Break your habits: be more empirical

Summary: The common attitude that “You think too much” might be better parsed as “You don’t experiment enough.” Once you’ve got an established procedure for living optimally in «setting», be a good scientist and keep trying to falsify your theory when it’s not too costly to do so.

Note 1: I originally posted this years ago on LessWrong, and it was my inspiration for the “Try Things” meme in the CFAR curriculum.

Note 2: in aspects of life where you’re impulsive, don’t introspect enough, or have poor self-discipline, this post is probably advice in the wrong direction.


Consider an Alice who is highly analytically minded. She always walks the same most-efficient route to work, only dances tango and salsa, and refuses to deviate even on rare occasions from her well-planned schedule. She has judged carefully from experience that the expected value of dating is too low to be worth her time, and will only watch a movie if at least 3 of her 5 closest friends recommend it. She travels only for work, to ensure the trip has a purpose and to minimize unnecessary transportation costs. Oh, and she also thinks a lot. About everything.

Bob often tells Alice that she “thinks too much” — advice that rarely, if ever, resonates….

But consider that Bob may be sensing a legitimate imbalance: Alice may be doing too much analysis with not enough data.

Bob can tell that Alice thinks way more than he does, and blames that for the imbalance, suggesting that Alice should “turn off her brain”. But Alice can’t agree. Why would she ever waste a resource as constantly applicable and available as her mind? That seems like a terrible idea.

So here’s a better one: Alice, if you’re reading this, don’t turn your mind off… turn it outward.

When (analysis:data) looks too big, just try turning up the data. There’s no need to get stupider or anything. When it’s not overly costly, you should deviate from your usual theories of optimal behavior for the sake of expected information gain. Even in theory, empiricism is necessary… For a Bayesian optimizing agent in an uncertain world, information has positive expected utility, and experiments have positive expected information. Ergo, do them sometimes!

What I mean is that probably, once in a while, Alice should dance freestyle. She should leave early and take a scenic route sometimes, and try a some new food along the way. She should visit somewhere she’s never been for a vacation, and try meeting some locals. And she needn’t feel discouraged when experimental behavior turns out to be “suboptimal as anticipated”. That just means she doesn’t have to try that particular thing again, at least for a while. The point is the rare occasion when an experiment does work out and you find something valuable that you can keep doing, or the less-rare occasion that the change of pace is simply inspiring.

So try to overcome that deep-rooted sense of suboptimality you get when you consider new things, or revisit old ones. Locally suboptimal behavior can be worth it for the global benefits of the information you gain. I’m not suggesting to take big risks like drug addictions or injuries… if you want a safe idea to start with, think of something you never do but which other people do all the time without ruining themselves.

A priori, classical mechanics could have explained pretty much any observation made before 1800. It looked great: every event could be imagined as a series of pushes and pulls acting on sufficiently small bits of matter and the right initial conditions. But we kept testing it anyway, and now we have nuclear power. What might you find in a new situation or environment that you never thought of before?

Go do something you wouldn’t normally do 🙂

Leave a Reply

Your email address will not be published. Required fields are marked *