The worst argument in the world already has a different name. Philosophers call it the logical fallacy of Accident.
Calling out the worst argument in the world is not useful in practice. It is really hard to stop it from being a fully general counterargument against any high level abstract argument. The article seems to hold that for communication to work properly all statements must refer to “archetypes”, central members of a cluster in thing space. If so, this conflicts with the very idea of parsing reality into clusters-in-thingspace, which is inevitable. Every cluster, being a cluster and not a point, has more and less central members. If arbitrarily marginal members of clusters are invalid members, arbitrarily many things said by humans are The Worst Argument In The World. To banish statements that don’t locate one cluster-in-thingspace right into the centre of another cluster-in-thingspace is faulty, especially when the statements are slogans and the words highly abstract. To use it properly you have to come up with an argument that shows that either the rule or generalization you are attacking is wrong or the case considered is sufficiently exceptional that it no longer applies. I wouldn’t trust myself to use that line of reasoning against an argument I already dislike to discount it. And if this really is a way to defend oneself from the dark arts as it presents itself doing, it should be good for precisely that! The article seems much more well made as a weapon to add to that arsenal but then it should be marked as such.
I think the most critical response to the worst argument in the world is that so many people are misunderstanding it (it was better explained on Yvain’s blog where he didn’t speak in LessWrongese). However, you are right that it is the logical fallacy of accident (as it is probably a form or child or parent of various other types of fallacies), but it’s been put in LessWrong’s clothes like Yudkowsky has done with other existing biases and fallacies, as such it assumes the LWian worldview and thus imports some nuances which kilobug partly noted.
To your second point, no line is ever drawn on what thing inside cluster-space is outside of the cluster for a given argument. Instead, the entire cluster is banished. Instead, you must argue for the tautology of which the cluster represents (e.g., murder cluster = tautologically bad), and even that’s assuming the cluster should be noncontinuous tautologies (shouldn’t things farther away from the center of the murder cluster be less bad?). This is no different than the philosophical process of unpacking statements to avoid begging the question.
Inference and context are annoyingly important in communication; you start off on the blog by making your definition more personal while on LW it’s more abstract and thereby it doesn’t convey your intention as well (although, it should be inferred from the rest of the post). It’s kinda the same throughout the LessWrong post.
Blog: “If we can apply an emotionally charged word to something, we must judge it exactly the same as a typical instance of that emotionally charged word.”
LW: “X is in a category whose archetypal member has certain features. Therefore, we should judge X as if it also had those features, even though it doesn’t.”
Suggestion:
Yesterday, I had a conversation with a friend about slavery. My friend said, “you know, capitalism is evil.” I replied, “Why is that?” He said, “You see, the Merriam-Webster dictionary defines slavery as ‘submission to a dominating influence’ which clearly includes working for a wage, so therefore capitalism is slavery!” I said, “you mean like slavery-slavery? Whips and shackles?” He said, “sure, see working for a wage is clearly submitting to a dominating influence, so it’s slavery all the same. But let’s not get into semantics...”
If David Stove can unilaterally declare a Worst Argument, then so can I. I declare the Worst Argument In The World to be this: “X is in a category whose prototypical member has certain features. Therefore, let’s presuppose X has all of those same features.”
A pretty bad argument is this widespread idea that one should never “get into semantics”, even if that is what is causing problems. Many even use “semantics” to mean something like “pointless pedantry”. I can remember when semantics was a respectable academic discipline...
Amen to this. Indeed, I fear that an actual majority of “people out there” may have no idea that “semantics” means anything other than “pointless pedantry”.
Actually, though semantics is perhaps the hardest hit, this is a general phenomenon, afflicting many unfortunate disciplines. You might call it the Argument from Circumscription of Subject Matter, or the ”...But That Would Get Us Into X” Fallacy. Essentially, it goes like this: “that line of inquiry can’t possibly be relevant, because it comes under the heading of a different academic discipline from the one our discussion falls under”. It is particularly common (and insidious) when the “other” discipline has some kind of “bad” reputation for some reason (as in the case of semantics, which is evidently regarded as “pointless pedantry”).
As a fictional (yet particularly illustrative) example of this fallacy, one could imagine EY and his colleagues at SIAI a decade ago saying “Well, we could worry about making sure future AI is Friendly, but....that would get us into philosophy [which is notoriously difficult, and not techno-programmer-sounding, so we won’t].”
To which the response, of course, is: “So it would. What’s your point?”
Many even use “semantics” to mean something like “pointless pedantry”.
I have even been in a conversation (with some MENSAns) where the primary subject was actually the meaning of a particular word. One person tried to support his position by retorting that the other person’s argument was “just semantics”. Well, obviously, yes. But that’s a literal description of the subject matter, not an excuse to use “Hah! Semantics!” as a general counterargument!
(Not that I endorse the conversation itself as especially useful, just that “Semantics! My side wins!” is very different to “Semantics! Let’s not have this conversation”.)
Minor wording point: labelling point 2 as “The worst argument in the world is not a useful argument in practice” sounds like you’re about to attack the WAitW, when you’re actually warning against labelling things as the WAitW. It might be less ambiguous to relabel point 2 as “Calling out the worst argument in the world is not useful in practice” or something similar.
I see a (subtle but significant) difference between Aristotle version and Yvain version.
In Aristotle version, it goes like “doing A is X”, “B does A” so “B is X”. That’s wrong, because (but Aristotle didn’t know it) words are not precise definitions but fuzzy clusters. That’s the main for which the “fallacy of accident” is a fallacy. And surgeons are not criminals.
The Yvain version is much more subtle. It acknowledges that words are fuzzy clusters, not fixed definitions. And that you can, without it being a fallacy (unlike in the first case) make claim like “abortion is murder” or “death penalty is murder”. But that even if that claim can be make (even if we can consider them to be part of the fuzzy cluster) it’s still a fallacy to use it as an argument, because while they are part of the cluster, they only share some of the problems that a typical member of the cluster has.
Now, if you consider my own point of view on those issues (but it could symmetric) : I’m pro-choice and against the death penalty. The WAitW idea is that I shouldn’t argue for the right to abortion by trying to prove “abortion is not murder” and against the death penalty by trying to prove “death penalty is murder”, being stuck in a definition match which is pointless, but that I should look deeper, dissolve what “murder” is and what it’s assumed to be wrong, and show that most of what make us reject murder doesn’t apply to abortion, and most of what make us reject it applies to death penalty. Or even completely discard the “murder” concept, and just look from a consequentialist point of view about the good and bad consequences of both.
“doing A is X”, “B does A” so “B is X”. That’s wrong, because (but Aristotle didn’t know it) words are not precise definitions but fuzzy clusters.
Everyone who does A is X.
B does A.
B is X.
That sounds like a valid argument to me. As such, if the premises are true, no god could make the conclusion false. The problem isn’t with this mode of argumentation. It is literally the opposite of fallacious. If there’s a problem, it’s just the very mundane problem that one of the premises is false.
In “doing A is X” (which kilobug wrote) X is an attribute of an action. In “everyone who does A is X” (which you wrote, apparently intending to echo what kilobug wrote) X is an attribute of people.
These aren’t equivalent. I’m not sure how relevant that is to your point, but then I’m not sure why you swapped one for the other.
I hesitated about that, but If Kilo had intended to hang something on that difference, then his subsequent comment probably wouldn’t have been about clusters in thing-space. ‘Fundamental attribution error’ isn’t relevant to that issue. That’s why I felt comfortable swapping them. But I’m not super confident about that.
I assume kilobug didn’t intend to hang anything on the difference between what they wrote and what you later wrote.
I assume you considered the difference significant, since if it wasn’t significant you could have actually referenced what he said to make your point, rather than referencing some other statement that he didn’t actually make.
I don’t know if the difference is actually significant.
The use of the Worst Argument in the World in practice is as a heuristic for tabooing words that don’t fit very well (and hence leaking inapplicable/misleading connotations). You are not refuting arguments with it, you are drawing more attention to certain parts and calling for unpacking. A good argument should be unpackable to significant degree, but in practice it’s too much work to unpack everything, so it’s useful to have heuristics that would point where to start digging.
If the argument remains sensible after you unpack, then there is no problem. The bad thing is when an argument was relying on not being unpacked and crumbles once you look inside. So the refutation or the lack thereof depends on what happens after you unpack, the heuristic for deciding what to unpack doesn’t itself perform any refutation.
I find myself questioning how many readers will actually do the unpacking you describe rather than just use the Worst Argument in the World as a club to beat their opponents over the head. Especially since title is such that it will probably attract many readers off LessWrong.
“Taboo murder.” works better than “Calling X murder is the worst argument in the world!”
I’ve heard anecdotes of philosophy professors dreading the lesson on logical fallacies because the students use them as a weapon. But even so, logical fallacies are pedagogically useful like the worst argument in the world. To know that you should taboo murder rather than continue presupposing “all murder = bad” requires a degree of sophistication, and learning logic and logical fallacies is exactly how you learn to unpack those presuppositions and actually argue rather than score political points.
I think the best practice is to taboo saying “X is Y bias or X is Y logical fallacy”, and rather require people to explain or question the exact flaw in reasoning and possibly why it’s important enough to bring up.
So, for example, if someone says “that’s murder, so it’s evil,” you should then reply with something like “why does something being murder necessitate it being evil?” (all the while internally thinking, “ah ha! I think that was the worst argument in the world.”)
Related to: List of public drafts on LessWrong
Draft of a critical response to this article
The worst argument in the world already has a different name. Philosophers call it the logical fallacy of Accident.
Calling out the worst argument in the world is not useful in practice. It is really hard to stop it from being a fully general counterargument against any high level abstract argument. The article seems to hold that for communication to work properly all statements must refer to “archetypes”, central members of a cluster in thing space. If so, this conflicts with the very idea of parsing reality into clusters-in-thingspace, which is inevitable. Every cluster, being a cluster and not a point, has more and less central members. If arbitrarily marginal members of clusters are invalid members, arbitrarily many things said by humans are The Worst Argument In The World. To banish statements that don’t locate one cluster-in-thingspace right into the centre of another cluster-in-thingspace is faulty, especially when the statements are slogans and the words highly abstract. To use it properly you have to come up with an argument that shows that either the rule or generalization you are attacking is wrong or the case considered is sufficiently exceptional that it no longer applies. I wouldn’t trust myself to use that line of reasoning against an argument I already dislike to discount it. And if this really is a way to defend oneself from the dark arts as it presents itself doing, it should be good for precisely that! The article seems much more well made as a weapon to add to that arsenal but then it should be marked as such.
I think the most critical response to the worst argument in the world is that so many people are misunderstanding it (it was better explained on Yvain’s blog where he didn’t speak in LessWrongese). However, you are right that it is the logical fallacy of accident (as it is probably a form or child or parent of various other types of fallacies), but it’s been put in LessWrong’s clothes like Yudkowsky has done with other existing biases and fallacies, as such it assumes the LWian worldview and thus imports some nuances which kilobug partly noted.
To your second point, no line is ever drawn on what thing inside cluster-space is outside of the cluster for a given argument. Instead, the entire cluster is banished. Instead, you must argue for the tautology of which the cluster represents (e.g., murder cluster = tautologically bad), and even that’s assuming the cluster should be noncontinuous tautologies (shouldn’t things farther away from the center of the murder cluster be less bad?). This is no different than the philosophical process of unpacking statements to avoid begging the question.
Can you explain what part on my blog you thought was better, so I can maybe replace it here?
Inference and context are annoyingly important in communication; you start off on the blog by making your definition more personal while on LW it’s more abstract and thereby it doesn’t convey your intention as well (although, it should be inferred from the rest of the post). It’s kinda the same throughout the LessWrong post.
Blog: “If we can apply an emotionally charged word to something, we must judge it exactly the same as a typical instance of that emotionally charged word.”
LW: “X is in a category whose archetypal member has certain features. Therefore, we should judge X as if it also had those features, even though it doesn’t.”
Suggestion:
Yesterday, I had a conversation with a friend about slavery. My friend said, “you know, capitalism is evil.” I replied, “Why is that?” He said, “You see, the Merriam-Webster dictionary defines slavery as ‘submission to a dominating influence’ which clearly includes working for a wage, so therefore capitalism is slavery!” I said, “you mean like slavery-slavery? Whips and shackles?” He said, “sure, see working for a wage is clearly submitting to a dominating influence, so it’s slavery all the same. But let’s not get into semantics...”
If David Stove can unilaterally declare a Worst Argument, then so can I. I declare the Worst Argument In The World to be this: “X is in a category whose prototypical member has certain features. Therefore, let’s presuppose X has all of those same features.”
A pretty bad argument is this widespread idea that one should never “get into semantics”, even if that is what is causing problems. Many even use “semantics” to mean something like “pointless pedantry”. I can remember when semantics was a respectable academic discipline...
Amen to this. Indeed, I fear that an actual majority of “people out there” may have no idea that “semantics” means anything other than “pointless pedantry”.
Actually, though semantics is perhaps the hardest hit, this is a general phenomenon, afflicting many unfortunate disciplines. You might call it the Argument from Circumscription of Subject Matter, or the ”...But That Would Get Us Into X” Fallacy. Essentially, it goes like this: “that line of inquiry can’t possibly be relevant, because it comes under the heading of a different academic discipline from the one our discussion falls under”. It is particularly common (and insidious) when the “other” discipline has some kind of “bad” reputation for some reason (as in the case of semantics, which is evidently regarded as “pointless pedantry”).
As a fictional (yet particularly illustrative) example of this fallacy, one could imagine EY and his colleagues at SIAI a decade ago saying “Well, we could worry about making sure future AI is Friendly, but....that would get us into philosophy [which is notoriously difficult, and not techno-programmer-sounding, so we won’t].”
To which the response, of course, is: “So it would. What’s your point?”
I have even been in a conversation (with some MENSAns) where the primary subject was actually the meaning of a particular word. One person tried to support his position by retorting that the other person’s argument was “just semantics”. Well, obviously, yes. But that’s a literal description of the subject matter, not an excuse to use “Hah! Semantics!” as a general counterargument!
(Not that I endorse the conversation itself as especially useful, just that “Semantics! My side wins!” is very different to “Semantics! Let’s not have this conversation”.)
Yeah, that was the joke.
Minor wording point: labelling point 2 as “The worst argument in the world is not a useful argument in practice” sounds like you’re about to attack the WAitW, when you’re actually warning against labelling things as the WAitW. It might be less ambiguous to relabel point 2 as “Calling out the worst argument in the world is not useful in practice” or something similar.
Obvious fix. Thank you!
I see a (subtle but significant) difference between Aristotle version and Yvain version.
In Aristotle version, it goes like “doing A is X”, “B does A” so “B is X”. That’s wrong, because (but Aristotle didn’t know it) words are not precise definitions but fuzzy clusters. That’s the main for which the “fallacy of accident” is a fallacy. And surgeons are not criminals.
The Yvain version is much more subtle. It acknowledges that words are fuzzy clusters, not fixed definitions. And that you can, without it being a fallacy (unlike in the first case) make claim like “abortion is murder” or “death penalty is murder”. But that even if that claim can be make (even if we can consider them to be part of the fuzzy cluster) it’s still a fallacy to use it as an argument, because while they are part of the cluster, they only share some of the problems that a typical member of the cluster has.
Now, if you consider my own point of view on those issues (but it could symmetric) : I’m pro-choice and against the death penalty. The WAitW idea is that I shouldn’t argue for the right to abortion by trying to prove “abortion is not murder” and against the death penalty by trying to prove “death penalty is murder”, being stuck in a definition match which is pointless, but that I should look deeper, dissolve what “murder” is and what it’s assumed to be wrong, and show that most of what make us reject murder doesn’t apply to abortion, and most of what make us reject it applies to death penalty. Or even completely discard the “murder” concept, and just look from a consequentialist point of view about the good and bad consequences of both.
Everyone who does A is X. B does A. B is X.
That sounds like a valid argument to me. As such, if the premises are true, no god could make the conclusion false. The problem isn’t with this mode of argumentation. It is literally the opposite of fallacious. If there’s a problem, it’s just the very mundane problem that one of the premises is false.
In “doing A is X” (which kilobug wrote) X is an attribute of an action.
In “everyone who does A is X” (which you wrote, apparently intending to echo what kilobug wrote) X is an attribute of people.
These aren’t equivalent.
I’m not sure how relevant that is to your point, but then I’m not sure why you swapped one for the other.
I hesitated about that, but If Kilo had intended to hang something on that difference, then his subsequent comment probably wouldn’t have been about clusters in thing-space. ‘Fundamental attribution error’ isn’t relevant to that issue. That’s why I felt comfortable swapping them. But I’m not super confident about that.
I assume kilobug didn’t intend to hang anything on the difference between what they wrote and what you later wrote.
I assume you considered the difference significant, since if it wasn’t significant you could have actually referenced what he said to make your point, rather than referencing some other statement that he didn’t actually make.
I don’t know if the difference is actually significant.
The use of the Worst Argument in the World in practice is as a heuristic for tabooing words that don’t fit very well (and hence leaking inapplicable/misleading connotations). You are not refuting arguments with it, you are drawing more attention to certain parts and calling for unpacking. A good argument should be unpackable to significant degree, but in practice it’s too much work to unpack everything, so it’s useful to have heuristics that would point where to start digging.
If the argument remains sensible after you unpack, then there is no problem. The bad thing is when an argument was relying on not being unpacked and crumbles once you look inside. So the refutation or the lack thereof depends on what happens after you unpack, the heuristic for deciding what to unpack doesn’t itself perform any refutation.
I find myself questioning how many readers will actually do the unpacking you describe rather than just use the Worst Argument in the World as a club to beat their opponents over the head. Especially since title is such that it will probably attract many readers off LessWrong.
“Taboo murder.” works better than “Calling X murder is the worst argument in the world!”
I’ve heard anecdotes of philosophy professors dreading the lesson on logical fallacies because the students use them as a weapon. But even so, logical fallacies are pedagogically useful like the worst argument in the world. To know that you should taboo murder rather than continue presupposing “all murder = bad” requires a degree of sophistication, and learning logic and logical fallacies is exactly how you learn to unpack those presuppositions and actually argue rather than score political points.
I think the best practice is to taboo saying “X is Y bias or X is Y logical fallacy”, and rather require people to explain or question the exact flaw in reasoning and possibly why it’s important enough to bring up.
So, for example, if someone says “that’s murder, so it’s evil,” you should then reply with something like “why does something being murder necessitate it being evil?” (all the while internally thinking, “ah ha! I think that was the worst argument in the world.”)
Right, it’s a bad name (but not a bad idea, if you correctly unpack the name).