Beware Natural Language Logic
Modus ponens is the logical operation which goes like:
If A then B
A
Therefore B
You might see examples like:
Or, in natural language
If there is a fire, you should leave the building
There is a fire
Therefore you should leave the building
I think you have to be very careful with natural language modus ponens. Natural language is not as precise as maths. Usually I see someone start a modus ponens I expect I’m about to be hit with some absolutely fudgey nonsense of the form.
If unlikely-sounding premise, then my preferred crazy-sounding conclusion
Unlikely-sounding premise
My preferred crazy-sounding confusion
Or even
If a thing has property X, then we should do Y about that thing
Thing that definitely sounds like it should not have property X has property X
Therefore we should do Y about the thing that definitely seems like it should not have property X
Usually, I think the culprit is that either the premise is dodgy, or the property X is dodgy, and by “dodgy” I mean poorly specified.
For Example
If an issue plausibly causes a million times more suffering than anything else in the world, then it’s the only thing worth working on
Insectg suffering plausibly causes a million times more suffering than anything else in the world
Therefore we should all be working on insect suffering
There’s two angles to this, the rhetorical angle and the actual logical problems with it.
It’s a useful rhetorical trick. It’s convincing. Why? I think it works so well because we tend to read things in the order they’re presented, to not think very far ahead, and to stick to things we feel we’ve committed to. When we see the first claim, it seems reasonable, and so it gets agreed with and cached. When we see the second claim (plus some evidence) it also seems reasonable, so it gets agreed with and cached. When we see the third claim, we’ve kinda forced ourselves into a corner where there’s friction to go back and disagree with our past selves.
In some ways this is good! Rhetoric is not a dirty word, convincing people of true things is how people end up believing true things.
My issue with this example is that it uses skilled rhetoric combined with faulty reasoning. In this case, the word “plausible” is doing a huge amount of work here.
Humans have a natural tendency to smooth over ideas into ones which make more sense. In this case, “first_sentence.plausible” means something like “a probability that’s reasonably large”. Meanwhile, “second_sentence.plausible” means something like “we can’t rule this out”.
(As an aside, “plausible” is my personal least favourite EA/rat word and I endeavor to never say it. If you catch me saying it in real life, you can let me know and I’ll squat to the perfect height for you to slap me across the face (or stand on something if you’re that much taller than me).
I think “plausible” also elides a map-territory distinction in a way that “possible” doesn’t: if I can plause something, you can’t really stop me. If I say something is “possible” then you can say “I don’t think it’s possible” and our disagreement becomes clearly factual.)
Asymmetric Weapons
Scott has written that useful “weapons” in arguments are ones which favour the side which is correct. I don’t think this kind of pseudo-logic is asymmetrical, at least when used in this particularly rhetorically-charged way.
I think a better way of presenting it is in a different order:
I think insect suffering might cause a million times more suffering than human suffering
…and that’s actually the whole interesting part of the claim, which isn’t being placed between two other claims like a trojan horse.
There’s another good way of presenting logic, which is (if I understand correctly) how Nick Bostrom actually used to talk about the simulation hypothesis:
Spacefaring civilizations might simulate many copies of pre-spacefaring ones
If they do, we’re more likely to be in a simulation than not
Therefore we’re more likely than not to be in a simulation
…That’s pretty crazy. Let’s all sit around and have a chat at whether one of the premises might be broken!
This is, again, more polite. You’re not trying to verbally corner your readers, you’re trying to enlighten them. So be mindful of when you’re being attacked by rhetoric, and equally mindful of when you’re firing off rhetorical shots into a residential neighbourhood.
Editor’s note, this post was written as part of Doublehaven (unaffiliated with Inkaven)
◆◆◆◆◆|◆◆◆◆◆|◆◆◆◆◇
◆◆◆◆◆|◆◆◆◆◆|◆◆◆◆◇
typo?
I think it’s fairly clearly a small joke.
Apparently you can select the text, and then add a ‘reaction’, and one of the reactions is ‘this is a typo’. And then a barely visible marker pops up next to the text. Maybe you already know this, but I just figured it out, and there could be other newbies that missed it.
I’m always scared of pointing out typos, because on LessWrong it feels like everything you say should be super important, so I am very glad the ‘typo’ reaction exists. Although I first need 5 karma, so I am not allowed to point out typos before I post some meaningful stuff.
Edit: Oh dear, now I read the text again, and it’s really not clear if ‘plause’ is a typo.
It’s pretty clearly a nonce backformation from “plausible”. There’s some small chance the author wasn’t aware that it wasn’t already an “established” word and would have written something else if they had known it wasn’t, but basically zero chance they meant to type some specific other word but typed “plause” instead by accident.
I’m not sure the ambiguity of “plausibility” is the problem with the conditional there. Suppose we replaced plausibility with “>10% chance as judged by an ideal reasoner knowing everything either of us know.” Then the conditional still couldn’t be true as a general principle, because there could be more than one cause area meeting that protosis, but the apodosis specifies uniqueness. Maybe insect suffering and AI suffering are both 15% likely to be millions of times greater in importance than anything else, but they can’t both be the only things worth working on.
OTOH a symmetric framing like
is much more, ah, plausible (although I might still be reluctant to embrace it.) On learning of the potential scope of insect suffering one should give it some pretty serious consideration.
(Agreed that replacing linear arguments with n-lemmas is in most cases an improvement.)
But that’s not the important rhetorical point that the claimant likely intends to make in this case. EAs know that insect suffering sounds obviously crazy to the uninitiated, so the likely target audience here are people who already think that animal suffering charities of some sort (e.g. factory farming-related) satisfy the “effective” criterion, which is equivalent to thinking that animal suffering (at some scale) could be more important than human suffering. From there the claim that suffering among one subset of animals swamps all other by orders of magnitude isn’t crazy at all.
Beyond the classic rhetorics that can happen with ‘possibilities’, hypotheticals, and very very high values, I definitely don’t resonate with the examples here.
The human world is deeply broken in coordination problems, not even able to fix many human suffering problems, so the path to (non-human) animal suffering might actually be to fix all human suffering first, whereas “therefore we should all be working on animal suffering” does not imply.
The simulation hypothesis implies consciousness can be created from simulation which may not be true at all, the number of books in the world does not actually create any fictional character suffering.
I’m not sure if you find that related to the article, but I guess it just didn’t land for me in any way I can see because the examples I find flawed due to some premise and not because of the rhetoric.