From Theory to Reality

As long as I can think back, I was motivated by a desire to find out about the inner workings of things. Not in the physically inspired sense of wanting to dissect objects and put them back together, like when you repair something. More in a purely curiosity-inspired sense of enjoying that moment of satisfaction when you have figured out a riddle or witness something behaving as expected, giving you that impression that you know something about at least a tiny part of the world and can use that knowledge to make sense of other parts or play around with in your head, doing mind-experiments, so to say.

Particularly interesting in that realm are systems, where you have various small, easily understandable building blocks which are put together in various ways. To see what happens when you plug together a few rules and objects with their individual motivations and properties interacting with each other and then letting it play out never becomes boring.

It does not matter whether it is board games (the more complex the more interesting), economy, politics, organization of a team or a company, or machine learning … The more diverse the problems are, the better. While at first it seems that each requires a different tool-set (and the figureheads of those fields make it seem so in a move to dig a moat around their expertise), I found that techniques taught in mathematics, physics and engineering were applicable to all of them. First and foremost, though, this also applies to my personal favorite: control theory.

I am of the stout belief that all we do when building a thing, company or society is to engineer feedback loops. Be it in the cadence of having elections, feedback talks with your superiors, having warning lights reminding you that your car engine needs oil or regularly check in with your partner how they feel.

It is all about setting up the right signals to learn about the state the system you have built is in and to be able to react, based on your previously built understanding of said system.

And that is where for me, it all comes together: You try to understand something, then you try to make use of that and build something - like an algorithm that tries to predict what happens next in any given esports match - with special care taken towards the feedback loops. When you flip the switch to test that algorithm out, I am always amazed that reality complies, and behaves as expected. Reality seems to (roughly) work according to our mathematical and physical models, and it always is surprising to me that it does.

Realizing your dream-job usually requires careful navigation between the poles of the enjoyable and the useful, which in my case is an easy cruise. At Bayes, I was very lucky to find a place where what I love and what the company needs overlap. I get to play around with ideas, bounce them off of very clever people which bring their own twist to things and thus digging always a tiny bit deeper into understanding a complex, feedback-driven system like competitive video games and eventually build a system which puts all that knowledge to the test, by predicting on a live-game what is about to happen, done by a complex, distributed state-of-the-art system. And to add a meta-layer to this: The prediction system itself is of course a sort of game, with its own feedback cycle telling us how good we were and where we could become better.

So, as a matter of fact, I do quite like Mondays!

‍

‍