What Gradient Descent teaches us about mediocrity
A few weeks ago, a friend and I had a long chat about ‘doing things for their own sake’ vs ‘doing things for an end-goal.’ It's the classic debate that makes you rethink everything you do in a day. My view has been that doing stuff for an end-goal is going to have a stronger ROI, and that doing stuff for its own sake is not.
But, I thought about it and wanted to see the other perspective of why doing things for their own sake may be better than doing them for an end-goal. So, here's the thing - In this post, I want to talk about mediocrity and why it may just be better than perfection.
If you know of gradient descent, read this; if not, skip to the next paragraph. So, my machine learning friends, take gradient descent. In that context, I learned that ‘mediocratizing’ is like staying on a hill, while optimizing is like going for the global minimum. The reason, then, that mediocratization is better than optimization is because mediocratization doesn’t assume that there are finite ends in an anisotropic (or random), continuous, unpredictable environment. By staying on the hill, you open yourself up to a host of opportunities you otherwise wouldn’t have come across. And when was the last time you did things for their own sake, anyway?
To play the finite game is to play the game to win; to play the infinite game is to continue playing the game itself. I’d always pick the latter, not the former. Mediocrity is playing the infinite game while perfection is playing the finite one. To perfect something means to only asymptotically approach perfection because if we were able to conceive of that which is perfect, we would know exactly when we got there. But we cannot describe — to painstaking detail — what perfection is, and so, we will never know when we get there.
In that case, optimizing for the global minimum will get us stuck in local minima. In other words, going for the best will prevent us from going for ‘good enough’ (something I talked about here.) It’s for the same reason why “done” is better than “best.” Look up ens?. In Zen philosophy, it’s a roughly drawn “circle.” I say “circle” because it doesn’t really matter if it even looks like a circle or if it’s even a complete circle. What matters is that you draw something, that, to your mind, is a circle.
Thinking about things this way — or even daring to think about things this way — is necessarily opening your mind to exponentially more creative outputs. I realized this when I read that we may barely even “finish” life itself, so there’s simply no point in trying to stress about finishing projects by optimizing for the global minimum.
In essence, you’re better off optimizing for mediocrity so you’re always open to the randomness (or anisotropicity) that the world throws at you. For when you’re optimizing to traverse many hills and go to different places, you will go places.
I'm still developing this idea.
Subscribe to Weekly Insights here.