Hacker News Reader

Comment by verdagon | Hacker News Reader

This technique really changed the way I engage with others' ideas [0], the way I approach language design [1], and even my relationships. It's so easy for us humans to get caught in a "me vs. you" mentality, but steel-manning really helps break out of that pattern and meaningfully connect with what others are saying.

It also unlocks the next level of rationality: the ability to admit you're wrong. After that one just needs to learn to actively test their own beliefs, and they can harness the true powers of the human mind.

My favorite snippet from Harry Potter and the Methods of Rationality: [2]

"But make no mistake, Draco, true science really isn't like magic, you can't just do it and walk away unchanged like learning how to say the words of a new spell. The power comes with a cost, a cost so high that most people refuse to pay it."

Draco nodded at this as though, finally, he'd heard something he could understand. "And that cost?"

"Learning to admit you're wrong."

[0] https://news.ycombinator.com/item?id=31937109

[1] https://vale.dev/

[2] https://www.hpmor.com/chapter/7

View on HN

Andrew Gelman offered a brief, mild critique of steelmanning [1] which you may find worth reading.

[1] https://statmodeling.stat.columbia.edu/2022/04/28/the-challe...

A good read, thanks for the link.

I think I'd disagree with his take on steelmanning though. He seems to think that in steelmanning, you need to assume the same assumptions as your opponent, thus precluding alternate possibilities.

But, in steelmanning, you need to ask the opposing side if your understanding of their point is correct, and also list out all the assumptions made along the way to get to that claim.

Once the assumptions are spoken out loud, it's easy to identify and focus on the first branch of disagreement. It saves a lot of time. That's the true power of steelmanning a hostile opponent.

I couldn’t agree more, well said

I agree with your assessment, but if your opponent is truly hostile it's not going to work. One person steelmanning and the other person strawmanning has a tendency to suffer from the same problem as always: to a superficially interested or poorly educated audience, defending against strawmen looks weak.

If you discover you operate from different axioms, or that even your ideas of how to evaluate the value of an axiom differ, what is the next step?

Draw parallels to other domains where those axioms come into conflict, to try and get some references on how to weigh them. Eventually you just live with the fact that some people have different moral axioms and weightings. It’s a big part of what it means to know someone well, and does not at all preclude having a strong relationship with someone.

As I commented when that article was posted to HN last time[1], it's important to recognize that there is an important distinction between accepting an argument and contemplating it as a hypothetical.

As long as one is able to maintain this distinction -- something most mature individuals should be perfectly capable of -- I think Andrew Gelman's criticism falls short.

[1] https://news.ycombinator.com/item?id=31244792

I think the "mature individuals" part precludes the possibility of even contemplating the position of others, for most people. I may be wrong though.

This; it’s been one of the biggest disappointments in my adult life to realize how many legal adults are so incredibly immature, with such inflated yet fragile egos, that merely apologizing to even someone in their own family is totally out of the question. Another is realizing how many people are so willfully ignorant yet dogmatic. I had some suspicions growing up, and even into my twenties I was still quite naive. Being surrounded by nerdy friends who were also generally professionals in a technical field definitely insulated me from the general public enough that it took another decade for it to really sink in. These days it just all feels a bit lonely (especially with COVID), but the spark of meeting kindred spirits still makes it worth it to attempt to have a social life. True friends are rare.

> how many legal adults are so incredibly immature

Yeah, some of us are like that. I think I've finally matured at about 30-34, now I'm 38 but still learning.

I think that you use "maturity" to mean "being perfect person" or term of approval rather then anything having to do with actual maturity.

It's a problematic word. Barring death or a rare physical condition, all human children biologically reach maturity. The colloquialism of 'mature' (or as the opposite of 'immature') varies notoriously person to person, to the point one could argue that it's immature to use it sincerely in debate.

I agree on it being a problematic word

As far as I am concerned, if one actually cares about the debate they should just replace "mature" for what they think it means or implies for the behaviour of the individual

Replacing the symbol for a definition is quite handy. People get all defensive over being called privileged, but change "privileged" for "having good circumstances" ( or similar ) and they don't feel called out for it despite having the same meaning

IMO it's something of a shame how much the human cognition is crap at that kind of stuff

From https://www.merriam-webster.com/dictionary/privileged :

    1 : having or enjoying one or more privileges, privileged classes
    2 : not subject to the usual rules or penalties because of some special circumstance

"having good circumstances" does not have the same meaning. For example you can have good circumstances by luck, or by good timing, without having any privileges.

I use the term "endorsement" to quotate when I believe a belief and whether I am merely acting as if I believe it when I discuss it.

There seems to be two subtly different definitions of steel manning.

The first one, and the one I think should always be done, is to engage with the best argument that has actually been presented by someone holding the opposing view.

The second version, and this one I don't fully understand, seems to involve defining your own strongest argument for the other view.

This version, in my experience, can go a few directions. Most often it turns into a more elaborate version of strawmanning ("and so we see, even with this strongest possible argument it still fails!"). It can also be a way toward compromise ("so if you're right, this much more limited form of my proposal still has value"). Or it can go to talking past each other (refuting vegetarianism without touching on the suffering of animals because you don't care about that, but many vegetarians do).

The first version is a requisite for truly debating rather than sloganeering. The second version is a mixed bag.

I think you've nailed why I've always felt slightly uncomfortable with "steelmanning" and I vastly prefer the alternative mentioned in the above rebuttal, "benefit of the doubt". So often "steelmanning" is indeed just another form of strawman, because at some level it's usually easy to misunderstand the perspective from which someone else is arguing (often the whole reason for getting into a heated debate in the first place). Giving someone the benefit of the doubt doesn't require any (possibly false) projection of having mutual understanding, but simply an attitude that "I may not understand it, but your argument is probably just as valid as mine". IMHO the real way to oppose the strawman fallacy or any other is simply to argue coherently in good faith, not to adopt a new potential fallacy on principle.

I think the second form is supposed to be a superset; you try to find all the best arguments you can for every position. That includes the ones other people are using, but if you can think of a better case to make for them then you should answer that as well. In no event should you ever say "you're arguing X but Y is a better argument and that's false so you're wrong" without also answering X, but "you're arguing X, which is wrong because X', but a better argument for your position is Y" is still legitimate.

he seems to present it as giving the "benefit of the doubt", but that's not correct. What you are doing is trying to make the best version of the argument before critiquing it. This avoids strawman type situations. It doesn't mean discarding weak arguments, but you may end up dropping weak parts of the argument that aren't necessary to the central argument. Basically it try's to avoid needless back and forth that doesn't really address actual argument at hand.

I think he's confusing accepting a conclusion with accepting a premise. Which is what steelmanning is supposed to be, granting your opposition a premise and then pointing out why even in that scenario, why their conclusion is still wrong. Or maybe it's not. In which case, you need to investigate/challenge the premise.

And then there's the issue that for some people, they just don't want to be reached. They want to believe a specific way on a specific issue for whatever personal reasons. You can't steelman crazy.

And let's not forget, steelmanning isn't about others, it's about yourself. It's about trying to find out if you are wrong.

> You can't steelman crazy.

THANK you.

I still can't rightly tell the difference between (the most generous definition of) steelmanning and just thinking things all the way through in a balanced way.

I am a bit puzzled that learning to admit that you are wrong is sometimes treated as attaining some kind of superpower. I would say it is just part of growing up. (Having said that, some people seem to never grow up.)

"This is true" and "Let us imagine that this is true" are radically, radically different positions. People who have a nuanced enough understanding to separate the two in an argument are rare.

Entertaining the even more nuanced idea that someone else has a strong argument that is wrong is, effectively, a superpower. Not many people who are willing to argue can do that. Especially if they are capable of doing that in a persuasive, public fashion.

>"This is true" and "Let us imagine that this is true" are radically, radically different positions.

What's the difference? Unintentional pretence vs intentional pretence? Strong imagining vs weak imagining?

Both make an assertion. Both assign a degree of confidence.

The second isn’t an assertion, it’s a hypothetical.

Eg, “let’s imagine that elves are real” is distinct from “elves are real”.

It's a confidence-modulated assertion. Hypothesis is low confidence. Fact is high confidence.

There is no magic point at which hypothesis suddenly becomes fact. It's simply a matter of confidence.

It’s confidence independent:

“Let X” is a context bound hypothetical — we’re assuming it is true within some argument; we have perfect confidence in the hypothetical, but only within that context.

That’s distinct from an assertion, which is a statement about how the world is with some associated confidence.

Have you never seen proofs by contradiction?

"It is the mark of an educated mind to be able to entertain a thought without accepting it." Aristotle

Not confidence, acceptance.

It’s not about admitting that you’re wrong. It’s about discovering what the right answer is without being swayed towards the thing you previously thought/said.

To be in a position to admit that you’re wrong you’ve already got to be trying to argue that you’re right.

What is growing up?

Realizing how pointless all of this is, for one.

Given that the method of engineering generally starts with something like, "These various assertions and principals may be taken as gospel". So admitting you might be wrong is like reversing gospel. And reversing gospel is quite a feat.

And HN is filled with engineers.

There's also the consideration that literal religious gospel is taken as fact by some people who use that assumption to push particular ideas in government and law

I couldn't invent a more suckass attitude for engineers to hold, but it is indeed common. I find it less common among really good engineers. YMMV.

HPMoR is a national treasure. If, as a Harry Potter and science fan, you ever find yourself with a few days free to read something ridiculous yet insanely entertaining, give it a go.

>"But steel-manning really helps break out of that pattern and meaningfully connect with what others are saying."

Steelmanning means that you ignore the argument your interlocuter makes, in favour of the argument you think they should have made. It runs diametrically counter to trying to 'meaningfully connect with what others are saying'.

If someone says 'X because Y', and you respond 'I think you mean X because Z but that's wrong', you have simply spoken past your interlocuter. Perhaps they actually meant 'X because Y2', but you didn't give them time to expand upon what '2' is. Or perhaps there's some other subtlety or depth that they have yet to expatiate.

There are lots of associated problems:

1. It is very difficult to say what a different version of the 'same' argument is. Perhaps you reason from different premises to the same conclusion. Perhaps you reason from the same premise to a better conclusion. Perhaps you use a similar style or tradition of argument. Really, only your interlocuter knows what 'their' argument is.

2. What you think is the strongest version of an argument is, may not be what a second or third person thinks the strongest argument is. In some realms - moral and political - there are no mind-independent facts in reference to which one argument can be objectively justified over another. 'Strongest' is therefore relative.

3. If you default to talking about an argument your opponent did not make, that is liable to create enormous confusion. If they did the same thing, you would end up never actually talking to one another. It would be like Chinese whispers, each rationally reconstructing the other's last statement.

What you really want to do is both 'interpret and understand' another's point of view, and then argue it through. The OP's link actually does a pretty good job of describing just that, but I think describing that as 'steelmannanning' is a misnomer. It's just philosophy as it has been understood for millennia. Steelmanning is supposed to be something more original.

Looking it up, this is claimed to be the original coinage of the word:

'Then we would be steelmanning, the art of addressing the best form of the other person’s argument, even if it’s not the one they presented.'[1]

[1] https://themerelyreal.wordpress.com/2012/12/07/steelmanning/

I strongly disagree with this interpretation. To the strength of your claims, the steelmanner is more engaging with themselves rather than the other person. Imagine a chess player playing both sides.

This of course isn't what steelmanning looks like. It's simply an extension of good faith. The other person needs to be there and needs to have their personal claims and beliefs represented in the steelman.

The most common steelman tactic is, after your opponent finishes a point, saying something to the effect of "Just so we're on the level, I'm going to restate what I believe your argument is and you can correct me if I've misinterpreted something".

Then you restate a summary of the argument in your own words and in good faith, bolstering it or lending structure where needed, or perhaps adding an obvious missing caveat. Then the person has a chance to correct you. They can kick out any supports they don't want you to add or recind any caveats.

From there, you can progress.

This is what steelmanning looks like. The form of an argument and the crux of it are not the same. It's good faith to interpret the form favourably.

Even in the most poorly stated arguments the core concept can be derived. You can give the person a better path to this central idea so as to ignore time waste arguments about the route, and instead attack that core idea.

You've ironically painted a hollow and one dimensional view of steelmanning.

> you restate a summary of the argument in your own words and in good faith, bolstering it or lending structure where needed [...] Then the person has a chance to correct you.

Where are y'all seeing these debates actually happen?

I always assumed steelmanning was a process you conduct entirely in your mind, while deciding your opinion or writing your editorial.

> Steelmanning means that you ignore the argument your interlocuter makes, in favour of the argument you think they should have made. It runs diametrically counter to trying to 'meaningfully connect with what others are saying'.

That's not quite steelmanning.

Steelmanning means that you rephrase the argument your interlocuter makes, with your own words, until they agree that it's the same or close enough. That's what creates the connection.

The agreement is the important distinction and what creates the connection. Even if you disagree, at least there is a level of respect that comes from hearing the person out and not misrepresenting their point.

Your definition is quite close but is skipping that last step that makes all the difference.

Thanks for the clarification. I recently read of steelmanning too by coincidence and have been fascinated with how it might work in practice.

I think it also means you should assume good faith too, right? So if someone says they are anti-abortion but pro-death penalty (for example) you don't jump on the hypocrisy as you see it but say "I see, so for you there is some material difference between the life of an unborn child and the life of a convict." When put this way it is easy to see where the other person's rationale is coming from.

I guess it like the Socractic method (except a little less confrontational) in which you restate the opponents statements in terms they can agree with, then follow up with questions exploring that until you trap them in a contradiction even they are forced to recognise.

Things in the realm of anti abortion and pro death penalty break down when you get down to the root of things, which is free will and innocence of the unborn soul

From my understanding that's what is behind both of these ideologies, and both of those things are 100% made up fiction

There is literally no evidence for an immaterial soul and there is actually evidence that free will is an illusion, but the people who believe things things have their beliefs rooted in religion

> Things in the realm of anti abortion and pro death penalty break down when you get down to the root of things, which is free will and innocence of the unborn soul

...no?

The essential point of steelmanning is understanding your opponent's perspective, perhaps even finding a better argument for it than they have, even if you disagree with them.

It's perfectly possible to do that even if you think their position is nonsense.

Sometimes you may even find that your perception of their arguments and ideas is incorrect.

> There is literally no evidence for an immaterial soul

There's no hard evidence, but there are things that are, let's say, weird. The hard problem of consciousness is one, and the steady trickle of reports of veridical observations from medical staff treating unconscious NDE patients is another. For that matter, NDEs themselves are odd in absentia the supernatural.

I don't expect people to grant souls or the supernatural based on those things, but they're certainly observed phenomena that some people interpret as evidence for those things.

> there is actually evidence that free will is an illusion

If you're referring to Libet's analysis of the Bereitschaftspotential experiments, there has been some acceptance in the past few years that this paper shows he misinterpreted the data:

https://www.pnas.org/doi/10.1073/pnas.1210467109

Popular summary of the debate in The Atlantic:

https://archive.ph/SHH0p

If you're referring to other evidence, I'd love to know what it is.

Consciousness is the experience if processing information, there is no reason to believe its anything more than that

There are also other studies besides the Libet ones, more recent ones using things like fMRI and in vivo electrophysiology in awake human patients.

If we think about how neurons and their networks function, there isn't a mechanism in the realm of biophysics that allows for nondeterminism. That alone precludes the possibility of what people classically think of as free will. I've had discussions with other neuroscientists about this and they came to the same conclusion.

Just because something seems "weird" doesn't actually mean anything and its likely the person's perception of the situation that instills that feeling.

Near death experiences are entirely subjective, again its perception and information processing. The brain is great at connecting concepts and generating ideas even while we're unconscious. Its not like we stop processing sensory information when we aren't active.

As someone who studies behavioral neuroscience, I would never use an NDE itself as evidence of anything besides the fact that it is a phenomena that happens sometimes

There is a decent overview on wikipedia, they do discuss criticisms as well:

https://en.m.wikipedia.org/wiki/Neuroscience_of_free_will

As a side note, I don't think that paper on pnas concludes what you think it does. They state that a stochastic model could account for the buildup, in reality the stochastic model is simulating the chaotic environment of biochemistry. That environment is still deterministic and we absolutely could predict the entire system if we are able to simulate all of the variables.

Neurons are complex, but no matter how you slice it they are still a deterministic system

thanks for the wiki link - interesting reading.

> Consciousness is the experience if processing information, there is no reason to believe its anything more than that

This reads to me as a baseless assertion. You could be right, but the body of your post hasn't done much to convince me.

Your understanding misses that there are secular and philosophical arguments to be made against both of those things, no need to invoke religion or go on an antitheist tirade.

Well philosophical arguments aren't based on evidence from what I've seen, so we can reject those right away

What are the secular arguments?

> I see, so for you there is some material difference between the life of an unborn child and the life of a convict.

Pro abortion position is that unborn fetus is not a child. You are conceding pretty large point when you are framing it like this.

Well, that's what they think isn't it? That's what I understand the steelmanning technique to be. Frame the argument as best you can in the way they see it, then you discover the assumptions that led them to their conclusion.

I know that assumption. What I object to is that OP uncriticslly accepts that assumption as true. His comment is literally making reader think equivalence between features and Baby is accepted fact.

That's certainly one pro-abortion position, but not the only one.

The more general position is that whatever interests a fetus may or may not have to continued life is outweighed by the bodily autonomy interests of the mother in a given scenario. The Violinist Argument is a classic thought experiment that argues even if the continued life in question belongs to fully-grown adult human who happens to be a world-class artist, bodily autonomy still wins.

Generally, pro life people arent going to find the violinists argument forceful: 1) the violinist is framed as being particularly valuable due to their talent, this is not a shared assumption between the two camps, and even if it was - every baby is a potential violinist.

2) the violinist is not offspring, and there is no parent-child relationship

3) extraordinary means were and are required to maintain the violinist, this is unlike a pregnancy - where extraordinary means are required to end it.

The Violinist Argument is not without its caveats and flaws. I'm just highlighting it as an example of a pro-abortion position that does not need to concede that the fetus is not a child.

Anti abortion people don't see mothers potential as something ever worth preserving - even when she is excessively young.

So no, this argument does not work, be abuse it is not about potential. Had it been, women who will loose potential or have higher risk of death in case of dangerous pregnancy would had some "potential" consideration too.

The argument is about who is entitled to potential or consideration and who does not deserve one.

Did you just… strawman the concept of steelmanning?

They did, and they did it by conflating steelmanning with strawmanning

And now I am going to have to re-read that masterpiece... le sigh

Definitely a masterpiece! I love this series. It's a wonderful blend of Ender's Game, cognitive science, and wizardly tomfoolery.

Readers loved it so much that an entire cast of volunteers banded together as voice actors to make an audiobook podcast out of it. [0]

Then, so many people listened to that and loved it, that folks even did an analysis podcast talking about the story chapter by chapter! [1]

And then, people did an analysis podcast of the analysis podcast! [2]

Anyone who hasn't read or listen to HPMoR should definitely give it a try.

[0] https://hpmorpodcast.com/

[1] https://hpmorpodcast.com/?p=2336

[2] https://www.patreon.com/posts/meta-mor-1-34064648

I love the original podcast and voices (and musical choices). There's another new audiobook version coming out chapter-by-chapter with a new narrator, which is also turning out quite well.

Hopefully some day we’ll have voice emulation so good that it can do convincing voice acting. I’d love these in the voice of Benedict Cumberbatch.

Oh my god, I'm so glad this is a thing. I stumbled on HPMoR last summer and marathoned the whole thing.

"You cannot always be right. You should try being wrong sometime, because you might learn something." -Johnny Depp

Reply on news.ycombinator.com