explore vs exploit

date initiated: 2023 September 13
last updated: 2024 June 29

In college, a friend patiently explained the concept of breadth vs depth-first search, and I thought it was a cute concept.

A number of years later, in The Origins of Order, I learned of the explore vs exploit paradigm, and somehow, even though they feel largely isomorphic, that has superceded breadth- vs depth-first search in terms of mental references/utility.

Perhaps it situated it better in terms of tradeoffs?

I notice that I’m generally biased towards exploration over exploitation, for some reason – maybe because a lot of the most meaningful things in my life were not explicitly planned for. I need to intentionally build in deadlines/incentives to get myself to finish things.

Meanwhile, I get the impression that exploitation is the more default/dominant strategy amongst ambitious people – the incentives of school and work are in executing a defined goal well.

I don’t think one is inherently better than the other; as with most things, it depends on context. But I tend to find it useful to make salient which tradeoff you prefer for a particular context, and calibrate your expectations appropriately.

In the exploitation paradigm, I think, works best when there is relatively minimal unknown unknowns relevant to completing the project.

In the exploration paradigm, you want to optimize for low stakes, to remove external pressures and considerations so that you can better focus on the internal sensor for ‘what is interesting/sparkly – I find that it is a relatively quiet signal. You want things to be low stakes so that you don’t judge an idea’s value prematurely, giving space for unknown unknowns. And it is often not very efficient; it is a longer-term strategy. The scenic route, if you will.

This topic became more salient recently because it’s been coming up in discussions with H. For example, knowing if it is an engineering (exploit) or research (explore) problem is important if you’re going to be working on a startup. And why aren’t more VCs independent thinkers? One of my guesses is that in the beginning, using your own naive models is worse than following the average/norm. It takes time to refine your model to one that’ll work better than the average (this also isn’t necessarily guaranteed). And since people tend to operate on a shorter time frame, the optimal choice is to exploit the norm.

I said that one isn’t inherently better than the other, but I think exploration is currently undervalued. Stumbling across unknown unknowns helps develop your faith in God. And in some sense, maybe that’s what independent thinking is – trusting your God, for better and worse.

[ ]