Steelmanning: Discover the truth by helping your opponent
themindcollection.com
themindcollection.com
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."
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.
[1] https://statmodeling.stat.columbia.edu/2022/04/28/the-challe...
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.
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.
Yeah, some of us are like that. I think I've finally matured at about 30-34, now I'm 38 but still learning.
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
1 : having or enjoying one or more privileges, privileged classes
2 : not subject to the usual rules or penalties because of some special circumstance
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.
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.
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.
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.
What's the difference? Unintentional pretence vs intentional pretence? Strong imagining vs weak imagining?
Both make an assertion. Both assign a degree of confidence.
Eg, “let’s imagine that elves are real” is distinct from “elves are real”.
There is no magic point at which hypothesis suddenly becomes fact. It's simply a matter of confidence.
“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.
Not confidence, acceptance.
To be in a position to admit that you’re wrong you’ve already got to be trying to argue that you’re right.
And HN is filled with engineers.
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/
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.
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.
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.
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.
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
...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:
If you're referring to other evidence, I'd love to know what it is.
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
> 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.
What are the secular arguments?
Pro abortion position is that unborn fetus is not a child. You are conceding pretty large point when you are framing it like this.
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.
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.
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.
I think a big exception is debate societies that meet regularly, where the folks doing the debating actually know each other and care about their long-term relationships with each other. But yeah, this isn't exactly a widespread hobby.
Also I'm not sure I agree that debates seeking truth are rare, they are just publicly broadcast less frequently so you come across more of them. Certainly I debate close friends and family matters on a fairly regular basis (for fun and learning).
The theory is essentially to refuse to participate in an echo chamber. If everyone in a group is feeding off each other in bashing an unpopular viewpoint, then you represent that viewpoint as strongly as possible rather than joining in.
I've used this technique with a lot of success, even in some pretty aggressive and opinionated groups. At a minimum it diffuses the echo chamber and the conversation turns to other topics, but I've also had good conversations come from it, where people leave understanding the other side better than they did before.
Because you're representing a party that is not present, you're less likely to end up making them feel patronized than with steelmanning, but it has a similar effect on your own understanding. It's even better if you're able to help someone else see the other perspective, too!
Once you both agree to what your opponent is claiming, you can identify and ask about all the assumptions that lead up to that claim, and from there, you can discuss and dismantle any false assumptions.
It's a way to "shortcut past the nonsense" and get straight to the root differences.
There's the occasional troll who is intentionally wasting your time, but I find that to be pretty rare, especially on HN. People are most often just bad at communication, not malicious.
Much of the time, the nonsense is there to keep things civil: you rehash tired tropes to talk around the issue for a while, get tired, and then somebody buys the next round of beers.
One time I was in an argument that was looking to get pretty heated, but with steelmanning I was able to find the underlying value difference, point out that we disagreed on it, and then we just kind of sat there in awkward silence for a second and moved on to play Starcraft.
It was a good day.
It's true that people can be too quick to dismiss, but that's for good reason. Bad faith arguments are catastrophically common.
The fact is most of us are eager to feel heard; and when we don't feel heard, we have an instinctive desire to shout our own viewpoint louder, making it more difficult to hear other people's viewpoints. The result is that situations where people feel strongly you get everyone shouting and nobody listening.
Stating the other person's point as clearly and strongly as you can not only helps you potentially be less wrong; it helps them feel heard, which makes them in turn more open to hearing other points of view.
There are a couple reactions you can have to that.
But I've found it a very effective approach when having a conversation about the best solution to a problem, for instance.
I also like the double crux technique: collaborate to find the crux for each party, the most important point upon which the argument rests, such that if it turned out to not hold you would change your mind. Turns an argument into a collaborative exploration.
I used to argue religion on Facebook a lot (I know, I know). But I stopped accepting invitations to debate when I started asked my opponents to agree before hand that they would agree to change their minds if I was right (as I agreed to change if I was wrong). When I realized they weren’t interested in learning anything, the point of debate went away.
In theory it's nice. In practice this doesn't get you anywhere nearly as often as you'd hope. Because IRL disagreements are almost often over the magnitude of a problem, not over their existence.
Trivial example for the sake of illustration:
Teammate 1: {writes some code}
Teammate 2: "That's hard to read, can you do it {this} way? It's much easier to understand the code that way if we find a bug."
Teammate 1: "That's too brittle though, can we do it {that} way? That way it's much less likely for the code to break in the first place."
Both sides can understand each others' positions perfectly fine. They just think their own concerns/priorities are a bigger deal than those of the other person.
Same goes with many political and other issues; this was just a silly example to illustrate.
And of course ultimately, if something is a subjective statement it's easy to reach agreements on the facts - Ie. "I feel like X", "well I feel like Y", where both are true statements - at which point you can stop pointlessly arguing and fall back on negotiation. "Well, I'll accept your PR if you at least Z."
Again, nice in theory, and sometimes that might work, but it has lots of failure modes. For example, you can end up with nonsensical Z's that neither of you would otherwise find particularly worthwhile or compelling, even though you both agree that, yes, Z would improve things slightly.
Exaggerated example: imagine the original debate had been over "we should leave the bike by the road so it's convenient to grab it and go" vs. "we should lock the bike by the house so it's harder to steal it", which is a debate over a 2-minute task. Your compromise Z might end up being an additional 1-week task like "well I'll let you leave it near the road if you/we first build a brown bikeshed for it so it at least looks nicer". You might both agree that it would indeed looks nicer that way, so there needn't be any disagreement on Z itself, but neither of you would've felt it appropriate to spend any time or resources on Z in the absence of the negotiation.
Usually it's not that stark of a contrast, but I've seen milder versions of this play out frequently in practice. Even when both people play along a few times (which often doesn't happen), they end up getting tired of having to go through it every time there's a disagreement.
I think that's where negotiation is a stronger frame though, because that sort of debate generally comes down to "how do I get through this faster", and the negotiative framing is "I disagree, but it doesn't matter, because the thing I actually want is for us to stop having this conversation, how can I achieve this with the greatest efficiency", or phrased otherwise, "what do I need to give you to make you go away?" Which there is much less to talk about.
HN was particularly bad about this with post flagging a few months ago. It was weaponized to hide posts which _someone_ disagreed with, even if the post was later restored the hour to twelve it was offline was enough to astroturf debates.
I'm seeing less of this since GP started getting his own posts flagged. Shame Dang didn't do his job till his boss told him to.
That being said, what steelmanning can offer, is to change the nature of the argument. If done correctly, you can sometimes de-escalate your opponent enough in order to tap into their humanity, and make the argument not about winning. Steelmanning is just a logical way of empathizing with your opponent.
In the presence of a motive for underhanded argumentation, strengthening your opponent's argument while allowing intentional weakening of yours is not a good recipe for useful outcomes.
How can we approach this problem?
To take your example of the abortion issue in America, your steel man argument has to explain why an acorn is the same as an oak tree, or perhaps the difference between an acorn and a fertilized human egg. They have to account for the mass murder of fertility clinics and researchers throwing out thousands of fertilized embryos per day.
Or you could choose to accept their philosophical argument about the start of life as a given, and discuss the consequences of it. E.g. you have a right to kill another human in self defence, unless the person threatening your life is inside a womb. Or the conditions under which you have a right to another person's life and body. Is there an innocence test we could do for organ transplant recipients, where the top end of the scale gives you a right?
Lots of room for good faith discussion there, even on immaterial subjects.
This an absolutely terrible option. It's rife with abuse, and even when fair, is only right slightly more than it's wrong. It doesn't resolve any of the underlying issues, which continue to exist. It seems to be better than any of the other last-ditch options, but it's incredibly bad.
Even among your allies, you're still stuck with the same problem. Being allies you're hoping that their core agreements with you will enable tactics like steelmanning to work, though in a lot of cases it still doesn't.
On the flip side, steelmanning is a really good "high ground maneuver". A good faith effort at trying to understand and strengthen the argument of someone who is being argumentative is a way to take the emotion out of the argument. You might find you have endeared yourself to them, and doing so have created a chance to shift their thinking a degree.
It's amazing what can happen when you don't assume someone who believes different things to you is an evil piece of shit.
Maybe there's hope for the world after all!
Just imagine the list of things you've seen flame wars about recently. It's one of those.
A Class Two disagreement is when each can explain to the other’s satisfaction the other’s point of view.
I've actually tried this, and it was pure torture for the subjects. They would start out, reluctantly, "He says X, Y, and Z ... [long pause] but he's wrong! "
You could see them try to state the other's POV, but they just couldn't do it.
I think reading a more in-depth explanation like these might help a lot
[1] https://www.latimes.com/business/hiltzik/la-fi-hiltzik-taylo...
I mean, a lot of people have trouble clearly articulating an argument for their own position, much less somebody else's. It's a skill like riding a bike or throwing a ball -- something that seems natural once you've had some practice, but not something we're born with. (NB I'm not talking about being correct; you can be right and know you're right without being able to articulate it well.)
If anyone wants some practice, I found the book "The Logic of Real Arguments" by Alec Fisher [1] really fun and useful.
[1] https://www.google.co.uk/books/edition/The_Logic_of_Real_Arg...
I prefer charitable interpretation (https://effectiviology.com/principle-of-charity/) which is a good platform for learning about the other's position, and potentially helping them strengthen their position for themselves.
This can happen because you don't understand it deeply enough, but it can also happen because of a difference in values. This is why I think steelmanning is difficult, and frankly a little dangerous. To truly do it, you have to embed yourself not just in their rational thought but also in their worldview.
And I really mean the part about it being dangerous. In all honesty, we aren't the purely rational creatures we think we are. Honestly steelmanning an argument changes you. And maybe that's for the better in some situations, but I certainly wouldn't take that for granted.
You're right that steelmanning is a bit different in that you're trying to formulate the argument in a way that is most convincing to you. That may not necessarily be the argument that is most convincing to your opponent.
That's why normal practice is to check: "So I hear you saying something like this: <steelman argument>. Is that right?"
> To truly do it, you have to embed yourself not just in their rational thought but also in their worldview.
Indeed; the best word I have for being truly persuasive is "incarnational": You have to enter into the situation of the person you want to convince, and see the way out of the cave they're in from where they are now.
> And I really mean the part about it being dangerous. In all honesty, we aren't the purely rational creatures we think we are. Honestly steelmanning an argument changes you. And maybe that's for the better in some situations, but I certainly wouldn't take that for granted.
I dunno; I have more faith in the power of truth. You certainly need to be careful that you're not taking information only from one source; if you only listened to Goebel's propaganda for 2 years you'd certainly start to have Nazi patterns of thought infect your thinking. But if you want to rescue someone who has only heard Goebel's propaganda for the last several years, you have to learn how to recognize the flaws in thinking from within it.
For me, most of the value of steelmanning is internal: questioning beliefs by dropping them and making the case for both sides. And when you find someone else doing this in their writing (for example Lyn Alden), you know you have found a useful source of information.
come on...
But it’s still a heady, rationalist approach to communication and will often stir up feelings of distrust, frustration, or resignation in the normal, everyday people that are pretty sure they’re right and see you doing intellectual performance art with their words.
Of course, that’s also what makes it a very effective way to get under the skin of online trolls.
Maybe I'm in a weird corner of the world, but I find use for this technique constantly in my professional life. You're frequently in situations where as a group you need to make decisions, and sometimes emotions run high; trying to articulate everyone's position as clearly as possible helps cool people down and opens them up to listening to other ideas, as well as making sure that all the relevant aspects have been covered.
Isn't steelmanning the practice of actively making your opponent's argument for them? This exercise goes beyond arguing in good faith, where you are arguing your own side, but in a way that doesn't strawman the other side.
You can absolutely do it for a position you don’t hold, and it has nothing particular to do with steelmanning/strawmanning (and indeed, you can use both techniques bad faith).
Nowadays, I try to read as much as I can and come to my own conclusions in quiet contemplation. The debate was merely a distraction from my real goal of improved understanding. In that context, I don't have any expectations on debating at all, let alone using a rhetorical technique like steelmanning.
[1] https://www.lesswrong.com/posts/HLqWn5LASfhhArZ7w/expecting-...
Often we know it won't work because of intuition/experience but providing a best effort steelman option when presenting design options is very helpful in speeding through stakeholder reviews. Stakeholders will eat you alive in a review if you present a strawman of what they asked for.
1. What is your central point?
2. What are your arguments? Rank them from best to worst.
3. What are your main claims? What explanations, supporting evidence and examples do you cite?
4. What are your emotional drivers, attitudes and beliefs, and motivations behind your viewpoint/worldview?
5. What are the vulnerabilities and weaknesses in any of the above? How can you fix them?
I really like this technique because it forces you to think about what assumptions the other side had to have made in order to reach their conclusion. I've often found that the other side's argument _is_ the logical choice based on their assumptions and background knowledge. From there I can examine those assumptions, and my own assumptions, and see which parts are incorrect, or id additional information would cause the argument to be less conclusive. I'll then focus my argument on those incorrect assumptions, because attacking the argument head-on often has counterproductive results. And of course, it's not uncommon to find that my own assumptions were incorrect while researching the other side's arguments and data
Steelmanning can also be used as preemptive self-defense for your position, assuming that it is actually defensible (if it is not, the sooner you come to that conclusion, the better for all concerned, especially yourself.)
Before the debate I would judge the quality of the position by the willingness of its proponents to honestly steelman its opponents, tending to give them a better grasp of the big picture.
A full steelman debate format would be very interesting.
This isn't practical mid-argument though, is it?
Closely related, point two "Help to fix your opponents argument," apart from being inherently escalatory, there is a pretty good chance that your opponent just made the argument like it is supposed to be and you're just not understanding it.
I've seen this approach in some oxford style debates and thought it made the debate better, as a viewer.
I like to think I have a pretty good idea of why people - whose ideas I find repulsive - believe what they do.
In my mind that means I have the capability to change their mind. The risk I take is that they can change mine.
Afraid there might be some really good reasons they're not worth talking about?
Steelmanning in practice may just be indistinguishable from trolling. The lazy redditor wants to look smarter than other redditors, by making a convoluted argument that no one believes in. It's intellectual posturing dressed up in "good faith".
A point of steelmanning is that your assumption that this person thinks an entire group of people is inferior might be incorrect.
Your grand plan will fail the moment you finally get to the "incorrect" part. Cause there you dealing with additional feeling of loss you yourself just caused.
Demonstrating that you fully understand an argument better than your opponent, and being able to charitably represent them, just makes it all the more devastating when you knock them down.
If you really disagree with an argument vehemently, then you should steelman it. Strawmanning makes you look dishonest and fragile, only capable of attacking a weakened form of your opposition.
If someone who has really abhorrent opinions tries making their argument, then you are letting them win if you strawman them.