Something I recently noticed: steelmanning is popular on LessWrong. But the sequences contain a post called Against Devil’s Advocacy, which argues strongly against devil’s advocacy, and steelmanning often looks a lot like devil’s advocacy. What, if anything is the difference between the two?
Steelmanning is about fixing errors in an argument (or otherwise improving it), while retaining (some of) the argument’s assumptions. As a result, the argument becomes better, even if you disagree with some of the assumptions. The conclusion of the argument may change as a result, what’s fixed about the conclusion is only the question that it needs to clarify. Devil’s advocacy is about finding arguments for a given conclusion, including fallacious but convincing ones.
So the difference is in the direction of reasoning and intent regarding epistemic hygiene. Steelmanning starts from (somewhat) fixed assumptions and looks for more robust arguments following from them that would address a given question (careful hypothetical reasoning), while devil’s advocacy starts from a fixed conclusion (not just a fixed question that the conclusion would judge) and looks for convincing arguments leading to it (rationalization with allowed use of dark arts).
A bad aspect of a steelmanned argument is that it can be useless: if you don’t accept the assumptions, there is often little point in investigating their implications. A bad aspect of a devil’s advocate’s argument is that it may be misleading, acting as filtered evidence for the chosen conclusion. In this sense, devil’s advocates exercise the skill of coming up with misleading arguments, which might be bad for their ability to reason carefully in other situations.
Then the main problem is that it produces (and exercises the skill of producing) arguments that are filtered evidence in the direction of the predefined conclusion, instead of well-calibrated consideration of the question on which the conclusion is one position.
So I’m still not sure what the difference with steelmanning is supposed to be, unless it’s that with steelmanning you limit yourself to fixing flaws in your opponents’ arguments that can be fixed without essentially changing their arguments, as opposed just trying to find the best arguments you can for their conclusion (the latter being a way of filtering evidence?)
That would seem to imply that steelmanning isn’t a universal duty. If you think an argument can’t be fixed without essentially steelmanning it, you’ll just be forced to say it can’t be steelmanned.
What leads you to believe that you disagree with Eliezer on this point? I suspect that you are just going by the title. I just read the essay and he endorses lots of practices that others call Devil’s Advocacy. I’m really not sure what practice he is condemning. If you can identify a specific practice that you disagree with him about, could you describe it in your own words?
Something I recently noticed: steelmanning is popular on LessWrong. But the sequences contain a post called Against Devil’s Advocacy, which argues strongly against devil’s advocacy, and steelmanning often looks a lot like devil’s advocacy. What, if anything is the difference between the two?
Steelmanning is about fixing errors in an argument (or otherwise improving it), while retaining (some of) the argument’s assumptions. As a result, the argument becomes better, even if you disagree with some of the assumptions. The conclusion of the argument may change as a result, what’s fixed about the conclusion is only the question that it needs to clarify. Devil’s advocacy is about finding arguments for a given conclusion, including fallacious but convincing ones.
So the difference is in the direction of reasoning and intent regarding epistemic hygiene. Steelmanning starts from (somewhat) fixed assumptions and looks for more robust arguments following from them that would address a given question (careful hypothetical reasoning), while devil’s advocacy starts from a fixed conclusion (not just a fixed question that the conclusion would judge) and looks for convincing arguments leading to it (rationalization with allowed use of dark arts).
A bad aspect of a steelmanned argument is that it can be useless: if you don’t accept the assumptions, there is often little point in investigating their implications. A bad aspect of a devil’s advocate’s argument is that it may be misleading, acting as filtered evidence for the chosen conclusion. In this sense, devil’s advocates exercise the skill of coming up with misleading arguments, which might be bad for their ability to reason carefully in other situations.
But what if you steelman devil’s advocacy to exclude fallacious but convincing arguments?
Then the main problem is that it produces (and exercises the skill of producing) arguments that are filtered evidence in the direction of the predefined conclusion, instead of well-calibrated consideration of the question on which the conclusion is one position.
So I’m still not sure what the difference with steelmanning is supposed to be, unless it’s that with steelmanning you limit yourself to fixing flaws in your opponents’ arguments that can be fixed without essentially changing their arguments, as opposed just trying to find the best arguments you can for their conclusion (the latter being a way of filtering evidence?)
That would seem to imply that steelmanning isn’t a universal duty. If you think an argument can’t be fixed without essentially steelmanning it, you’ll just be forced to say it can’t be steelmanned.
As far as I can tell...nothing. Most likely, there are simply many LessWrongers (like me) that disagree with E.Y. on this point.
What leads you to believe that you disagree with Eliezer on this point? I suspect that you are just going by the title. I just read the essay and he endorses lots of practices that others call Devil’s Advocacy. I’m really not sure what practice he is condemning. If you can identify a specific practice that you disagree with him about, could you describe it in your own words?