Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Why I’m skeptical of “steelmanning”

statmodeling.stat.columbia.edu

Post reply on HN
Print view View original

61–70 of 131 posts

Re: Why I’m skeptical of “steelmanning”

#61

Post by anjbe »

TulliusCicero wrote: post #56
anjbe wrote: post #54

Earlier quoted context omitted.

> It's usually described as, "construct the best possible form of your opponent's argument." If they made some weak points in their argument, steelmanning would involve either ignoring those or reconstructing them to be stronger. Those two sentences seem contradictory. What definition of steelmanning requires ever ignoring weak points that have been made by one’s opponent?

I think this may depend on what definition you're using, but generally people have multiple points while arguing their case, and it can make sense to drop weaker or more vulnerable points to strengthen the overall argument. Thus, the best possible form of an argument may involve ignoring or reconstructing certain points. If someone says something vaguely racist while arguing against, say, multi-family housing, the be…

> the best possible form of their argument would probably involve dropping that -- even if it's actually a significant motivation

If a point is so significant, then the result of dropping it would no longer be “their argument.” Doing so would not qualify as steelmanning by any definition I’m aware of—it would be a different argument altogether.

To steelman is effectively to practice the principle of charity, by not assuming irrationality when a plausible rational view exists that fits the same arguments. But if one’s opponent has made an irrational argument, no assumption is necessary!

> I think this may depend on what definition you're using

What definition do you use? It doesn’t match how I’ve ever seen the term used.

Re: Why I’m skeptical of “steelmanning”

#62

Post by mensetmanusman »

Traster wrote: post #59

Honestly I see "steelmanning" as an emergent property of a completely broken discourse in the first place. The whole concept derives from the idea that we need specific terminology for when we're going to actually discuss things in a reasonable manner. It's basically a response to a group of public intellectuals who have emerged on the internet that spend more time building their brand and dunking on their opponents…

The idea of understanding your opponent’s arguments/beliefs better than they do is a centuries old technique.

Re: Why I’m skeptical of “steelmanning”

#63

Post by roenxi »

jan_Inkepa wrote: post #47
faeriechangling wrote: post #28

Earlier quoted context omitted.

I agree the author is massively overthinking this. Steelmanning just means treating your opponent charitably in good faith. If you interpret your opponents beliefs as having absurd outcomes which are gravely insulting to other people, such as believing that all doctors are part of some global conspiracy, than you honestly probably are pretty bad at imagining a good counterargument to your own position. I can come up…

>I agree the author is massively overthinking this. Steelmanning just means treating your opponent charitably in good faith. I don't mean this to sound pernickety: steelmanning from what I've seen normally is a lot more active than just treating the opponent charitably. Often it involves a lot of work trying to rework the other people's ideas from the ground up. It goes beyond charity in a way that's probably too lab…

It is easier to treat steelmanning as a hazy middle ground and pick examples that clearly are or clearly aren't.

Eg, the strongest pro-ivermectin argument was "There is all this evidence that lots of people who take ivermectin have much better outcomes after COVID. We should let people take ivermectin". Scott Alexander was literally the first person I saw to acknowledge the evidence before arguing against taking ivermectin to treat COVID. Which is quite an impressive first since I'd seen a lot of weirdly authoritarian people trying and failing to argue effectively against the drug up until then. Those people had failed to steelman the pro-ivermectin position.

Re: Why I’m skeptical of “steelmanning”

#64

Post by TulliusCicero »

anjbe wrote: post #61
TulliusCicero wrote: post #56

Earlier quoted context omitted.

I think this may depend on what definition you're using, but generally people have multiple points while arguing their case, and it can make sense to drop weaker or more vulnerable points to strengthen the overall argument. Thus, the best possible form of an argument may involve ignoring or reconstructing certain points. If someone says something vaguely racist while arguing against, say, multi-family housing, the be…

> the best possible form of their argument would probably involve dropping that -- even if it's actually a significant motivation If a point is so significant, then the result of dropping it would no longer be “their argument.” Doing so would not qualify as steelmanning by any definition I’m aware of—it would be a different argument altogether. To steelman is effectively to practice the principle of charity, by not a…

> If a point is so significant, then the result of dropping it would no longer be “their argument.”

I think that's debatable. People do shift which points they bring up and how they frame things while still feeling that their essential argument is unchanged.

> What definition do you use? It doesn’t match how I’ve ever seen the term used.

That's probably the rub here: the idea of "best possible form of the argument" leaves a fair amount of room for interpretation. I don't have some single authoritative source for my conception of steelmanning, I've just seen people use the term and I've read it explained in various places a few times.

In my defense, here's a source that agrees with my conception: https://debate.fandom.com/wiki/Steelmanning

> Steelmanning involves arguing against the strongest possible version of the other side's argument, whether they made it or not.

And here's a definition that's rather ambiguous: https://www.lesswrong.com/tag/steelmanning

> Steelmanning is the act of taking a view, or opinion, or argument and constructing the strongest possible version of it. It is the opposite of strawmanning.

Re: Why I’m skeptical of “steelmanning”

#66

Post by jwarden »

I find this paragraph in the article to be, ironically, a straw man.

> But it’s not “extremely plausible” that there’s a vast conspiracy among medical researchers (many of whom stated open to the view that ivermectin has some benefit) and public health officials to perpetrate the most significant fraud in human history

Because Alexander’s actual argument is the following.

“Sometimes these people even have a specific theory for why elites are covering up ivermectin, like that pharma companies want you to use more expensive patented drugs instead. This theory is extremely plausible.”

The latter does seem plausible given the examples Alexander cites of Big Pharma doing unethical things to promote use of more expensive drugs. But I don’t know if these examples are actually legit: if the author was steelmanning Alexander, he might have explored those questions.

Re: Why I’m skeptical of “steelmanning”

#67

Post by anjbe »

TulliusCicero wrote: post #64
anjbe wrote: post #61

Earlier quoted context omitted.

> the best possible form of their argument would probably involve dropping that -- even if it's actually a significant motivation If a point is so significant, then the result of dropping it would no longer be “their argument.” Doing so would not qualify as steelmanning by any definition I’m aware of—it would be a different argument altogether. To steelman is effectively to practice the principle of charity, by not a…

> If a point is so significant, then the result of dropping it would no longer be “their argument.” I think that's debatable. People do shift which points they bring up and how they frame things while still feeling that their essential argument is unchanged. > What definition do you use? It doesn’t match how I’ve ever seen the term used. That's probably the rub here: the idea of "best possible form of the argument" l…

> In my defense, here's a source that agrees with my conception: https://debate.fandom.com/wiki/Steelmanning

> > Steelmanning involves arguing against the strongest possible version of the other side's argument, whether they made it or not.

How bizarre! I would agree that by that definition, “steelmanning” is much less attractive—not something I would never do, but certainly not a practice I would attempt to follow by default.

Re: Why I’m skeptical of “steelmanning”

#68

Post by sokoloff »

anjbe wrote: post #54
TulliusCicero wrote: post #51

Earlier quoted context omitted.

That's not how steelmanning is generally described. It's usually described as, "construct the best possible form of your opponent's argument." If they made some weak points in their argument, steelmanning would involve either ignoring those or reconstructing them to be stronger. "Avoid attacking points that your opponent never made" is just the absence of strawmanning, it's not steelmanning.

> It's usually described as, "construct the best possible form of your opponent's argument." If they made some weak points in their argument, steelmanning would involve either ignoring those or reconstructing them to be stronger. Those two sentences seem contradictory. What definition of steelmanning requires ever ignoring weak points that have been made by one’s opponent?

Suppose I argue: “When SAT-math style comparing (square root of 9) to (1 + 2), it cannot be determined which is larger, therefore the correct answer would be “D-cannot be determined”, in part because a salmon is a mammal.”

Steelmanning would be agreeing with my conclusion while disputing the incorrect salmon point via pointing out that it’s entirely irrelevant to the correctness of my overall argument, even though I directly said the incorrect thing.

Re: Why I’m skeptical of “steelmanning”

#69

Post by pdpi »

TulliusCicero wrote: post #50
pdpi wrote: post #48

Earlier quoted context omitted.

To put steelmanning in software terms: if we’re arguing about what algorithm to use to solve a certain problem, I can assume you’ll work out all the bugs in your implementation (in your example: e.g. getting capacity/speed numbers slightly wrong for some modes of transportation), and focus on the core algorithm instead (e.g. that transitional period is something that definitely needs to be addressed)

That's not actually analogous, though, because often the weak points aren't "bugs to be worked out". For example, take this Parks & Rec clip: https://www.youtube.com/watch?v=Psi5m4x5aT0 While this lady was courageous enough to be willing to put her bigotry out in the open, for every person like this, there's 10 more locals with similar sentiments who are self-aware enough to not say the quiet part out loud. This is a…

There's two things at play here. One is your political opinions vs theirs, the other is your arguments for how you should handle a specific concrete situation. Crucially, the former doesn't (edit: necessarily) change the merits of the latter.

Would you accept the thrust of their argument if it came from a different (non-racist/classist) speaker? Then their racism is just a bug you can dismiss. Focusing on it is actually a straw man, because that's not germane to the argument they're actually making!

Does the argument hinge, at its core, on unspoken racist assumptions? Then it's a fundamental part of the argument and you need to tackle it.

Post reply on HN