Virtuous Cycles

May 17, 2021

When you are trying to do something hard – become the best in the world at something, start a company, master an unfamiliar domain – your own current inadequacy dominates your prediction of the future. You project failure, lose motivation, and quit. The dangerous part: you will look correct to yourself, because your lack of motivation caused the failure you predicted.

A virtuous cycle is one where each stage of the process improves the next, and the last stage feeds back into the first. Self-play in machine learning is the clearest example: an agent plays against itself, its failures become signal, the signal improves the agent, and a slightly better agent plays against itself again. After enough iterations, the agent surpasses the best humans. The key property is that improvement compounds.

Recognizing that you are inside a virtuous cycle changes the prediction.

In a situation like that the task is so hard that it isn’t reasonable to expect that you will achieve your goal in a short time frame. Yet you think in real time. You can’t afford the cost of fully thinking through the consequences of everything that you need to do. So you are going to end up making an approximation.

If you don’t recognize that virtuous cycles exist then you are likely to make a bad approximation. Your approximation is going to be conditioned on where you are versus where you expect to be, but your own current inadequacy is going to dominate the prediction. It will be easy to project a future which looks similar to the past, but if you don’t already have a history of success to motivate you this will result in a lack of motivation. That same lack of motivation can lead to giving up on the hard thing.

The scary thing here isn’t that you were wrong: you aren’t wrong to project your failure. The danger is that you will look correct to yourself. When you give up because of a lack of motivation you are justified in doing so because your lack of motivation is causally connected to your failure. You don’t just fail. You fail because you thought you would fail.

A recognition of virtuous cycles produces a different outcome. While you can’t know the whole future you can take a shortcut by appealing to the properties of virtuous cycles to know that improvement occurs and that improvement is substantial. Now, despite being limited in time, you have a cost effective way for your brain to make a decent prediction about the future that suggests success. So now you don’t give up, because of a lack of motivation. You go on to accomplish all that you dreamed and more.

Now your prediction also looks good in hindsight.

Notice that you are going to think you are right whether you go down the path without understanding and the path with understanding. In both cases your overarching prediction is justified by your most probable experience, because your most probable experience is a function of what you do and what you do is conditioned on your expectations about what the future holds for you.

So understanding of the cycle is critical for success. The difference between failure and success was perseverance when operating under a virtuous cycle. You’ve likely heard it said before that the trick to succeed is to keep failing until you don’t. That is because learning is a type of virtuous cycle wherein your failures are signal which improves your future actions. Wise people recognize this and love getting signal about their failures. People who don’t recognize this see their failures in a different way.

When failure and weakness isn’t opportunity, people can think of it like damage. Something to defend themselves from. Something to avoid. They can quit going down the route they were going down. Worse, they are no longer inside the virtuous cycle when they do so. A kind of inevitability of success is now lost to them, because they didn’t recognize the processes that were operating on them.

Since recognizing that cycles exist and their properties are critical to our success it would be good to be able to discriminate between the cycles we are in. In order to do so I have two tools that I used to understand what is happening in a cycle.

The first is measurement. Measuring what is actually happening tells you what the cycle is actually doing at the present moment and in the past. If you do that for a while you end up getting a signal that is telling you how things are going.

What it doesn’t tell you, but what it seems to tell you, is how things will go in the future. For that historical data can be misleading. A trick I’ve found to getting around this is to work backward. Just assume that whatever is rewarded is going to be the thing that ends up happening. If the reward indicates that you are going to trend in the direction you want to go and the measurements are also saying you are heading in that direction then you have a lot of reason to be confident that the cycle is actually taking you in the direction you think it is.

This gives you a tool. You can build virtuous cycles deliberately. The diagnostic questions are simple: Are there things you should be rewarding but aren’t? Are there places where feedback loops should be created? If you are already in a cycle, is there a way to go through it more quickly?

Knowing without applying means little. So get going.