[Previous] Open Discussion (2019) | Home | [Next] Andy Discussion

Errors Merit Post-Mortems

After people make errors, they should do post-mortems. How did that error happen? What caused it? What thinking processes were used and how did they fail? Try to ask “Why?” several times to get to deeper issues than your initial answers.

And then, especially, what other errors would that cause also cause? This gives info about the need to make changes going forward, or not. Is it a one-time error or part of a pattern?

Effective post-mortems are something people generally don’t want to do. What causes errors? Frequently it’s irrationality, including dishonesty.

Lots of things merit post-mortems other than losing a debate. If you have an inconclusive debate, why didn’t you do better? No doubt there were errors in your communication and ideas. If you ask a question, why were you ignorant of the answer? What happened there? Maybe you made a mistake. That should be considered. After you ask a question and get an answer, you should post-mortem whether your understanding is now adequate. People usually don’t discuss thoroughly enough to effectively learn the answers to their questions.

Regarding questions: If you were ignorant of something because you hadn’t yet gotten around to learning about it, and you knew the limits of your knowledge, that can be a quick and easy post-mortem. That’s fine, but you should check if that’s what happened or it’s something else that merits more attention. Another common, quick post-mortem for a question is, “I asked because the other person was unclear, not because of my own ignorance.” But many questions relate to your own confusions and what went wrong should be post-mortemed. And if you hadn’t learned something yet, you should consider if you are organizing your learning priorities in a reasonable way. Why learn this now? Why not earlier or later? Do you have considered reasoning about that?

What if you try to post-mortem something and you don’t know what went wrong? If your post-mortem fails, that is itself something to post-mortem! Consider what you’ve done to learn how to post-mortem effectively in general. Have you studied techniques and practiced them? Did you start with easier cases and succeed many times? Do you have a history of successes and failures which you can compare this current failure to? Do you know what your success rate at post-mortems is in general, on average? And you should consider if you put enough effort into this particular post-mortem or just gave up fast.

You may wonder: We make errors all the time. Should we post-mortem all of them? That sounds like it’d take too much time and effort.

First, you can only post-mortem known errors. You have to find out something is an error. You can’t post-mortem it as an error just because people 500 years from now will know better. This limits the issues to be addressed.

Second, an irrelevant “error” is not an error. Suppose I’m moving to a new home. I’m measuring to see where things will fit. I measure my couch and the measurement is accurate to within a half inch. I measure where I want to put it and find there are 5 inches to spare (if it was really close, I’d re-measure). The fact that my measurement is an eighth of an inch off is not an error. The general principle is that errors are reasons a solution to a problem won’t work. The small measurement “error” doesn’t prevent my from succeeding at the problem I’m working on, so it’s not an error. It would be an error in a different context like doing a science experiment that relies on much more accurate measurements, but I’m not doing that.

Third, yes you should try to post-mortem all your errors that get past the previous two points. If you find this overwhelming, there are two things to do:

  1. Do easier stuff so you make fewer errors. Get your error rate under control. There’s no benefit to doing stuff that’s full of errors – it won’t work. Correctness works better both for immediate practical benefits (you get more stuff done that is actually good or effective instead of broken) and for learning better so you can do better in the future.
  2. Learn and write down recurring patterns/themes/concepts and reuse them instead of trying to work out every post-mortem from scratch. If you develop good ideas that can help with multiple post-mortems, that’ll speed it up a ton. Reusing ideas is a major part of Paths Forward and is crucial to all of life.

Elliot Temple on April 8, 2019

Messages (9)

"You aren't done when the problem is gone. You're done with the problem will never come back." — Max Kanat-Alexander


Alisa at 11:11 PM on February 21, 2020 | #15586 | reply | quote

#15586 is a verbatim quote from Max's Twitter. I guess he meant to write: "when the problem will never come back", not "with the problem will never come back".


Alisa at 11:12 PM on February 21, 2020 | #15587 | reply | quote

https://www.scottaaronson.com/blog/?p=4695

Very good that Scott Aaronson not only changed his mind on coronavius and started taking it seriously, but is self-aware enough to be reflecting on why he was wrong and why he's been too trusting of the authorities in general. He's considering how he should change as a person/thinker.

This is something of an intellectual post-mortem – those are very important – and it also contains some good comments on the pandemic.


curi at 1:15 AM on April 5, 2020 | #16265 | reply | quote

#16265 SA adds more in comments:

> Incidentally, if anyone ever wondered why I despise SneerClub and woke Twitter, in ways that might seem wildly out of proportion to their actual importance in the world … well, you now have your answer. The people who get called clueless techbros, asshats, and a thousand other names on those forums were overwhelmingly represented among the people who turned out to be the clearest and rightest about the coronavirus from the beginning. And this is so directly consequential that, even if the rationalist techbro asshats were wrong about everything else in their entire lives (which I don’t think they were), being right this one time would more than cancel it out. Furthermore, the very qualities that the techbros get attacked for—e.g., arrogant reliance on their own reason and math abilities, refusal to defer to the authorities in other fields—were precisely the qualities that caused them to be unpopularly right this time. I might someday forgive the sneerers for targeting me personally, but I can never forgive them for so directly targeting our civilization’s error-correcting machinery, at what might end up being a huge cost in lives.


curi at 1:19 AM on April 5, 2020 | #16267 | reply | quote

#16267 SA wrote a bunch of comments. Here's another quote with some post-morteming and self-awareness:

> (And how should people update about me? I think the following would be fair: “He’s too fearful of embarrassment to serve as a reliable early warning system for civilization-wide catastrophes, not that he ever claimed to be anything more than a quantum complexity theorizing nebbish. On the other hand, like Bertrand Russell abandoning his pacifist absolutism in 1940, at least he’s able to recognize when others are righter than he is and course-correct.”)


curi at 1:23 AM on April 5, 2020 | #16268 | reply | quote

#16268 More SA updating his thinking:

> For my part, here’s what I’ve learned from this episode: that I’d rather trust a single medical expert who seems smart and honest and whose logic I fully understand, than an organization of a thousand medical experts that’s vulnerable to political and bureaucratic pressure.

Also SA:

> marxbro #19: I’m prepared to make one of the largest concessions that I’ve ever made to Marxism about anything. I think that right now, a command-and-control system like China’s would be preferable to what we have in the US.

jfc, fooled by China? Maybe he'll learn better in a few months when he finds out China has 50k+ dead and is lying. SA also hates and flames Trump, a lot, which I didn't quote.


curi at 1:27 AM on April 5, 2020 | #16269 | reply | quote

#16269 More self-awareness from SA:

> I’ll often spend hours rewording things to try to offend readers as little as I can, consistently with sticking to my principles


curi at 1:45 AM on April 5, 2020 | #16270 | reply | quote

Spending time rewording things to avoid offending readers

#16270

I wanted to know whether SA thinks that spending hours rewording things to avoid offending readers was good or a waste of time. I couldn't tell from the quote you gave. So I looked at the whole paragraph. Here's the paragraph.

> I disagree with at least some of the views of all the people I listed, certainly including Cochran. Beyond that, I often find Cochran’s style to be needlessly aggressive. Where I’ll often spend hours rewording things to try to offend readers as little as I can, consistently with sticking to my principles, Cochran often seems like a (much more intellectual) radio shock jock in his evident delight in offending anyone who he considers an idiot.

So SA contrasted his style ("spend hours rewording things to try to offend readers as little as I can") against Cochran's style ("needlessly aggressive"). He used the word "needlessly", which I think implies that he disagrees with Cochran's style and prefers his own style. So I think SA thinks it's useful to spend hours rewording things to try to offend readers less, rather than it being a waste of time.


GISTE at 5:44 AM on April 5, 2020 | #16273 | reply | quote

Decent post-mortem example:

https://www.lesswrong.com/posts/WKzGhKbtJAXntXEdB/sometimes-it-can-take-a-while-to-notice-confusion

also i think some of their covid precautions were high-effort optimization of non-bottlenecks which led them to be biased about how safe they were.


curi at 9:18 AM on April 12, 2021 | #20372 | reply | quote

Want to discuss this? Join my forum.

(Due to multi-year, sustained harassment from David Deutsch and his fans, commenting here requires an account. Accounts are not publicly available. Discussion info.)