Reply to Duncan Sabien on Strawmanning
DAVID: [...] Because that’s all we can ever have of each other: an imitation, a Copy. All we can ever know about are the portraits of each other inside our own skulls.
LORAINE: Is that all you think I am? An idea in your head?
DAVID: No! But if it’s all I have, then it’s all I can honestly love. Don’t you see that?And, miraculously, she does. She finally understands.
—”A Kidnapping” by Greg Egan
In the comments on “Basics of Rationalist Discourse” (where I am prohibited from commenting), Duncan Sabien writes, concerning my post “Aiming for Convergence Is Like Discouraging Betting”, that it’s “important to [him] that [I] be understood to not actually be responding to [him], rather than to [my] own strawman”.
I don’t think I understand the implied semantics of “actually” responding to an author. I don’t consider myself any kind of authority on Sabien’s views, and don’t see myself as attempting to speak on his behalf.
Rather, I consider “Aiming for Convergence Is Like Discouraging Betting” to be responding to the published text of “Basics of Rationalist Discourse”—not Sabien’s full private belief-state, which I obviously don’t have access to. That is, I was attempting to use language to communicate some thoughts I had that were partially inspired by language that I read in “Basics of Rationalist Discourse”.
Where I wasn’t sure how to interpret the text because Sabien’s thoughts and worldview differ so much from my own, I attempted to explicitly indicate my uncertainty, especially around the concepts of defection/cooperation and good/bad faith, of which I wrote (in the third summary bullet point) “I don’t think I understand how these terms are being used in this context” and (in the body of the post) that the meanings “in my vocabulary” and “in my ontology” “don’t seem consistent with the way Sabien seems to be using the words.” I also included a paragraph “emphasiz[ing] [...] that this discussion is extrapolating a fair amount from the text that was written”, and noting that Sabien might be using language differently from me “such that [my] objections don’t apply”.
It’s quite possible that my reading comprehension of the post was poor, either “innocently” or “motivatedly”. That is, it could be that a reader might compare my quotation or paraphrasing of “Basics of Rationalist Discourse” to the text of that post, and conclude that my construal of the text was unreasonable, based on how the words would be typically understood by English speakers.
If so, I eagerly welcome corrections: if Sabien or anyone else thinks I’ve misinterpreted the published text of “Basics of Rationalist Discourse”, I encourage them to write a comment explaining the misinterpretation, either on this post, or on “Aiming for Convergence Is Like Discouraging Betting”. I furthermore encourage voters to upvote what they see as correct explanations of misinterpretations, and to downvote posts that they believe to contain misinterpretations. Furthermore, if anyone has any specific suggestions for how I might edit the text of the post to correct such misinterpretations, I will happily consider them.
As it happens, I did receive a comment from Sabien contesting a paragraph in which I “suppose[d]” that a speaker who said that he was “not treating [his interlocutor] as a cooperative partner” and was “motivated [...] to raise [his] status at the expense of [theirs]” “would not be engaging in the ‘collaborative truth-seeking’ that the ‘Basics of Rationalist Discourse’ guideline list keeps talking about.” Sabien said that such a speaker would not be breaking any of the listed guidelines, and described the paragraph as “an excellent demonstration that [I] cannot pass the [Ideological Turing Test] of the post [I thought I was] objecting to, and [was] in fact objecting to a strawman of [my] own construction.”
I thanked Sabien for the clarification, adding that I agreed that I wasn’t passing his Ideological Turing Test (which I don’t think should be required before replying to a post). A few hours later, I asked a followup question: it still wasn’t clear to me why my hypothetical speaker wouldn’t be violating Sabien’s Fifth Guideline. As of the time of the present post’s publication, my followup question has not received a reply.
At the time of the present post’s publication, I have not edited the text of “Aiming for Convergence Is Like Discouraging Betting”, because it’s not clear to me that doing so is necessary: in the one case above where I received an explicit objection to the text I wrote, it seems plausible to me that my use of the first person and the verb suppose in the sentence “I suppose such a person would not be engaging in the ‘collaborative truth-seeking’ that the ‘Basics of Rationalist Discourse’ guideline list keeps talking about”, is sufficient to communicate to a reasonable reader that the sentence refers to my interpretation of the text of “Basics of Rationalist Discourse”, and doesn’t necessarily represent Sabien’s personal judgement of that particular scenario.
After all, how could it? The scenario of “such a person” was something I made up while writing “Aiming for Convergence Is Like Discouraging Betting”. I could hardly be expected to have solicited Sabien’s feedback on the post in advance of publishing, given that he doesn’t seem to want to communicate with me (him having blocked me on Facebook in December 2021, and having blocked me from commenting on his Less Wrong posts).
In consideration of the above, I’m inclined to tentatively reject Sabien’s characterization of my post as “objecting to a strawman of [my] own construction”. I think if that were true, it would be easy to quote a passage from the text of “Aiming for Convergence Is Like Discouraging Betting” and convincingly explain to the median Less Wrong reader specifically how it mischaracterizes the text of “Basics of Rationalist Discourse”, without directly appealing to Sabien’s personal judgement, and I don’t think this has been done. Less Wrong user Said Achmiz seems to concur with this assessment (writing, “then I read Zack’s post, thought ‘yep, sounds about right’”), as does Less Wrong administrator Ray Arnold (writing, “I generated an understanding of 5 that was similar enough to Zack’s understanding to also nod along with his post and think ‘yup, sounds about right’(ish)”).
The reason I think an explanation is necessary to sustain a strawmanning allegation, and not just the say-so of an author that a critic is not correctly representing them, is because critics can’t be expected to be mind-readers. Sabien writes that’s it’s possible to “underestimate the corrosive power of a flood of bullshit, and how motivated some people are to produce that flood.”. But I suggest this goes both ways: it’s also possible to underestimate the corrosive power of people being motivated to categorize things as bullshit in order to escape the obligation to reply to intellectually substantive objections.
It’s no doubt true that I don’t have a good understanding of Sabien’s views, and I don’t claim to. Unfortunately, I am not a mind-reader. I don’t have direct access to his beliefs. All I have to go on is the text he’s published—and if that’s all I have, then it’s all I can honestly reply to. If Sabien or anyone else disagrees with anything I write, they’re welcome to leave a comment explaining why! But I don’t see how the strawmanning allegation presently under consideration is credible in the absence of a more specific explanation than has heretofore been provided.
I don’t in general expect this to be easy, even if the accusation is true. (This is rule 0, right?)
I do remember reading your post and thinking you were misunderstanding Duncan’s, though I’d already read his comment calling yours a strawman so this doesn’t say as much as if I hadn’t. I don’t remember details now.
I might at some point get to rereading and see if I can either argue for that, or convince myself I was wrong. But I think it’ll take significant time and effort if I do.
Is “strawmanning” the same thing as “misunderstanding”, though?
To me, in my vocabulary, “strawmanning” indicates not just that a critic didn’t understand what the author thought they were trying to say (which is very common), but furthermore that this misunderstanding is culpable, a norm violation; the critic should have had better reading comprehension.
Correspondingly, I don’t accuse people of strawmanning unless I expect to be able to convince onlookers with an argument of the form, “The critic claimed in this-and-such paragraph that I claimed such-and-this, but if you look at the text of my post, I clearly claimed no such thing; no reasonable reader could walk away with that impression.”
Very often, I don’t think critics of my writing understood what I thought I was trying to say. But in the vast majority of cases, I don’t think it’s productive to try to litigate it as some sort of norm violation, when (if I care) I can just reply, “Nah, that’s not what I meant; to rephrase or elaborate, I meant such-and-this; does that make more sense?”
I think it’s important that I do it that way, because I don’t expect my audience to be mind-readers. Sometimes, I think a critic is being motivatedly dumb (in a way that I don’t think I can convincingly intersubjectively prove as a violation of norms against strawmanning). That’s very frustrating. But oftentimes, I didn’t make myself clear; the actual text that I actually published wasn’t good enough to convey my ideas to all reasonable readers. Sometimes, I didn’t make myself clear because my thoughts weren’t clear; the actual text that I actually published was bad because my ideas were bad. I think it’s important for people to be able to tell me that. I think if I were to expect my readers to be pre-emptively charitable to me, accusing them of “strawmanning” when they didn’t understand the text I wrote the way I hoped they would understand it, that would interfere with them being able to tell me when my ideas are bad. I think that would be bad for my intellectual development, and bad for the quality of this website.
Perhaps (I’m speculating; I don’t know) Duncan is using the word “strawmanning” in a weaker sense than the sense in which I understand the word, such that readers do have a duty to be pre-emptively charitable in order to not be guilty of strawmanning? This hypothesis seems (I’m inferring; I’m not sure) to be supported by his claim that I “cannot pass the ITT of the post [I thought I was] objecting to, and [was] in fact objecting to a strawman of [my] own construction”?
If I’m understanding correctly (and I may very well not be), the idea seems to be that a critic first needs to be able to rephrase an author’s idea in terms the author will agree with (“pass the author’s ITT”), before their criticism can even be considered to be a reply to the author at all (rather than to a strawman of the critic’s own construction)? If I’ve understood correctly (and—once again—I might not have), this seems like a hugely important crux: I just don’t think that’s a reasonable standard for intellectual discourse.
I think can see why someone might think it was a reasonable standard? The reasoning might go, “In order to debate a proposition, we first have to agree on what the proposition is exactly, and the ITT is the test to establish that we’re both talking about the same thing.”
The reason I disagree with this is because I think it’s surprisingly often possible to find flaws in ideas without too fine-grained of a shared understanding; I think expecting critics to pre-emptively pass my ITT would interfere with people telling me when my ideas are bad, and I need people to tell me when my ideas are bad.
(I’ve been limiting my internet access to focus on another project and likely won’t see replies for several days.)
I think I agree with some of what you say here and would prefer to give a more in-depth reply. I will briefly agree that “strawmanning” and “misunderstanding” have different valence in my head, though Duncan said “objecting to a strawman” rather than “strawmanning” and that has different valence from either of those two.
But also… I don’t really see how this relates to my comment? It still seems to me that
would not in general be easy. And if one were to do so with the aim of showing that you’re “strawmanning” Duncan, rather than simply “misunderstanding” him, where “strawmanning” is narrower than “misunderstanding”; then that surely makes the job harder.
These seem semantically equivalent to me? Duncan’s description of me as “enthusiastically leap[ing] to recalcitrant strawmanning” would seem to suggest they’re equivalent in his usage, too?
Hm, maybe I shouldn’t have said “easy.” I think I want to claim that “The critic is misunderstanding me” is something an author can credibly claim without an explanation (because the author is presumed to be a privileged authority on what their ideas are), but “The critic is strawmanning me” is an accusation that needs an explanation in order to be credible?
(But this position is somewhat awkward insofar as it implies that strawmanning is a crime that can’t be discussed literally; I never expect to say “I think he’s strawmanning me, but I can’t prove it” the way I would sometimes say “I think he’s lying, but I can’t prove it.” Interestingly, I don’t have this intuition about the word “mischaracterize”!)
Maybe some concrete examples will help? (Sorry that these are “political”, specifically gender-political; it’s what I’ve been working on for the past few years.)
Example 1. In “Challenges to Yudkowsky’s Pronoun Reform Proposal”, I accused Yudkowsky of “haughtily implying that people like [Barra] Kerr [the author of “Pronouns Are Rohypnol”] are making an elementary philosophy mistake that they are clearly not making”—namely, thinking that use of gender-identity-based pronouns constitutes “lying”—”if you actually read what they write.” (I didn’t use the specific word “strawman”, but it’s essentially the same complaint.)
I think I adequately support that accusation to my readers: I cited Kerr as a representative example of opponents of gender-identity-based pronouns, quoted text from Kerr’s article to demonstrate that she wasn’t just simplistically calling pronouns “lies”, and pointed out that the article was frequently referenced as evidence that Kerr really was a representative example.
Example 2. In Ozy’s criticism of my criticism of “The Categories Were Made for Man, Not Man for the Categories”, Ozy claimed that I “argue[d] for the existence of a third definition [of gender], based on psychology”, and claimed that my position implied that “any group of women whose interests and personality traits, on average, observably differ from that of women as a whole ought to be classified as not actually women at all” which would imply that lesbians aren’t women, reductio ad absurdum.
I thought Ozy’s post was quite bad, and very much failed to understand my position. I did not accuse them of stawmanning or similar, but if I were to make such an accusation, I would support that to my readers by pointing out that nowhere in the text of my post did I say that I wanted to define gender based on psychology, and that the text of my post explicitly anticipated and replied to the subgroup-different-from-women-as-a-whole objection in the paragraph starting with “To this it might be objected that [...]”.
Contrast of Examples 1 and 2 to the present case. I think it’s important that in both these cases where I either accused a critic of strawmanning or have explained how I would have done so, I provided an explanation of how the actual text the critic actually published substantively mischaracterizes the actual text the criticism-target actually published. I would not expect readers to just take my word for it; I tried to make it easy for a reader who doubted me to read the quotes for themselves and confirm for themselves whether my explanation made sense, and if someone did question my explanation, I would expect to be able to satisfactorily answer them.
I don’t think Duncan has met this standard, of providing a satisfactory explanation for how the text of “Aiming for Convergence Is Like Discouraging Betting” substantively mischaracterizes the text of “Basics of Rationalist Discourse.” As I explained in this post, in the one case where Duncan objected to the text of “Aiming for Convergence”, he said that the scenario I posed “doesn’t break any of the guidelines listed”. I find this reply very puzzling; it seems to me that the blunt person in my scenario is very much not “behav[ing] as if [his] interlocutor[ ] [is] also aiming for convergence on truth” (from the summary statement of the Fifth Guideline) nor “keep[ing] in touch with the possibility that you might be misunderstanding each other, or that the problem might be in your models, [...] etc.” (from the detailed expansion of the Fifth Guideline). It does not seem to me that the claim that the scenario I posed “doesn’t break any of the guidelines listed” is something a reasonable reader of the text of “Basics of Rationalist Discourse” should have been able to infer.
Perhaps Duncan’s claim that I’m strawmanning him isn’t about that particular paragraph. But if so … what is it, specifically? In what way, specifically does the text of “Aiming for Convergence Is Like Discouraging Betting” mischaracterize the text of “Basics of Rationalist Discourse”? If I were a lawyer prosecuting this on Duncan’s behalf, I would try to argue that “Aiming for Convergence”’s arguments against the Fifth Guideline are undermined by the fact that the expansion of the Fifth Guideline includes this paragraph:
That is, if “Aiming for Convergence” claimed that the Fifth Guideline claimed that people should attempt to converge with each other—effectively ignoring the “on truth” part—then it would be mischaracterizing the Fifth Guideline. But I don’t think a reasonable reader of “Aiming for Convergence” would characterize the text as ignoring the “on truth” part. Rather, I said:
That is, I’m not claiming that the Fifth Guideline says people should just try to converge with each other. That would be stupid, and I know Duncan is not claiming the stupid thing. Rather, the post is trying to illustrate why even though I know Duncan is not saying the stupid thing, it still seems super-weird to me to summarize “both of them to try to see more of what’s true” as “[a]im[ing] for convergence on truth, and behav[ing] as if your interlocutors are also aiming for convergence on truth”, because I just don’t think that’s what approximations of Bayesian updating look like, even though we have theorems about how perfect Bayesian reasoners with common priors would in fact converge on truth.
Did I do a perfect job of explaining this? No, I don’t think I did; thanks to critical commenters including Duncan, I have some vague ideas on how I might revise the post to be clearer if I get around to it. Should I, perhaps, have made it clearer that I know Duncan doesn’t believe the stupid thing, perhaps by quoting and acknowledging the paragraph starting with “If two people disagree”? Maybe! I will definitely consider that if I get around to revising the post! Is such a revision necessary in order for my post to not be guilty of strawmanning? Again, I argue No.
(Again, replying relatively-quickly, not in depth.)
Not to me. I would say someone is “objecting to a strawman” if someone else had constructed the strawman, but I wouldn’t say they’re “strawmanning” in that situation.
(E.g. Alice: “X”. Bob: “Alice says Y.” Carol: “According to Bob, Alice says Y. She’s wrong because...”.)
(Even if they were semantically equivalent, that doesn’t mean they have the same valence. But I think in this case the valency difference is mostly downstream of the semantic difference.)
...But in this case: there’s no third party, the quote continued ”...of his own creation”, and Duncan explicitly says “strawmanning” elsewhere, so sure. This point doesn’t seem relevant.
“Easy” is a large part of what I was replying to.
Paraphrasing, you said something like: “I’m inclined to tentatively reject (accusation). If it were true, it would be easy to prove. This hasn’t been done.” If it’s not easy to prove, even when true, then the justification to reject is weaker.
Like, from a Bayesian perspective? I don’t see why this would be the case—if we agree that these accusations aren’t easy to prove even when true, then lack of proof isn’t strong Bayesian evidence of falsehood.
(Even when they are easy to prove, someone might have various reasons not to provide the proof, so I still don’t think it’s particularly strong evidence.)
(Replies to possible objections. 1. You said “explanation” rather than “proof”, but I think explanations aren’t easy to give either. 2. We could say that if an accusation isn’t easy to explain/prove then the accuser also shouldn’t be confident in it—but I don’t think that would be true.)
Another thing I want to say is that with accusations of strawmanning, I don’t think the relevant context is limited to “here is what I said, here is what you said when replying”. It might be the case that there are patterns appearing over a long timeframe and many conversations. Such patterns, if they appear, are the sort of thing that an accuser can be very confident in while also being very hard to prove. Especially given that things on the internet sometimes disappear.
And from my understanding of the history between you and Duncan, I’d say that’s almost certainly part of the picture, in this instance.
So to summarize:
If you are mischaracterizing Duncan in “Aiming for convergence...”, I think this is something Duncan could demonstrate “locally”, i.e. by quoting your text and quoting his text and comparing. I don’t think this would be easy.
If you are strawmanning Duncan in that piece, I think this is not necessarily—in fact, probably not—something Duncan could demonstrate “locally”. And any demonstration, if possible at all, would be harder than in the above case.
In both cases, I think failure on Duncan’s part to demonstrate the accusation, is not strong evidence that the accusation is false. (Which obviously doesn’t mean it’s true, either.)
Communication is only “productive” when you either ask for more work to be done in whichever areas you suggest, or you provide the results of the work you have done in those areas. Everything else is just “drama.” Drama only serve the people involved in them, not the people who would benefit from knowledge in general. That doesn’t mean the drama is beneficial to them either. People who are addicted to opiates aren’t really making the best of choices even though they may think opiates benefit them. Yes, they do benefit, but certainly there are better choices out there.
Brainstorming sessions, meetings, etc. these are scenarios where people ask work to be done and people receive those requests and do the work afterwards. Educating yourself is work you do for yourself that you don’t need to communicate to others. You may seek guidance, but that’s just asking others to put in the work to analyze your own knowledge base and provide feedback so you can go back and improve on your learning for your own benefit. You don’t have to learn either. You don’t need to go to school, but it will make it difficult for you to live in the modern society. You owe it to yourself when the environment is changing. Or you knew nothing before but have to educate yourself on things that you didn’t know. Growing knowledge base means there are more things to learn. Need to learn? That depends on what you ask of yourself. You don’t have to ask yourself much either, and with that you don’t have to learn anything for anyone else.
I think I’m pretty good at reading minds, actually. You can probably do it too. I just attend to the personality of the person I’m talking to, ask myself what people who are like that usually think, and if I turn out to be wrong then my thought-prediction model gets an update.
I used to agree with this, and now strongly disagree; reading minds refers specifically to inferring structural, coherent, complex information—of the type typically encoded into language—directly from contents of someone’s brain, with low bias and low error rate. This is not possible without brain scanning technology or implants. I don’t think you’re claiming otherwise, of course—you’re using “mind reading” to refer to “inferring what someone is thinking by correlation from sidechannels in their behavior, as interpreted by models that picked up those sidechannel correlations from a variety of people, potentially including previous history with the person you’re observing”. While that’s a reasonable thing to claim to be able to do sometimes, it is much higher bias and error rate than real reading of private neural representation codes, and will often make catastrophically false inferences, especially about people who have unusual structural patterns in their concept-relational cognition, which does not get encoded into sidechannels significantly under most circumstances. Eg, even if you were in the room with me, I would be able to easily surprise you about what I choose to do next.
Once you know that I know I should make my beliefs pay rent in anticipated experiences, what more do you want me to know?
hmm, I’m not actually disagreeing with your empirical predictions about whether one can read sidechannels; I’m disagreeing that it’s relevant—Zack’s claim that sidechannel reading is insufficient to mindread Duncan does seem reasonable. And, in fact, duncan made a post expressing frustration about what appears to me issues with people failing to correctly infer his thinking, so I think Zack’s point is in fact well warranted—Duncan expressed he’s typically out-of-model for people who are surprised by this, so it makes sense that Zack would have issues interpreting his meaning from text. Duncan’s post expressing being out of model did request more assumption of uncertainty, and perhaps there’s a real disagreement in ideal approach here, but I don’t think failure to make mindreading guesses are why.
I think Duncan would be fine with people modelling him if those models were exceptionally discriminatory.
“It especially annoys me when racists are accused of ‘discrimination.’ The ability to discriminate is a precious faculty; by judging all members on one ‘race’ to be the same, the racist precisely shows himself incapable of discrimination.”
— Christopher Hitchens
So the advice should be to model more discriminately rather than to not model at all. More mindreading, not less.