You don’t want your interlocutor to feel like you are either misrepresenting or humiliating him. Improving an argument is still desirable, but don’t sour the debate.
There are a couple different things I sometimes see conflated together under the label “steel man”.
As an example, imagine you’re talking to the mother of a young man who was killed by a drunk driver on the way to the corner store, and whose life could likely have been saved if he had been wearing a seat belt. This mom might be a bit emotional when she says “NEVER get in a car without your seat belt on! It’s NEVER safe!”, and interpreted completely literally it is clearly bad advice based on a false premise.
One way to respond would be to say “Well, that’s pretty clearly wrong, since sitting in a car in your garage isn’t dangerous without a seat belt on. If you were to make a non-terrible argument for wearing seat belts all the time, you might say that it’s good to get in the habit so that you’re more likely to do it when there is real danger”, and then respond to the new argument. The mother in this case is likely to feel both misrepresented and condescended to. I wouldn’t call this steel manning.
Another thing you could do is to say “Hm. Before I respond, let me make sure I’m understanding you right. You’re saying that driving with a seat belt is almost always dangerous (save for obvious cases like “moving the car from the driveway into the garage”) and that the temptation to say “Oh, that rarely happens!”/”it won’t happen to me!”/”it’s only a short trip!” is so dangerously dismissive of real risk that it’s almost never worth trusting that impulse when the cost of failure is death and the cost of putting a seat belt on is negligible. Is that right?”. In that case, you’re more likely to get a resounding “YES!” in response, even though that not only isn’t literally what she said, it also contradicts the “NEVER” in her statement. It’s not “trying to come up with a better argument, because yours is shit”, it’s “trying to understand the actual thing you’re trying to express, rather than getting hung up on irrelevant distractions when you don’t express it perfectly and/or literally”. Even if you interpret wrong, you’re not going to get bad reactions because you’re checking for understanding rather than putting words in their mouth, and you’re responding to the thing they are actually trying to communicate. This is the thing I think was being pointed at in the original description of “steel man”, and is something worth striving for.
In writing this article I did some digging into the history of steelmanning in order to find the original meaning and I got to say I’m not sure your description totally fits. The Black Belt Bayesian quote I mentioned wasn’t talking about that, but maybe that doesn’t count because he didn’t call it a steelman (I’m not sure the term was invented yet). The Lukeprog post I linked to did say
2 Sometimes the term “steel man” is used to refer to a position’s or argument’s improved form. A straw man is a misrepresentation of someone’s position or argument that is easy to defeat: a “steel man” is an improvement of someone’s position or argument that is harder to defeat than their originally stated position or argument.
but that might’ve been edited in later. I found this post by ‘the merely real’:
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.
This is a very early mention and it’s also the source wikipedia uses.
But I’ve also seen plenty of people use steelmanning the way you explained it. Your definition of steelmanning would be fine during the debate but the other definition wouldn’t be. I solved this dilemma by just not mentioning the term steelmanning in my post and instead using the original Black Belt Bayesian quote.
a “steel man” is an improvement of someone’s position or argument that is harder to defeat than their originally stated position or argument.
This seems compatible with both, to me. “You’re likely to underestimate the risks, and you can die even on a short trip” is a stronger argument than “You should always wear your seat belt because it is NEVER safe to be in a car without a seat belt”, and cannot be so easily defeated as saying “Parked in the garage. Checkmate”.
Reading through the hyperbole to the reasonable point underneath is still an example of addressing “the best form of the other person’s argument”, and it’s not the one they presented.
There are a couple different things I sometimes see conflated together under the label “steel man”.
As an example, imagine you’re talking to the mother of a young man who was killed by a drunk driver on the way to the corner store, and whose life could likely have been saved if he had been wearing a seat belt. This mom might be a bit emotional when she says “NEVER get in a car without your seat belt on! It’s NEVER safe!”, and interpreted completely literally it is clearly bad advice based on a false premise.
One way to respond would be to say “Well, that’s pretty clearly wrong, since sitting in a car in your garage isn’t dangerous without a seat belt on. If you were to make a non-terrible argument for wearing seat belts all the time, you might say that it’s good to get in the habit so that you’re more likely to do it when there is real danger”, and then respond to the new argument. The mother in this case is likely to feel both misrepresented and condescended to. I wouldn’t call this steel manning.
Another thing you could do is to say “Hm. Before I respond, let me make sure I’m understanding you right. You’re saying that driving with a seat belt is almost always dangerous (save for obvious cases like “moving the car from the driveway into the garage”) and that the temptation to say “Oh, that rarely happens!”/”it won’t happen to me!”/”it’s only a short trip!” is so dangerously dismissive of real risk that it’s almost never worth trusting that impulse when the cost of failure is death and the cost of putting a seat belt on is negligible. Is that right?”. In that case, you’re more likely to get a resounding “YES!” in response, even though that not only isn’t literally what she said, it also contradicts the “NEVER” in her statement. It’s not “trying to come up with a better argument, because yours is shit”, it’s “trying to understand the actual thing you’re trying to express, rather than getting hung up on irrelevant distractions when you don’t express it perfectly and/or literally”. Even if you interpret wrong, you’re not going to get bad reactions because you’re checking for understanding rather than putting words in their mouth, and you’re responding to the thing they are actually trying to communicate. This is the thing I think was being pointed at in the original description of “steel man”, and is something worth striving for.
In writing this article I did some digging into the history of steelmanning in order to find the original meaning and I got to say I’m not sure your description totally fits. The Black Belt Bayesian quote I mentioned wasn’t talking about that, but maybe that doesn’t count because he didn’t call it a steelman (I’m not sure the term was invented yet). The Lukeprog post I linked to did say
but that might’ve been edited in later. I found this post by ‘the merely real’:
This is a very early mention and it’s also the source wikipedia uses.
But I’ve also seen plenty of people use steelmanning the way you explained it. Your definition of steelmanning would be fine during the debate but the other definition wouldn’t be. I solved this dilemma by just not mentioning the term steelmanning in my post and instead using the original Black Belt Bayesian quote.
This seems compatible with both, to me. “You’re likely to underestimate the risks, and you can die even on a short trip” is a stronger argument than “You should always wear your seat belt because it is NEVER safe to be in a car without a seat belt”, and cannot be so easily defeated as saying “Parked in the garage. Checkmate”.
Reading through the hyperbole to the reasonable point underneath is still an example of addressing “the best form of the other person’s argument”, and it’s not the one they presented.