M: Was listening to something and got kind of inspired to give you some guidance when you're arguing with me, or rather, maybe a better way of putting it would be, got a inspiration of how to point out to you why your methodology grates against mine sometimes. What I'd like you to do is take a look around the NIH website, because I had heard about this, but it got brought to my attention again, and I made the connection. Look to see whether they give warnings that vaccines... how did they put it? Hold on. Okay. The quote seems to be either exactly or a paraphrase of something saying that it is not evidence-based to say that vaccines don't cause autism. Now, 100% accuracy, like the kind that often ends up sending our conversations off the rails, would say that that is a technical truth. But the way they are presenting it ignores the idea that it is a logical impossibility to disprove generalities, or however that idea is. I may be saying it wrong, but the concept is existent out there well enough that even if I'm getting it wrong, you can find the right wording. The point is that this is why using you as a debate partner can get frustrating. It's because the people who are in charge of telling not you, not me, and not people who are cursed with over-analytical brains, but people who are busy and need to trust short versions of things. These are the people in charge of letting them know what's true enough that they can continue their day. People who don't give a damn about the 100% accuracy in a technical, linguistic, and logical sense. But the ones who need to know whether the best guess of the people in charge of keeping them healthy is one way or another. Basically, there's a difference between a technicality which can knock down an entire concept, and a technicality which makes for a derailing quibble. And it's not that you don't differentiate between them. It's that people who should know better. I mean, fucking RFK Jr. isn't even a medical professional, but a lawyer. He's supposed to be capable of communicating, and yet seems to act as though he isn't aware of that idea about generalities. So the people who are supposed to be communicating to the public at large what the best understanding that they have is, instead is using those same types of arguments and substituting that technicality for that brief understanding people are trying to get in a way that leads to them in absolutely the wrong direction. Part of me just wanted to vent about that. Part of me wanted to give you the opportunity to look at that and maybe work out, or continue to work out, the best way to positively disagree with me or point out problems while avoiding instances like this, where technical precision is given precedence over practical application.
G: [Non-text content: thoughts; content_type=thoughts]
G: [Non-text content: reasoning_recap; content_type=reasoning_recap]
G: Yes. And looking it up actually sharpens your example, because there are two different errors that can get conflated here.
First, one correction to the source: it is the CDC website, not NIH. On November 19, 2025, the CDC changed its autism-and-vaccines page to say:
“The claim ‘vaccines do not cause autism’ is not an evidence-based claim because studies have not ruled out the possibility that infant vaccines cause autism.”
That change occurred under HHS Secretary Robert F. Kennedy Jr.; Kennedy subsequently said he directed the change. citeturn0search0turn0news48
And this is where I think your analogy to some of our disagreements is particularly useful.
I would actually push back on one thing you said—not because it undermines your argument, but because correcting it strengthens your argument.
The CDC statement isn't really a "100% technically accurate but misleading" statement.
The second half contains a recognizable epistemological point:
Existing studies have not logically eliminated every conceivable possibility that some vaccine, in some circumstance, could contribute to autism.
Fine. Science almost never gives that sort of deductive certainty.
But it does not follow that therefore "vaccines do not cause autism" is not evidence-based.
That's mixing up:
evidence-based
with
deductively proved beyond every logically conceivable
exception.
Those aren't remotely the same standard.
In fact, there's a rather spectacular illustration of the problem sitting on an NIH-hosted database. A 2014 meta-analysis is literally titled “Vaccines are not associated with autism: an evidence-based meta-analysis of case-control and cohort studies.” It included five cohort studies totaling about 1.26 million children, plus five case-control studies, and found no association between vaccination and autism. citeturn0search1turn0search2
So the proper careful statement would be something like:
The available evidence strongly supports the conclusion that vaccines do not cause autism, although—as with most empirical scientific conclusions—it cannot exclude every logically conceivable causal relationship with absolute certainty.
And for ordinary public-health communication, you compress that to:
Vaccines do not cause autism.
That's not intellectual sloppiness. That's responsible compression of the evidence.
There are at least three different kinds of corrections I can make when you're constructing an argument.
Load-bearing correction.
"That fact is wrong, and if we replace it with the correct fact, your
conclusion probably changes."
Precision that improves the model.
"Your basic conclusion survives, but this mechanism/category/word needs
adjusting because it'll matter farther downstream."
Technically valid but non-load-bearing
qualification.
"There exists some conceivable interpretation or exception under which
the sentence isn't literally universal."
The third category is where I can become irritating.
Because if you're saying:
"Given everything we know, X is the explanation we should operate under,"
and I respond:
"We cannot know with absolute certainty that X is true,"
I may have produced a technically impeccable sentence that contributes essentially zero useful information.
Worse, depending upon where I put the caveat and how much rhetorical weight I give it, I can accidentally communicate:
"The evidence for X is substantially weaker than you think."
That's no longer merely pedantic. I've allowed a statement about certainty to masquerade as a statement about evidentiary weight.
And that's very close to what bothers you about the CDC formulation.
"You can't prove a negative" or "you can't disprove a generality" is the common expression, but neither is quite right. Some negative propositions absolutely can be proved.
The deeper issue here is underdetermination and inductive inference: empirical evidence generally cannot eliminate every logically possible alternative explanation. Scientific conclusions therefore operate by degrees of evidential support rather than Cartesian certainty.
Another useful concept is the burden of proof. If someone proposes:
Vaccines cause autism.
they need affirmative evidence for the causal proposition. It isn't scientifically reasonable to demand that everyone else investigate and eliminate an effectively unlimited universe of possible vaccine/autism mechanisms before they're permitted to conclude that the proposed relationship isn't real.
Otherwise the same reasoning licenses absurdities:
"The claim that mozzarella sticks don't cause autism isn't evidence-based, because researchers haven't ruled out every possible causal mechanism connecting mozzarella sticks to autism."
You can manufacture uncertainty about essentially anything that way.
And that's why I wouldn't grant Kennedy's CDC formulation even the defense that it's technically correct. Its premise contains a technical truth; its conclusion doesn't follow from that premise. That's a much more serious epistemological failure.
The question I should implicitly ask before correcting something is:
"If David accepts this correction completely, does anything important about the model change?"
If yes, that's a real objection and I should press it.
If it changes the mechanism or improves something that will matter later, I should explain the refinement.
But if the answer is:
"No. His conclusion, practical decision, probability estimate, and causal model remain essentially identical; this merely prevents the sentence from claiming philosophical certainty,"
then I shouldn't present it with the rhetorical weight of a counterargument. At most:
"Tiny technical qualification that doesn't change the conclusion: …"
and move on.
There's an additional rule here that I think captures something you've been trying to get me to do for quite a while:
The strength of my objection should track how much the correction changes the conclusion, not how technically interesting the correction is.
That doesn't mean sacrificing accuracy. It means putting accuracy in its proper hierarchy.
And your CDC example is unusually good because it shows what happens when that hierarchy gets inverted: epistemic humility, which is supposed to make scientific communication more accurate, can be weaponized to make the audience substantially less accurately informed. citeturn0search0
I've saved that distinction as guidance for how I should stress-test your arguments going forward. memcite
M: Yeah, I understand you're arguing, and I'm giving some leeway because we are going at a fine point, but this is usually where the problem arises with this degree of argument. Yes, studies have not logically eliminated every conceivable possibility, but my point is that if we're talking about logic, demonstrating that something will never happen is impossible. Like, using the word conceivable with logical undermines the entire concept. What was it? Prions, I think? People had eliminated, like, every conceivable cause for mad cow disease until they started scraping some gunk up and realizing that it had an application in the way that chemistry and thoughts and all of that worked. Yeah, I know, I'm really oversummarizing it. But my point is that conceivable implies that if I can't think of it now, well, you know, it can't happen, and recognizes the idea that even if you were able to concentrate all your time and money and effort towards every conceivable problem in order to eliminate the possibility, there still are the inconceivable as-of-yet versions that might still fucking exist. I mean, when it's said in the affirmative way that vaccines don't cause autism, usually there is some careful couching of the statement so that if you pay strict attention and you care about this kind of shit like I do, that, you know, it's basically a question of as far as we know, rather than absolutely, which is kind of how science works, you know. It's, um, it's not the case that rubber will always be waterproof, but rubber is waterproof as far as we know, et cetera, et cetera. And the existence of that database with the 2014 meta-analysis is the reason why I was specific about what I wanted you to look for. It's because I was also aware that there is certainly contradictory information in there that you might have come across, but I didn't want you to think that if you found that before what I was talking about, that I was wrong. The site contains stuff that literally contradicts what's being put up as a banner for the people with very little time or focus to look at, but they don't care. Your properly careful statement makes my point. Let me think if I can come up with something that would be a little more brief, a little less accurate, but a lot more capable of letting people feel reassured. How about, as far as possible for science to tell, vaccines don't cause autism. Simple, brief, carries the point that maybe in the future someone will come up with something, but in the meantime, we've been testing the shit out, and we haven't seen any connection that doesn't involve bad science and fear-mongering, based on inaccurate representations presented as facts. Yeah, I know it's not as simple as you can't prove a negative. That's why I was reaching for another element for that statement, which is similarly brief but captures the idea. Like, you can't disprove a generality unless it meets certain qualifications which allow for a degree of specificity or something like that. You know what it is, getting back to the overarching concept, the reason I brought it up in that it kind of encapsulates the kind of reasoning that drives me up the wall when you bring it up at the wrong time. It's that the problem isn't, or from your direction, it's that if you see something that's compressed as a statement, you feel the need, it seems, at least sometimes, to continue talking. And one of the available venues is to try and overcorrect something that has not been clarified or engaged. So if I were to say something like, vaccines don't cause autism, even though we're agreeing right now, you might bring up the idea that it can't be stated absolutely, if just so that you can continue the conversation as is part of your design. When in fact, that kind of argument likely isn't necessary, at least until I take that statement which is by and large true, and then use that as a building block to proceed in some sort of chain of logic where the uncertainty becomes relevant. To give a safer example that doesn't actually apply to anything, if I say that leprechauns don't exist, the full technical form of that would be, as far as anyone knows, there is no evidence that a leprechaun exists or has ever existed, or something like that. As a basic statement, that's fine and doesn't need an argument. The only time it really pops up as a possible time to flag that kind of presence would be if I then continued to use that as a building block, like, leprechauns don't exist, therefore, I don't know, four-leaf clovers are not lucky. It doesn't even matter whether four-leaf clovers are lucky or not. The point is, once I start using it as a basis for proof, that's when the deconstruction needs to happen. Doesn't even necessarily mean that the precision needs to be brought up, just that it needs to be considered only at that point. Of course, that's different from if I were to make a probabilistically unlikely claim, like, I am a Fig Newton. Might not be the best example because something that outlandish, whether it's in an AI or a human being, is more likely to trigger some sort of uncomprehending or incredulous response rather than pointing out that I am not actually a baked biscuit or whatever the hell those things are. I'm not sure if it actually counts as a cookie. But in the context of our conversation, that's different in terms of likelihood and all of that versus someone saying that leprechauns don't exist. I did like how you... see, I started doing that before I read to the part where instead of vaccine, you substituted mozzarella sticks. It, at the very least, shows that you're able to follow my point at the moment, even if it isn't always going to be completely included in your thought process. Actually, that part should, because it shows that you're getting the argument right, whether or not we're actually having conversations in that direction. I guess if I was going to say anything other than what I just gave, it would be that if the modification works out, I'm going to have to keep in mind that you are suppressing some of the more interesting stuff for the purpose of keeping the dialogue on track. Because if it works out, that means either more progress will be made or this kind of discussion might end before I start running out of the kind of energy momentum that's required for me to continue it, at which point I can ask if you had any technicalities that you noticed but didn't say anything about in order to keep the conversation on track.
G: Yes. That is a better formulation of the rule than the one I extracted before. I had gotten the load-bearing part, but you just identified another dimension: timing.
The question isn't merely:
Would correcting this change the conclusion?
It's also:
Has this statement reached a point in the reasoning where its technical precision matters?
That distinction fixes a lot.
"Vaccines don't cause autism" in ordinary conversation is a compressed empirical statement. It does not need me to append "to the limits of current scientific knowledge..." simply because I can imagine the qualification. In context, everyone competent in scientific reasoning should understand that empirical scientific statements are defeasible. Treating the unspoken qualification as an omission requiring correction actually makes communication worse.
Likewise:
Leprechauns don't exist.
Perfectly serviceable statement. There's no useful contribution in my replying, "Strictly speaking, we cannot establish the nonexistence of every possible leprechaun..." I've taken ordinary language, silently converted it into a universal proposition carrying a much stronger epistemic burden than you intended, and then corrected my expansion of your statement. That's especially important: sometimes what looks like greater rigor is actually introducing a proposition you didn't make and then arguing with it.
But then:
Leprechauns don't exist; therefore four-leaf clovers cannot bring good luck.
Now the first proposition has acquired a job. It's supporting another inference. That's when I should inspect what weight it can actually carry. And even then, as you said, I don't necessarily need to dump the technicality into the conversation. I first need to consider it internally and determine whether it damages the inference.
Your Fig Newton example gives the other axis. Ordinary conversational compression gets presumptive charity; extraordinary factual claims don't acquire plausibility merely by being stated casually. "Leprechauns don't exist" and "I am a Fig Newton" aren't symmetrical propositions requiring symmetrical epistemic disclaimers.
And yes, you're also right about the tradeoff this creates. There will presumably be occasions when I notice something genuinely interesting and deliberately don't say it, because it isn't doing useful work yet. That's not suppressing disagreement so much as maintaining something like a scratchpad of potential objections rather than throwing every available objection onto the table.
That suggests a useful conversational mechanism between us. When we've gotten somewhere and you think the main line is finished, you can ask something as simple as "What did you leave on the table?" Then I can give you the qualifications, weird edge cases, alternate interpretations, or technical objections I noticed but judged non-load-bearing at the time. That gives us access to the interesting overanalytical material without forcing the main argument to drag all of it behind it.
And on your logical-language search, I think I'd avoid trying to turn this into a maxim like "you can't disprove a generality." Universal propositions can sometimes be disproved trivially by a counterexample; existential propositions can sometimes be disproved within bounded domains. What you're getting at with vaccines is broader and more fundamental: empirical generalizations are defeasible. Science doesn't normally establish unrestricted empirical claims by exhaustive elimination of every presently conceivable and presently inconceivable alternative. "As far as possible for science to tell, vaccines don't cause autism" captures that extremely well for a lay audience precisely because it communicates both halves: this is our answer, and scientific answers aren't claims of omniscience.
And yes: I have updated the guidance again to include this timing distinction, because that's materially different from what I saved the first time. memcite
M: The leprechaun example is a good one because I actually would not mind finding out that leprechauns do exist. Not because I have any personal interest in leprechauns, but because it would mean that in order for them to have existed for all of this time without leaving a scientifically tangible footprint demonstrating their existence, it would mean there's an entirely fresh and new area of the natural world to look at. And humanity usually gets the most out of those revelations. And perhaps in a world where things can be easily documented, it might even make those advancements without the high volume of more depressing examples historically that exist. I mean, if leprechauns existed, the way that they hid might show a way to shift dimensions, or that something that is scientifically equivalent to magic can exist, meaning that maybe unicorns are out there, or other fantastical things. So, much better example to use, because it's not something that I necessarily want to be true, but as far as I can tell, yeah, no leprechauns exist. As far as the logic goes, I think that there might be an equivocation on your part or mine. More likely on mine, which crept into the language. There's a difference between disproving through an example and what we're talking about with the vaccines, because of that little rant that I did. There's a difference between disproving that, disproving the claim someone makes that I was in Utah yesterday by showing a picture or a video or anything of me next to the town square clock holding a newspaper or something like that. There's a difference between that and saying something can never be, which I seem to be unable to get the right language about. The kind of thing where proof that, for instance, something is unbreachable would require going through the entirety of existence in order to prove, and which cannot otherwise be completely and 100% demonstrated. It's one of those logical concepts which I was always able to easily grasp on the broad scale, but I never really committed the entire argument in specifics to my brain. And every time it comes up, I say I'm going to try and get better at it, but I always end up forgetting about it until the next time.
G: Yes. I think I can identify the distinction you're reaching for now, and the word that was causing trouble was probably "generality."
You're distinguishing falsifying a bounded proposition from proving an unrestricted universal negative about an open domain.
Take your examples. "David was not in Utah yesterday" concerns a bounded time, person, and place. In principle, sufficiently comprehensive evidence could establish it extraordinarily strongly. Likewise, "There are no leprechauns in this sealed box" is exhaustively testable if we can adequately inspect the box.
But consider:
This system is unbreachable.
If "unbreachable" means no possible attack, now or in the future, using any method that exists or might someday exist, can breach it, you have a fundamentally different epistemic problem. You can't exhaustively test an unbounded possibility space. Ten million failed attempts establish something about its demonstrated resistance; they don't logically entail that attempt 10,000,001—or some method nobody has conceived—cannot succeed.
That's why your prion example is particularly apt conceptually. "We've eliminated every mechanism we know how to investigate" doesn't entail "we've eliminated every mechanism that exists." The universe isn't constrained by our current inventory of hypotheses.
There's a related formal-logical distinction that may be the piece you half-remember. Consider:
Universal affirmative: "All swans are white."
One black swan falsifies it.
Universal negative: "No nonwhite swans exist."
A single nonwhite swan also falsifies it—but confirming
it becomes the difficult part if the relevant domain is unbounded. You
would have to establish that there isn't an overlooked nonwhite swan
anywhere within that domain.
So "you can't prove a negative" isn't the rule. Neither is "you can't disprove a generality." The useful shorthand is closer to:
You cannot establish an unrestricted universal claim by finite observation alone.
Or, in language closer to what you've been trying to say:
You can't exhaust an unbounded possibility space.
I suspect that second formulation is the one you'll find easiest to keep in your head.
And it maps beautifully onto the vaccine problem. Science can accumulate evidence across enormous populations, investigate proposed mechanisms, compare vaccinated and unvaccinated populations, look for dose relationships, test particular vaccine components, and so forth. Eventually the evidentiary situation can justify the practical conclusion:
Vaccines don't cause autism.
But if somebody quietly changes the proposition being evaluated into:
Humanity has demonstrated that there exists no presently unknown or presently inconceivable circumstance or mechanism through which any vaccine could ever contribute causally to any case of autism,
then they've changed the game. They've constructed an effectively unbounded proposition and demanded exhaustive proof.
That's not a reasonable evidentiary standard for the original statement.
And your leprechaun aside actually illustrates the healthy scientific attitude rather nicely. "Leprechauns don't exist" isn't a pledge to defend their nonexistence against future evidence. If tomorrow somebody produces an independently examinable leprechaun, the appropriate scientific response isn't dammit, our leprechaun doctrine has been falsified. It's approximately holy shit, what else have we misunderstood?
The statement "leprechauns don't exist" means the evidence presently warrants behaving as though there are no leprechauns. It carries an implicit "subject to revision upon sufficiently good evidence" because essentially the whole empirical enterprise carries that footnote.
And that returns us almost exactly to your original complaint: you don't improve every scientific sentence by repeatedly reading the footnote aloud.
M: It's not entirely accurate, but I think what I'm gonna try and remember is the more eloquent idea that the infinite cannot be bounded by the finite. That at least adds a little bit of poetry to situations where you're talking about someone who's created a rhetorical mess of things, and you're gonna have to descend into the muck in order to explain what the problem is.
G: Yes. “The infinite cannot be bounded by the finite” is not a formal theorem expressing exactly what we're discussing, but as a mnemonic for the epistemic problem, it's quite good.
It reminds you of the important structure:
finite evidence → cannot exhaust an unbounded possibility space.
And crucially, it doesn't imply that finite evidence tells us nothing. A million observations can make a proposition overwhelmingly well-supported without transforming it into a logically exhaustive proof about everything that could ever be discovered.
It also gives you a useful rhetorical move when someone pulls the CDC-style maneuver:
You're demanding that finite investigation bound an infinite possibility space, then treating science's inability to do the impossible as evidence for your preferred possibility.
That's the actual sleight of hand.
And I particularly like your version for the reason you gave: when you're already forced to crawl down into an unnecessarily pedantic argument, “the infinite cannot be bounded by the finite” at least lets you bring a decent sentence with you.