Plausible, but can you potentially say more on whether any of the linked articles actually fail to provide substantial arguments? I do agree it’s a tempting thing to do, but it seems to me that providing references for implicit arguments made in a story seems overall substantially better than just leaving them implicit.
Gish gallop is a rhetorical technique in which a person in a debate attempts to overwhelm their opponent by providing an excessive number of arguments with no regard for the accuracy or strength of those arguments
As an attempt to model Gwern’s likely motivations, this seems terrible. You really think there’s no reason to include lots of details in scenario-building or fiction-writing outside of wanting to deceive debate opponents??
You really think the primary motivation of Gwern Gwern.net Branwen for finding the fine details of ML scaling laws interesting (or for wanting to cite sources) is ‘I really want to deceive people into thinking AI is scary’?
I think it’s pretty common in internet writing, and don’t think it should be a hypothesis that people can’t consider.
You really think there’s no reason to include lots of details in scenario-building or fiction-writing outside of wanting to deceive debate opponents??
Clearly this is not the standard of evidence necessary to call something “plausible”. Of course there are other reasons, but I don’t see how that has much of an effect on the plausibility of a hypothesis.
You really think the primary motivation of Gwern Gwern.net Branwen for finding the fine details of ML scaling laws interesting is ‘I really want to deceive people into thinking AI is scary’? Have you met Gwern???
Again, thinking a hypothesis is plausible has very little to do with “what I believe”. It certainly doesn’t take that much evidence to convince me that in a single case, Gwern was executing on some habit that tends to result in overwhelming the reader with enough information that it’s hard for them to really follow what is happening. I would be surprised if Gwern was being super agentic about this, but also don’t even find that hypothesis implausible, though of course quite unlikely.
In the Gish Gallop, you present a bunch of perhaps somewhat related, but fundamentally independent arguments for a position. In the classic Gish Gallop, you give just one or maybe two to start with, wait for people to debunk it, then ignore the knockdown and present another one. Usually you act as if the new one is support for the old one, or as if the new one was what you were saying all along… but you’rereally giving a completely different argument.
The idea is to eventually exhaust the opponent, who is forced to invest time and effort to refute every new argument. It works best if the arguments are hard to understand and even better if they claim to be supported by facts, so the opponent has to do research to try to disprove factoid statements.
Presenting a single argument with support for each step isn’t really like a Gish Gallop. And the hyperlinks in the story are a lot more like a single argument with support for each step than they are like independent arguments for a single position.
If you don’t allow any complicated arguments with lots of steps that need support, you degrade the discussion even more than if you let people change their arguments all the time. And tossing around phrases like “Gish Gallop” (and “Sealion”) is its own kind of rhetorical dirty pool.
Maybe we’re using the word ‘plausible’ differently? Based on context/tone, I read
Plausible, but can you potentially say more on whether any of the linked articles actually fail to provide substantial arguments?
as basically saying “This is probably true, but can you potentially say more on whether any of the linked articles actually fail to provide substantial arguments?”
I would have had no objection to, e.g., ‘This kind of hypothesis is plausible on priors, because a lot of long Internet argument contain some amount of Gish-galloping. But can you potentially say more on whether any of the linked articles actually fail to provide substantial arguments?’
Again, thinking a hypothesis is plausible has very little to do with “what I believe”.
I don’t understand this part. Generally, I interpret “plausible” as meaning “at least ~10% likely” (with a connotation that it’s probably not e.g. 95% likely), and the tone/phrasing/context of your comment made it sound to me like ’50+% likely’ in this case.
It certainly doesn’t take that much evidence to convince me that in a single case, Gwern was executing on some habit that tends to result in overwhelming the reader with enough information that it’s hard for them to really follow what is happening.
It sounds like your prototype for Gish-galloping might be ‘autistic kid who talks way too much about their special interest and makes it hard for other people to get a word in edgewise’, whereas mine is ‘creationist who takes advantage of the timed-debate format to deliberately try to trick people into believing some claim’? The latter scenario is where the term comes from. I wouldn’t normally say “It’s a Gish Gallop.” about anything that wasn’t primarily motivated by a goal of deceiving and manipulating others.
I think part of my negative reaction at the inferential leap here is the lack of imagination I feel like it exhibits. It feels roughly the same as if I’d heard someone on the street say ‘Scott Sumner wrote a blog post that’s more than three pages long?! That’s crazy. The only possible reason someone could ever want to write something that long is if they have some political agenda they want to push, and they think they can trick people into agreeing by exhausting them with excess verbiage.’
It’s not that people never have political agendas, or never inflate their text in order to exhaust the reader; it’s that your cheater-detectors are obviously way too trigger-happy if your brain immediately confidently updates to ‘this is the big overriding reason’, based on evidence as weak as ‘this blog post is at least four pages long’.
If someone then responds “Sounds plausible, yeah”, then I update from ‘This one person has bad epistemics’ to ‘I have been teleported to Crazytown’.
I do feel like you are speaking too confidently of the epistemic state of the author. I do think the opening sentence of “I have a less charitable description of the links” feels like it weakens the statement here a good amount, and moves it for me more into the “this is one hypothesis that I have” territory, instead of the “I am confidently declaring this to be obviously true” territory.
Hmmm. I’d agree if it said “a less charitable hypothesis about the links” rather than “a less charitable description of the links”. Calling it a “description” makes it sound even more confident/authoritative/objective.
To be clear, I think a comment like this would have been great:
I clicked on your first three references, and in all three cases the details made me a lot more skeptical that this is a plausible way the future could go. Briefly, reference 1 faces problem W; reference 2 faces problems X and Y; and reference 3 faces problem Z. Based on this spot check, I expect the rest of the scenario will similarly fall apart when I pick at the details.
The whole story feels like a Gish gallop to me. Presumably at least part of your goal in telling this story was to make AI doom seem more realistic, but if a lot of the sense of realism rests on you dumping in a river of details that don’t hold up to scrutiny, then the sense of realism is a dishonest magic trick. Better to have just filled your story with technobabble, so we don’t mistake the word salad for a coherent gearsy world-model.
If I were the King Of Karma, I might set the net karma for a comment like that to somewhere between +8 and +60, depending on the quality of the arguments.
I think this would have been OK too:
I clicked on your first three references, and I didn’t understand them. As a skeptic of AI risk, this makes me feel like you’re Eulering me, trying to talk me into worrying about AI risk via unnecessarily jargony and complex arguments. Is there a simpler version of this scenario you could point me to? Or, failing that, could you link me to some resources for learning about this stuff, so I at least know there’s some pathway by which I could evaluate your argument if I ever wanted to sink in the time?
I’d give that something like +6 karma. I wouldn’t want LWers to make a habit of constantly accusing Alignment Forum people of ‘Eulering’ bystanders just because they’re drawing on technical concepts; but it’s at least an honest- and transparent-sounding statement of someone’s perspective, and it gives clear conditions that would let the person update somewhat away from ‘you’re just Eulering me’.
Jiro’s actual comment, I’d probably give somewhere between −5 and −30 karma if I were king.
“This is probably true, but can you potentially say more on whether any of the linked articles actually fail to provide substantial arguments?”
No, that’s very far from how I would use the word plausible. I use it to mean “doesn’t seem implausible”, e.g. something closer to “seems like a fine hypothesis to think about”. I don’t know of any other word that communicates an even lower level of probability. My guess is I am currently at around 1% on the hypothesis that Jiro proposes.
Good to know! In colloquial English, I think people would typically say “That’s possible, but...” or “That’s a valid hypothesis, but...” instead of “That’s plausible, but...”, given the belief you were trying to convey.
Unfortunately, this collides with the technical meanings of “possible” and “valid”...
FWIW plausible is actually ambiguous to me. One sense means, “this is sort of likely; less likely than mainline, but worth tracking as a hypothesis, though maybe I won’t pay much attention to it except now that you bring it up”, or something. This would probably be more likely than something called “possible” (since if it were likely or plausible you probably would have called it such). The other sense means “this seems like it *might be possible*, given that I haven’t even thought about it enough to check that it’s remotely meaningful or logically consistent, let alone likely or worth tracking, but I don’t immediately see a glaring inconsistency / I have some sense of what that would look like / can’t immediately rule that out”. The second sense could imply the thing is *less* likely than if it were called “possible”, since it means “might be possible, might not”, though model uncertainty might in some contexts mean that something that’s plausible_2 is more likely than something you called definitely possible.
I’m a native English speaker, and I think of ‘plausible’ as connoting higher probability than ‘possible’ - I think I’d use it to mean something like ‘not totally crazy’.
I think if I have a space of hypotheses, I’ll label ‘probable’ the ones that have >50% probability, and ‘plausible’ the ones that are clearly in the running to become ‘probable’. The plausible options are the ‘contenders for probableness’; they’re competitive hypotheses.
E.g., if I’m drawing numbered balls from an urn at random, and there are one hundred balls, then it’s ‘plausible’ I could draw ball #23 even though it’s only 1% likely, because 1% is pretty good when none of the other atomic hypotheses are higher than 1%.
On the other hand, if I have 33 cyan balls in an urn, 33 magenta balls, 33 yellow balls, and 1 black ball, then I wouldn’t normally say ‘it’s plausible that I’ll draw a black ball’, because I’m partitioning the balls by color and ‘black’ isn’t one of the main contender colors.
See this is exactly the situation where I would say ‘plausible.’ To me ‘plausible’ implies a soon-to-be-followed ‘but’: “It is plausible that I would draw a black ball, but it is unlikely.” It is nearly synonymous with ‘possible’ in my mind.
Native speaker, and my understanding of ‘plausible’ agrees 100% with Kaj. It’s about the lowest possible assessment you can give, while still admitting that it is a possible hypothesis. I believe this is because under normal circumstances you would use literally any other word to give a more charitable assessment, if you wanted. E.g. you could say it is ‘likely’ (but you didn’t), or you could say it is ‘valid’ (which in non-technical English tends to connote some sort of likelihood), etc.
If I come at you with some argument or theory, and you reply “well, I guess that’s plausible” I take the hint that you are actually ending the conversation out of disinterest. You are conceding that it is technically plausible, maybe, but you don’t think it is likely nor even worthy enough to take your time debating.
Good feedback. I will try using “That’s possible” more, instead of plausible, though in my internal monologue “possible” sounds slightly more confident than “plausible”.
Your response would make sense to me if Jiro had said something like ‘I wonder if some part of Gwern was influenced by a desire to Gish-gallop opponents (among other motivations)’. This is really importantly different, in my mind, from the bald assertion “It’s a Gish Gallop.”
It’s also radically different from a neutral warning to readers, ‘hey, be cautious of updating too much on all these fictional details and authoritative-looking references’. In my book, that sort of claim usually has a way lower evidential bar to pass than speculating on someone’s motives, which in turn has a lower bar to pass than asserting an acquaintance with a virtuous track record has highly adversarial motives. (Without feeling a need to argue for your hypothesis, and without first trying to engage in any sort of object-level discussion about any part of the post.)
Oh, I think the comment I am responding to is quite bad, but I don’t think in terms of pure conceptual content, saying “I wonder if X” and “X” is that different. In either case, downvoting, then asking for more evidence seems like a reasonable thing to do (and I think is better than going up to the meta level and talking about whether the comment was phrased the right way, which I think is generally not super productive).
It’s plausible you are reacting to a different social context than I am. When I responded to the comment, the comment was at −6 karma.
In either case, downvoting, then asking for more evidence seems like a reasonable thing to do
Yep, agreed!
(and I think is better than going up to the meta level and talking about whether the comment was phrased the right way, which I think is generally not super productive).
The way a Gish gallop works is that it’s pointless to refute one of the references, because there are too many others that would take too much time and effort to refute.
Are you… also against citing references in scientific papers, which usually cite vastly more than this post? Just because there are many links, does not mean it’s necessary to respond to all content of all links. If anything, citing your references on average makes it easier to respond.
Just because there are many links, does not mean it’s necessary to respond to all content of all links.
I read the phrase
And the fact that they lead to clear examples of the feasibility of such proposal in every single example was impressive.
as implying that each link is evidence (at least to that person, not to the OP) and therefore refuting the initial post would require responding to all of them.
Plausible, but can you potentially say more on whether any of the linked articles actually fail to provide substantial arguments? I do agree it’s a tempting thing to do, but it seems to me that providing references for implicit arguments made in a story seems overall substantially better than just leaving them implicit.
How is this plausible? Per Wikipedia:
As an attempt to model Gwern’s likely motivations, this seems terrible. You really think there’s no reason to include lots of details in scenario-building or fiction-writing outside of wanting to deceive debate opponents??
You really think the primary motivation of Gwern Gwern.net Branwen for finding the fine details of ML scaling laws interesting (or for wanting to cite sources) is ‘I really want to deceive people into thinking AI is scary’?
Have you met Gwern??
I think it’s pretty common in internet writing, and don’t think it should be a hypothesis that people can’t consider.
Clearly this is not the standard of evidence necessary to call something “plausible”. Of course there are other reasons, but I don’t see how that has much of an effect on the plausibility of a hypothesis.
Again, thinking a hypothesis is plausible has very little to do with “what I believe”. It certainly doesn’t take that much evidence to convince me that in a single case, Gwern was executing on some habit that tends to result in overwhelming the reader with enough information that it’s hard for them to really follow what is happening. I would be surprised if Gwern was being super agentic about this, but also don’t even find that hypothesis implausible, though of course quite unlikely.
In the Gish Gallop, you present a bunch of perhaps somewhat related, but fundamentally independent arguments for a position. In the classic Gish Gallop, you give just one or maybe two to start with, wait for people to debunk it, then ignore the knockdown and present another one. Usually you act as if the new one is support for the old one, or as if the new one was what you were saying all along… but you’rereally giving a completely different argument.
The idea is to eventually exhaust the opponent, who is forced to invest time and effort to refute every new argument. It works best if the arguments are hard to understand and even better if they claim to be supported by facts, so the opponent has to do research to try to disprove factoid statements.
Presenting a single argument with support for each step isn’t really like a Gish Gallop. And the hyperlinks in the story are a lot more like a single argument with support for each step than they are like independent arguments for a single position.
If you don’t allow any complicated arguments with lots of steps that need support, you degrade the discussion even more than if you let people change their arguments all the time. And tossing around phrases like “Gish Gallop” (and “Sealion”) is its own kind of rhetorical dirty pool.
Yeah, I think this is part of why the claim seemed out-of-left-field to me.
Maybe we’re using the word ‘plausible’ differently? Based on context/tone, I read
as basically saying “This is probably true, but can you potentially say more on whether any of the linked articles actually fail to provide substantial arguments?”
I would have had no objection to, e.g., ‘This kind of hypothesis is plausible on priors, because a lot of long Internet argument contain some amount of Gish-galloping. But can you potentially say more on whether any of the linked articles actually fail to provide substantial arguments?’
I don’t understand this part. Generally, I interpret “plausible” as meaning “at least ~10% likely” (with a connotation that it’s probably not e.g. 95% likely), and the tone/phrasing/context of your comment made it sound to me like ’50+% likely’ in this case.
It sounds like your prototype for Gish-galloping might be ‘autistic kid who talks way too much about their special interest and makes it hard for other people to get a word in edgewise’, whereas mine is ‘creationist who takes advantage of the timed-debate format to deliberately try to trick people into believing some claim’? The latter scenario is where the term comes from. I wouldn’t normally say “It’s a Gish Gallop.” about anything that wasn’t primarily motivated by a goal of deceiving and manipulating others.
I think part of my negative reaction at the inferential leap here is the lack of imagination I feel like it exhibits. It feels roughly the same as if I’d heard someone on the street say ‘Scott Sumner wrote a blog post that’s more than three pages long?! That’s crazy. The only possible reason someone could ever want to write something that long is if they have some political agenda they want to push, and they think they can trick people into agreeing by exhausting them with excess verbiage.’
It’s not that people never have political agendas, or never inflate their text in order to exhaust the reader; it’s that your cheater-detectors are obviously way too trigger-happy if your brain immediately confidently updates to ‘this is the big overriding reason’, based on evidence as weak as ‘this blog post is at least four pages long’.
If someone then responds “Sounds plausible, yeah”, then I update from ‘This one person has bad epistemics’ to ‘I have been teleported to Crazytown’.
I do feel like you are speaking too confidently of the epistemic state of the author. I do think the opening sentence of “I have a less charitable description of the links” feels like it weakens the statement here a good amount, and moves it for me more into the “this is one hypothesis that I have” territory, instead of the “I am confidently declaring this to be obviously true” territory.
Hmmm. I’d agree if it said “a less charitable hypothesis about the links” rather than “a less charitable description of the links”. Calling it a “description” makes it sound even more confident/authoritative/objective.
To be clear, I think a comment like this would have been great:
If I were the King Of Karma, I might set the net karma for a comment like that to somewhere between +8 and +60, depending on the quality of the arguments.
I think this would have been OK too:
I’d give that something like +6 karma. I wouldn’t want LWers to make a habit of constantly accusing Alignment Forum people of ‘Eulering’ bystanders just because they’re drawing on technical concepts; but it’s at least an honest- and transparent-sounding statement of someone’s perspective, and it gives clear conditions that would let the person update somewhat away from ‘you’re just Eulering me’.
Jiro’s actual comment, I’d probably give somewhere between −5 and −30 karma if I were king.
No, that’s very far from how I would use the word plausible. I use it to mean “doesn’t seem implausible”, e.g. something closer to “seems like a fine hypothesis to think about”. I don’t know of any other word that communicates an even lower level of probability. My guess is I am currently at around 1% on the hypothesis that Jiro proposes.
Good to know! In colloquial English, I think people would typically say “That’s possible, but...” or “That’s a valid hypothesis, but...” instead of “That’s plausible, but...”, given the belief you were trying to convey.
Unfortunately, this collides with the technical meanings of “possible” and “valid”...
I’m a non-native speaker, but to me both “possible” and “valid” connote higher probability than “plausible”.
non-native, to me possible is “technically possible, but not necessarily probable”, while plausible is “possible and slightly probable”.
! Woah!!
I’m a native speaker and I agree with Kaj about the connotations, and use “plausible” to mean roughly the same thing as habryka is.
Woah! Maybe I’m the crazy one! :o
(I would still predict ‘no’, but the possibility has become way more likely for me.)
FWIW plausible is actually ambiguous to me. One sense means, “this is sort of likely; less likely than mainline, but worth tracking as a hypothesis, though maybe I won’t pay much attention to it except now that you bring it up”, or something. This would probably be more likely than something called “possible” (since if it were likely or plausible you probably would have called it such). The other sense means “this seems like it *might be possible*, given that I haven’t even thought about it enough to check that it’s remotely meaningful or logically consistent, let alone likely or worth tracking, but I don’t immediately see a glaring inconsistency / I have some sense of what that would look like / can’t immediately rule that out”. The second sense could imply the thing is *less* likely than if it were called “possible”, since it means “might be possible, might not”, though model uncertainty might in some contexts mean that something that’s plausible_2 is more likely than something you called definitely possible.
Yeah, I think that’s a more complete view of its meaning.
I’m a native English speaker, and I think of ‘plausible’ as connoting higher probability than ‘possible’ - I think I’d use it to mean something like ‘not totally crazy’.
(I think this is how I use it)
I think if I have a space of hypotheses, I’ll label ‘probable’ the ones that have >50% probability, and ‘plausible’ the ones that are clearly in the running to become ‘probable’. The plausible options are the ‘contenders for probableness’; they’re competitive hypotheses.
E.g., if I’m drawing numbered balls from an urn at random, and there are one hundred balls, then it’s ‘plausible’ I could draw ball #23 even though it’s only 1% likely, because 1% is pretty good when none of the other atomic hypotheses are higher than 1%.
On the other hand, if I have 33 cyan balls in an urn, 33 magenta balls, 33 yellow balls, and 1 black ball, then I wouldn’t normally say ‘it’s plausible that I’ll draw a black ball’, because I’m partitioning the balls by color and ‘black’ isn’t one of the main contender colors.
See this is exactly the situation where I would say ‘plausible.’ To me ‘plausible’ implies a soon-to-be-followed ‘but’: “It is plausible that I would draw a black ball, but it is unlikely.” It is nearly synonymous with ‘possible’ in my mind.
Where/when did you guys learn to speak English at (I’m wondering if this is regional/generational)? I grew up in the American Midwest and am 22.
Native speaker, and my understanding of ‘plausible’ agrees 100% with Kaj. It’s about the lowest possible assessment you can give, while still admitting that it is a possible hypothesis. I believe this is because under normal circumstances you would use literally any other word to give a more charitable assessment, if you wanted. E.g. you could say it is ‘likely’ (but you didn’t), or you could say it is ‘valid’ (which in non-technical English tends to connote some sort of likelihood), etc.
If I come at you with some argument or theory, and you reply “well, I guess that’s plausible” I take the hint that you are actually ending the conversation out of disinterest. You are conceding that it is technically plausible, maybe, but you don’t think it is likely nor even worthy enough to take your time debating.
Good feedback. I will try using “That’s possible” more, instead of plausible, though in my internal monologue “possible” sounds slightly more confident than “plausible”.
Your response would make sense to me if Jiro had said something like ‘I wonder if some part of Gwern was influenced by a desire to Gish-gallop opponents (among other motivations)’. This is really importantly different, in my mind, from the bald assertion “It’s a Gish Gallop.”
It’s also radically different from a neutral warning to readers, ‘hey, be cautious of updating too much on all these fictional details and authoritative-looking references’. In my book, that sort of claim usually has a way lower evidential bar to pass than speculating on someone’s motives, which in turn has a lower bar to pass than asserting an acquaintance with a virtuous track record has highly adversarial motives. (Without feeling a need to argue for your hypothesis, and without first trying to engage in any sort of object-level discussion about any part of the post.)
Oh, I think the comment I am responding to is quite bad, but I don’t think in terms of pure conceptual content, saying “I wonder if X” and “X” is that different. In either case, downvoting, then asking for more evidence seems like a reasonable thing to do (and I think is better than going up to the meta level and talking about whether the comment was phrased the right way, which I think is generally not super productive).
It’s plausible you are reacting to a different social context than I am. When I responded to the comment, the comment was at −6 karma.
Yep, agreed!
Agreed.
The way a Gish gallop works is that it’s pointless to refute one of the references, because there are too many others that would take too much time and effort to refute.
Are you… also against citing references in scientific papers, which usually cite vastly more than this post? Just because there are many links, does not mean it’s necessary to respond to all content of all links. If anything, citing your references on average makes it easier to respond.
I read the phrase
as implying that each link is evidence (at least to that person, not to the OP) and therefore refuting the initial post would require responding to all of them.