Actually Practicing Rationality and the 5-Second Level

[I first posted this as a link to my blog post, but I’m reposting as a focused article here that trims some fat of the original post, which was less accessible]


I think a lot about heuristics and biases, and I admit that many of my ideas on rationality and debiasing get lost in the sea of my own thoughts. They’re accessible, if I’m specifically thinking about rationality-esque things, but often invisible otherwise.

That seems highly sub-optimal, considering that the whole point of having usable mental models isn’t to write fancy posts about them, but to, you know, actually use them.

To that end, I’ve been thinking about finding some sort of systematic way to integrate all of these ideas into my actual life.

(If you’re curious, here’s the actual picture of what my internal “concept-verse” (w/​ associated LW and CFAR memes) looks like)

MLU Mind Map v1.png


Open Image In New Tab for all the details

So I have all of these ideas, all of which look really great on paper and in thought experiments. Some of them even have some sort of experimental backing. Given this, how do I put them together into a kind of coherent notion?

Equivalently, what does it look like if I successfully implement these mental models? What sorts of changes might I expect to see? Then, knowing the end product, what kind of process can get me there?

One way of looking it would to say that if I implemented techniques well, then I’d be better able to tackle my goals and get things done. Maybe my productivity would go up. That sort of makes sense. But this tells us nothing about how I’d actually be going about, using such skills.

We want to know how to implement these skills and then actually utilize them.

Yudkowsky gives a highly useful abstraction when he talks about the five-second level. He gives some great tips on breaking down mental techniques into their component mental motions. It’s a step-by-step approach that really goes into the details of what it feels like to undergo one of the LessWrong epistemological techniques. We’d like our mental techniques to be actual heuristics that we can use in the moment, so having an in-depth breakdown makes sense.

Here’s my attempt at a 5-second-level breakdown for Going Meta, or “popping” out of one’s head to stay mindful of the moment:

  1. Notice the feeling that you are being mentally “dragged” towards continuing an action.

    1. (It can feel like an urge, or your mind automatically making a plan to do something. Notice your brain simulating you taking an action without much conscious input.)

  2. Remember that you have a 5-second-level series of steps to do something about it.

  3. Feel aversive towards continuing the loop. Mentally shudder at the part of you that tries to continue.

  4. Close your eyes. Take in a breath.

  5. Think about what 1-second action you could take to instantly cut off the stimulus from whatever loop you’re stuck in. (EX: Turning off the display, closing the window, moving to somewhere else).

  6. Tense your muscles and clench, actually doing said action.

  7. Run a search through your head, looking for an action labeled “productive”. Try to remember things you’ve told yourself you “should probably do” lately.

    1. (If you can’t find anything, pattern-match to find something that seems “productive-ish”.)

  8. Take note of what time it is. Write it down.

  9. Do the new thing. Finish.

  10. Note the end time. Calculate how long you did work.

Next, the other part is actually accessing the heuristic in the situations where you want it. We want it to be habitual.

After doing some quick searches on the existing research on habits, it appears that many of the links go to Charles Duhigg, author of The Power of Habit, or B J Fogg of Tiny Habits. Both models focus on two things: Identifying the Thing you want to do. Then setting triggers so you actually do It. (There’s some similarity to CFAR’s Trigger Action Plans.)

B J’s approach focuses on scaffolding new habits into existing routines, like brushing your teeth, which are already automatic. Duhigg appears to be focused more on reinforcement and rewards, with several nods to Skinner. CFAR views actions as self-reinforcing, so the reward isn’t even necessary— they see repetition as building automation.

Overlearning the material also seems to be useful in some contexts, for skills like acquiring procedural knowledge. And mental notions do seem to be more like procedural knowledge.

For these mental skills specifically, we’d want them to go off, time irrespective, so anchoring it to an existing routine might not be best. Having it as a response to an internal state (EX: “When I notice myself being ‘dragged’ into a spiral, or automatically making plans to do a thing”) may be more useful.


(Follow-up post forthcoming on concretely trying to apply habit research to implementing heuristics.)