefore performance in one skill occasionally leaks over to improve performance of adjacent skills. Your real-world performance at leaks over from adjacent skills on rungs above and below on the ladder of abstraction.

These adjacent skills increase your real world performance on by a quantity independent of . Since , there will inevitably come a time when increasing increases less than increasing .

It follows that quantity of avocations correlates positively with winning Nobel Prizes, despite the time these hobbies take time away from one's specialization.

When I want to improve my ability to write machine learning algorithms, my first instinct is to study machine learning. But in practice, it's often more profitable to do something seemingly unrelated, like learning about music theory. I find it hard to follow this strategy because it is so counterintuitive.

## 3 comments

Comments sorted by top scores.

I think that a hidden assumption here is that improving in a weak skill always has a positive spillover affect on other skills. There might be a hidden truth within this. Namely, sometimes unlearning things will be the best way to make progress.

Perhaps this can be connected with another recent post. It was pointed about in Subspace Optima [LW · GW] that when we optimize we do so under constraints external or internal. It seems like you had an internal constraint stopping you from optimizing over the whole space. Instead you focused on what you thought was the most correlated trait. This almost reads like an insight following the realization you’ve been optimizing a skill along a artificial sub-space.

It follows that quantity of avocations correlates positively with winning Nobel Prizes, despite the time these hobbies take time away from one's specialization.

An alternative explanation is that there are factors that both allow having hobbies and increase the probability of winning a Nobel Prize. Things like "having lots of free time" and "not having to worry about various problems".

It is true that hobbies and specialization compete against each other for time and attention. But there are also other things that compete with both hobbies and specialization; and if you succeed to eliminate them (e.g. by being rich, having a spouse that takes care of everything, or living an ascetic life), it provides more space to both.