Ergodicity and Survival


I’ve been thinking more about Monday and Tuesday’s blog posts and how they relate to ergodicity.

What is ergodicity? The simple statement is that the space average is the same as the time average. Which means that if you look at one point over time you’ll get the same answer if you look at all of space at one time.

That’s very convenient when trying to figure out something - it’s hard to run an experiment when you have to wait for all of time to see what the answer is. It’s much quicker to look everywhere at once.

Since I come from a theoretical chemistry background, we spend a lot of time trying to proving that a system is ergodic. For molecules, that’s usually true. But for people, it’s not. And the funny thing is that our cultural heuristic is that everything is ergodic - It’s not! We have a bad set of assumptions.

I wrote about this once before in the context of strength training. The place where ergodicity breaks down is when being wrong takes you out of the game. This is also what makes a population better and stronger (at the cost of individuals).

In the context of strength training, if you do the wrong thing and get injured, it’s bad for you. But for the population, we can learn what you did and not repeat that (unless you’re stuck in the conventional wisdom that sore shoulders and knees is a good thing because you’re working hard). That’s why I’ve gravitated to a kettlebell-heavy programming style. They tend to get better results with a lower risk of taking you out of the game.

When the cost of running an experiment is really low and the payoff for being right is huge, taking advantage of non-ergodicity is the best way to win. Think about VC firms in the valley. They invest in everything that comes their way in the hopes that they run into one Uber that explodes.

It’s harder for us in the health and fitness world because the cost of experimenting is a little more. But the internet can help us because the world can see all the experiments and see what fails (USDA nutrition advice, e.g.). That’s the one place that CrossFit did it correctly. Instead of having top-down franchise-like control, they let all the gyms experiment and we saw what succeeded and who destroyed their shoulders by too many kipping pull-ups.

The problem was that what was considered “success” in CrossFit was producing games athlete, not making regular people fit and healthy.

In fact, most of the things we know comes from running many low cost experiments and harvesting the winners - as opposed to directed research in search of a specific answer to a specific problem. The narrative is that science goes looking for something, but in reality it’s more random tinkering finds something and then science goes back to figure out why it worked.

That’s why I run lots of experiments on programming. I usually have 2-3 experimental programs running at any given time for my advanced students. Things that survive the advanced students come to the regular classes. If it’s no fun or not terribly effective, I throw it away and we don’t see it again. If we like it, I remember and bring it back.

That’s why we’re doing Olympic lifting this fall. Everyone loved that and wanted it back, so we’re doing it. We’ll experiment with a new idea in January and see how that goes. And then I have some of my favorites to revisit in the spring cued up.

Michael Deskevich