Minding Our Way – Deliberate Once

Link post

Here’s a question I get asked pretty regularly:

OK, I’m sold on this whole “do the best you can” thing, but how do you actually commit? When I look at my available options, none of them look great. I can take the one that seems best (despite its flaws), but then I keep doubting myself the whole time, and wondering whether there isn’t something better I could be doing. For example, I’m currently [doing a PhD|running a startup|earning to give|working at an EA org], and I keep wondering whether I should instead be [switching majors|running a startup|earning to give|working at a different org], and I can’t access conviction or resolve. How is it that you actually commit to what you’re doing?

My solution is fairly simple: Deliberate once, and then don’t deliberate again until you get new information that would have changed the result of your deliberation.

If you have a big decision to make, set aside some time to do some serious thinking. Ask people you trust for the obvious advice, and then actually do it. Set some five-minute timers, spend time describing the problem before you spend time describing possible solutions, brainstorm a wide variety of solutions. Figure out where you’re highly uncertain, find the places where your decision turns on a critical piece of information you’re missing, and add actions like “run thus-and-such an experiment” to your set of available actions. Build models. Make your best guesses about the probabilities of various outcomes in response to your actions, make your best guesses about how good those outcomes are, multiply out an expected value calculation, throw the numbers out, and consult your updated intuitions. And so on.

Whatever your “think seriously” process is, run it, and then pick the best action available to you (given the information you currently have, having taken into account the incompleteness of your situation). Then do that. Then don’t go back into deliberation mode until you encounter information that would have actually changed your answer.

If you do it right, there’s no need to go back to worrying about which thing you should be doing unless you encounter new evidence or information. When you do, deliberate again — or, better, plug the new evidence into the model you built when you deliberated the first time, instead of re-doing all that work. Half the reason for really deliberating hard the first time is to build a versatile model, anyway. (Though of course, evidence and experience will also cause you to update that model as you go.)


Many people seem plagued by “am I doing the right thing?” thoughts which get in the way of them committing themselves to any action. I think there are at least three different types of “am I doing the right thing?” thoughts, each worth treating differently.

First are the “but I am uncertain!” thoughts, which I refer to as “commitment-aversion thoughts”. They appear when someone isn’t confident in what they’re doing in an absolute sense, regardless of whether they see any good alternatives, and the above advice is more or less advice geared towards shifting you into a mode where you aren’t getting pinged with “but I am still uncertain!” thoughts all the time. However, it’s important to distinguish the commitment-aversion thoughts (which I think are often in error) from two other similar-sounding thoughts which are quite important.

Next are the “I just got new evidence that we’re doing the wrong thing!” thoughts, which I refer to as “abort notices.” Perhaps you decided to complete your econ PhD, and, three months in, you observe that you’ve been demotivated the whole time and that you’ve made almost no progress (which comes as a surprise to you). Or perhaps your friend from silicon valley tells you how high programmer salaries are these days, and that information would have changed your decision if you had know it six months ago, and you think “maybe it should also change my decision now.” Or perhaps you read a blog post on the internet about a cognitive bias that you notice was distorting your decisions, and you are pretty sure that, without the bias, you’d be doing something else instead. Dealing with these sorts of abort notices is easy, so long as you don’t quash them: Simply change your action, or (if you need to) go back into “heavy deliberation” mode. In these cases, be especially wary of the sunk cost fallacy and consider taking measures to avoid it.

Third are the “I have a vague sense that I deliberated incorrectly but I can’t articulate it yet” thoughts, which I refer to as “confusion pings.” These thoughts are very important, and they can be pretty hard to distinguish from commitment-aversion thoughts. Make sure that you don’t steamroll your confusions in an attempt to squash commitment-aversion thoughts. Noticing and respecting confusion pings is a core rationalist skill, and if some part of you feels an ineffable hard-to-articulate concern with what you’re doing, then the solution is not to shove it under the rug, the solution is to pay very close attention to it. Leave yourself a line of retreat, ask yourself what thoughts you’re not allowing yourself to think and what actions you’re not allowing yourself to consider, and ask yourself questions like “If I changed the course of my life and looked back on this moment in two years, what would I think I was missing?” Human deliberation is a flawed and fallible process, and confusion pings are sometimes the only hint you ever get of a giant gaping blind spot in your deliberation process. In my experience, handling confusion pings (and learning to articulate your inarticulable concerns) is a skill that requires practice, and it’s a skill that I have found very valuable.

On my current best model of where the “am I doing the right thing?” thoughts come from, they tend to come from one of two places. Either (a) some part of you is concerned that your deliberation was dangerously biased, flawed, or otherwise invalid; or (b) some part of you is not yet comfortable in the face of uncertainty. On this model, I suggest a two-pronged method for dealing with commitment-aversion thoughts. First, treat them as if they might be confusion pings: Learn to inspect them and extract content from them, especially content of the form “I think I have been ignoring factor X” or “I think I have been under-weighing concern Y.” Second, learn to be comfortable in the face of high uncertainty, and develop a deliberation procedure of your own that deserves appropriate meta-confidence.

Operating under uncertainty is the norm, not the exception. No matter how hard you deliberate, your deliberation procedure is going to be flawed and biased, and there are going to be considerations you’re missing and evidence you failed to take into account. It is very hard to predict the consequences of your actions, and you aren’t going to get a perfect answer. However, some answers are still better than others, and you can construct a process for choosing between actions that leaves you comfortable that you’re doing the best you can do with the information and time available.


When you’ve committed yourself to a new action, and you start wondering whether you’re really doing the right thing, the relevant question is not “were my thoughts when I chose this action perfectly unbiased?” Nor is it “did I find literally the best available action?” No, the relevant question is, “am I in a better position now to pick a good action than I was then?”, or alternatively, “could I do a significantly better job picking an action this time around than I did last time around, enough so to make up for the opportunity cost of deliberating again?”

The only information that affects your actions is information that changes which action you would think is better after deliberating. This is true regardless of how much uncertainty you have. Think hard, pick the best action, and then don’t worry about what action you’re doing until you see something that would have changed the result of the hard thinking. If you understand this on a gut level, and if you’re one of the people who has found my advice helpful in the past, then I predict that this mindset is one where you won’t experience constant doubt about whether you’re doing the right thing. Deliberate well once, and then don’t deliberate again until you come across new information that would have changed your answer.