The so-called "scientific method" is, I think, rather poorly understood. For example, let us consider one of the best-known laws of nature, often simply referred to as the Law of Gravity:
Newton's Law of Universal Gravitation: Every object in the universe attracts every other object toward it with a force proportional to the product of their masses, divided by the square of the distance between their centers of mass.
Now here is a series of questions for you, which I often ask audiences when I give lectures on the philosophy of science:
- Do you believe Newton's Law of Universal Gravitation is true?
- If so, how sure are you that it is true?
- Why do you believe it, with that degree of certainty?
The most common answers to these questions are "yes", "very sure", and "because it has been extensively experimentally verified." Those answers sound reasonable to any child of the Enlightenment -- but I submit, on the contrary, that this set of answers has no objective basis whatsoever. To begin with, let us ask, how many confirming experiments do you think would have been done, to qualify as "extensive experimental verification." I would ask that you, the reader, actually pick a number as a rough, round guess.
Whatever number N you picked, I now challenge you state the rule of inference that allows you to conclude, from N uniform observations, that a given effect is always about from a given alleged cause. If you dust off your stats book and thumb through it, you will find no such rule of inference rule there. What you will find are principles that allow you to conclude from a certain number N of observations that with confidence c, the proportion of positive cases is z, where c < 1 and z < 1. But there is no finite number of observations that would justify, with any nonzero confidence, that any law held universally, without exception (that is, z can never be 1 for any finite number of observations, no matter how small the desired confidence c is, unless c = 0). . And isn't that exactly what laws of nature are supposed to do? For Pete's sake it is called the law of universal gravitation, and it begins with the universal quantifier every (both of which may have seemed pretty innocuous up until now).
Let me repeat myself for clarity: I am not saying that there is no statistical law that would allow you to conclude the law with absolute certainty; absolute certainty is not even on the table. I am saying that there is no statistical law that would justify belief in the law of universal gravitation with even one tenth of one percent of one percent confidence, based on any finite number of observations. My point is that the laws of the physical sciences -- laws like the Ideal gas laws, the laws of gravity, Ohm's law, etc. -- are not based on statistical reasoning and could never be based on statistical reasoning, if they are supposed, with any confidence whatsoever, to hold universally.
So, if the scientific method is not based on the laws of statistics, what is it based on? In fact it is based on the
Principle of Abductive Inference: Given general principle as a hypothesis, if we have tried to experimentally disprove the hypothesis, with no disconfirming experiments, then we may infer that it is likely to be true -- with confidence justified by the ingenuity and diligence that has been exercised in attempting to disprove it.
In layman's terms, if we have tried to find and/or manufacture counterexamples to a hypothesis, extensively and cleverly, and found none, then we should be surprised if we then find a counterexample by accident. That is the essence of the scientific method that underpins most of the corpus of the physical sciences. Note that it is not statistical in nature. The methods of statistics are very different, in that they rest on theorems that justify confidence in those methods, under assumptions corresponding to the premises of the theorems. There is no such theorem for the Principle of Abductive Inference -- nor will there ever be, because, in fact, for reasons I will explain below, it is a miracle that the scientific method works (if it works).
Why would it take a miracle for the scientific method to work? Remember that the confidence with which we are entitled to infer a natural law is a function of the capability and diligence we have exercised in trying to disprove it. Thus, to conclude a general law with some moderate degree of confidence (say, 75%), we must have done due diligence in trying to disprove it, to the degree necessary to justify that level confidence, given the complexity of the system under study. But what in the world entitles us to think that the source code of the universe is so neat and simple, and its human denizens so smart, that we are capable of the diligence that is due?
For an illuminating analogy, consider that software testing is a process of experimentation that is closely analogous to scientific experimentation. In the case of software testing, the hypothesis being tested -- the general law that we are attempting to disconfirm -- is that a given program satisfies its specification for all inputs. Now do you suppose that we could effectively debug Microsoft Office, or gain justified confidence in its correctness with respect to on item of its specification, by letting a weasel crawl around on the keyboard while the software is running, and observing the results? Of course not: the program is far too complex, its behavior too nuanced, and the weasel too dimwitted (no offense to weasels) for that. Now, do you expect the source code of the Universe itself to be simpler and friendlier to the human brain than the source code of MS Office is to the brain of a weasel? That would be a miraculous thing to expect, for the following reason: a priori, if the complexity of that source code could be arbitrarily large. It could be a googleplex lines of spaghetti code -- and that would be a infinitesimally small level of complexity, given the realm of possible complexities -- namely the right-hand side of the number line.
In this light, if the human brain is better equipped to discover the laws of nature than a weasel is to confidently establish the correctness an item in the spec of MS Office, it would be a stunning coincidence. That is looking at it from the side of the a priori expected complexity of the problem, compared to any finite being's ability to solve it. But there is another side to look from, which is the side of the distribution of intelligence levels of the potential problem-solvers themselves. Obviously, a paramecium, for example, is not equipped to discover the laws of physics. Nor is an octopus, nor a turtle, nor a panther, nor an orangutan. In the spectrum of natural intelligences we know of, it just so happens that there is exactly one kind of creature that just barely has the capacity to uncover the laws of nature. It is as if some cosmic Dungeon Master was optimizing the problem from both sides, by making the source code of the universe just simple enough that the smartest beings within it (that we know of) were just barely capable of solving the puzzle. That is just the goldilocks situation that good DM's try to achieve with their puzzles: not so hard they can't be solved, not so easy that the players can't take pride in solving them
There is a salient counterargument I must respond to. It might be argued that, while it is a priori unlikely that any finite being would be capable of profitably employing the scientific method in a randomly constructed universe, it might be claimed that in hindsight of the scientific method having worked for us in this particular universe, we are now entitled, a posteriori, to embrace the Principle of Abductive Inference as a reliable method. My response is that we have no objective reason whatsoever to believe the scientific method has worked in hindsight -- at least not for the purpose of discovering universal laws of nature! I will grant that we have had pretty good luck with science-based engineering in the tiny little spec of the universe observable to us. I will even grant that this justifies the continued use of engineering for practical purposes with relative confidence -- under the laws of statistics, so long as, say, one anomaly per hundred thousand hours of use is an acceptable risk. But this gives no objective reason whatsoever (again under the laws of statistics) to believe that any of the alleged "laws of nature" we talk about is actually a universal law. That is to say, if you believe, with even one percent confidence, that we ever have, or ever will, uncover a single line of the source code of the universe -- a single law of Nature that holds without exception -- then you, my friend, believe in miracles. There is no reason to expect the scientific method to work, and good reason to expect it not to work -- unless human mind was designed to be able to uncover and understand the laws of nature, by Someone who knew exactly how complex they are.
Jump in the discussion.
No email address required.
Notes -
Thank you.
In return, I'll save you some effort about getting to grips with Popperian notions of falsifiability by pointing out that they're obsolete.
Popper claimed that a single contradictory finding is sufficient to sink a hypothesis, whereas no amount of evidence can ever prove it with 100% confidence. In other words, you can't prove anything, only disprove it.
Bayesianism goes even further. It asserts, with the maths to prove it, that it is impossible, mathematically so, to achieve either 0 or 100% confidence in a hypothesis without starting there, at which point literally no finite amount of evidence will sway you.
Starting anywhere in between, it would take an infinite amount of evidence to raise confidence in a hypothesis to 100%, or to reduce it to 0%. And if you start with 100% credence or 0% levels of disbelief, nothing anyone can do to you short of invasive neurosurgery (or maybe a shit ton of LSD) can change it.
What are the practical ramifications? Well, here, what Nelson is trying to argue is a waste of time. If you demand 100% confidence that the laws of physics are "universal" and timeless, you're SOL unless you assume the conclusion in advance. But we can approach arbitrarily close, and the fact that modern technology works is testament to the fact that we can be goddamn bloody confident in them. And worst part is that it's not the poor laws of physics at stake here, it's everything you don't hold axiomatic.
Skip Popper. Get on the Bayes Boat, baby, it's all you need.
Here's a few links if you're curious:
0 And 1 Are Not Probabilities (at least in the Bayesian sense)
And Scott on the Predictive Processing theory of cognition which holds that all human cognition is fundamentally Bayesian, even when it breaks.
This is mistaken. There are two quantifiers in an assertions about laws of nature: one might be called generality, which refers to the uniformity with which the law is believed to hold, and the other might be called confidence, which refers to the degree of belief that the law holds with the given generality. For example, if I say I firmly believe that at least 1% of crows are black, this statement would have high confidence and low generality -- whereas if I said, It is plausible that at least 99% of crows are black, that statement would have lower confidence and higher generality. Nothing in any of my posts mentioned 100% confidence; my thesis is about nonzero confidence in 100% generality.
Funny thing: everybody loves Bayes rule; but they never state their priors. To that extent they never consciously use it. Nor is there any evidence that it models the unconscious process of real life rational cognition. The evidence to support that would need to be quantitative; not just "Hey I believed something, then I saw something, and I altered my degree of belief. Must have been using Bayes!"
You evidently don't hang around LessWrong enough.
While Predictive Processing theory, which posits that human cognition is inherently Bayesian, has not been established to the extent it's nigh incontrovertible, it elegantly explains many otherwise baffling things about human cognition, including how it breaks when it comes to mental illnesses like depression, autism, OCD, and schizophrenia. I've linked to Scott on it before. I think it's more likely to be true than not, even if I can't say with a straight-face that it's gospel truth. It is almost certainly incomplete.
In other words, humans are being imperfect Bayesians all the time, and you don't need to explicitly whip out the formula on encountering evidence to get by, but in situations where the expected value of doing so in a rigorous fashion is worth it, you should. The rest of the time, evolution has got you covered.
Besides, the best, most accurate superforecasters and people like quants absolutely pull it out and do explicit work. In their case, the effort really is worth it. You can't beat them without doing the same.
I know quants do this, but I think it is a special case. Show me a hundred randomly selected people who are making predictions they suffer consequences for getting wrong, and are succeeding, I will show you maybe 10 (and I think that's generous) that are writing down priors and using Bayes rule. Medical research, for example, uses parametric stats overwhelmingly more than Bayes (remember all those p-values you were tripping over?), as do the physical sciences.
If the effective altruism (EA) crowd are in the habit of regularly writing down priors (not just "there exist cases"), then I must be mistaken in the spirit of my descriptive claim that nobody writes them down. On the other hand, I would not count EA as people who pay consequences of being wrong, or that is doing a demonstrably good job of anything. If they aren't doing controlled experiments (which would absolutely be possible in the domain of altruism), they are just navel gazing -- and making it look like something else by throwing numbers around. I have a low opinion of EA in the first place; in fact, in the few cases where I looked at the details of the quantitative reasoning on sites like LessWrong, it was so amateurish that I wasn't sure whether to laugh or cry. So an appeal to the authority if LessWrong doesn't cut much ice with me.
I should give an example of this. Here is an EA article on the benefits of mosquito nets from Givewell.org. It is one of their leading projects. (https://www.givewell.org/international/technical/programs/insecticide-treated-nets#How_cost-effective_is_it). At a glance, to an untrained eye, it looks like an impressive, rigorous study. To a trained eye the first thing that jumps out is that it is highly misleading. The talk about "averting deaths" would make an untrained reader think that they are counting the number of "lives saved". But this is not how experts think about "saving lives" and there is a good reason for it. Let's suppose that we take a certain child, that at 9 AM our project saves him from a fatal incident; at 10 Am another, at 11 AM another, but at noon he dies from exactly the peril our program is designed to prevent. Yay, we just averted 3 deaths! That is the stat that Givewell is showing you. Did we save three lives? no, we saved three hours of life.
This is the way anyone with a smidgeon of actuarial expertise thinks about "saving lives" -- in terms of saving days of life, not "averting deaths", and the Givewell and Lesswrong people either know that or ought to know it. If they don't know it, they are incompetent; and if they know it, then talking about "averting deaths" in their public facing literature is deliberately deceptive because it strongly suggests "saving lives", meaning whole lives, in the mind of the average reader. To be fair to givewell, their method of analyzing deaths averted apply to saving someone from malaria for a full year (not just an hour), but (1) that would not be apparent to a typical donor who is not versed in actuarial science, and (2) the fact remains that you could "avert the death" of the same person nine times while they still died of malaria (the peril the program is supposed to prevent) at the age of 10. The analysis and language around it is either incompetent or deceptive -- contrary to either one word or the other in the name of the endeavor, effective altruism.
That's not a cherry picked example; it was the first thing I saw in my first five minutes of investigating "effective altruism". It soured me and I didn't look much further, but maybe I'm mistaken. Maybe you can point me to some EA projects that are truly well reasoned, that are also on the top of the heap for the EA community.
I can't imagine you've spent any time in the EA community and never encountered the concept of the QALY. So yes, they do know it.
Either your being dishonest or you've hardly spent anytime reading anything they've written.
Or your wrong that:
That's your opinion. Mine is that anyone who can understand the rest of the article can understand the difference.
Oh. Ok then.
If you are an EA buff, I'd be happy if you'd share with me the case for one of their better, or best projects.
The thing is I'm not. I've just read Scott's posts regularly, and the occasional link from his subreddit, yet even I knew enough to know they understood the concept of a qaly.
And then by doing a two second google search for: "qaly" mosquito nets site:givewell.org I found this:
https://files.givewell.org/files/ExternalReviews/CompletedReviews/FIECON_AMF_CEM_Technical_Report_December_2017_FINAL.pdf
I have no idea how rigorous it is, but it at least seems the information you are looking for can be easily found if you put in a modicum of effort.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I'm not an EA. I couldn't give less of a fuck about the moral valence of shrimp.
I don't make nearly as strong a claim as you like, so I'm not compelled to submit the arguments you desire. As I've said before, intelligent, functional humans are rational enough that explicit training in the "arts" of rationality, including Bayesian reasoning, provides little mundane value in everyday life to justify the added effort of engaging in it. The opportunity cost of doing so every day is simply not worth it, it's an arsenal best reserved for Big Ticket decisions, and even then, cultural wisdom and "common sense" are not terrible. You don't need a black belt from Yudkowsky's dojo to know not to finance a 2024 Dodge Charger at 30% APR or rack up credit card debt, or even to make sensible middle class decisions.
But the fact is that when you can justify it, such as in forecasting (financial or otherwise), you see the pros doing it. Hence it must be superior to fudging it, all the more so when the stakes are high. The existence of any alpha at that level is surprising enough.
More options
Context Copy link
More options
Context Copy link
I'm skeptical of this. I think predictive processing theory posits a model with certain qualitative features that Bayesian updating would also have, but there are scads of non-Bayesian approaches that would also have those qualitative properties. They would only look Bayesian from the point of view of someone who doesn't know any other theories of belief updating. Does PPT posit a model that have the quantitative properties of Bayesian updating in particular, and experimentally validate those? That would be a very interesting find. If you know of a source I'd be curious to look at it.
I haven't looked into it in that much detail, but I genuinely find the elegance of its explanations in multiple, highly different cognitive disorders, to be hard to explain if it didn't have a kernel of truth.
But as I've admitted, it could be a just-so story or false pattern matching. I'm not that confident in it just yet.
If I see anything more convincing, I'll try and spread it around. For now, it's the domain of better psychiatrists, neurologists and researchers than me.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
How can we approach arbitrarily close? As stated, this does nothing to address Hume's argument against induction, which holds equally whether you are aiming for probability or for certainty, and does not address the retro skeptical argument that every reason you can give is either based on something else or based on nothing, leading to infinite regress. I don't see how Bayesianism helps with this. Justification is not to be had, with any level of confidence or probabilty. Which is why you need Popper, who explained how you can maintain the use of logic and reason and maintain truth as the aim of science, while also accepting Hume's and the skeptical arguments as correct and consequently discarding justification alltogether.
Another issue Bayesianism often runs into is that many variants of Bayesianism give up on truth - I'm not interested in the confidence we can assign to a theory given our priors and the evidence, I'm interested in whether the theory in question is actually true. Even if we could be justified in Bayesian calculations of probabity/confidence (which we can't be), this would tell us exactly nothing about whether this probable theory is actually true, which is what we are really interested in. There is no logical connection between probable truth and truth (just because something is probably true, it need not be true), and Bayesianism often focuses on subjective calculations of probable truth and abandons actual truth as the goal of science. But if Bayesianism aims at truth rather than solely at subjective calculations of confidence unmoored from reality, if it is interested in what is true rather than just what we can be confident in, it is in no better a position to provide justification than any other epistemology.
By amassing more evidence from observations and updating accordingly. Physics demands 5 sigmas of confidence in experimental results before accepting an experiment as valid. For most purposes, you can get away with a lot less.
Nobody has a solution to infinite regress, barring "I said so". As far as I can tell, you've got to start somewhere, and Bayesianism leads to more sensible decision theories and is clean and simple.
"The next sentence is false. The previous sentence is true." Good luck.
Given that English is an imprecise language, feel free to interpret my 99.9999% confidence that the Sun will rise tomorrow as being equivalent to "it's true the Sun will rise tomorrow".
The universe we live in does not provide us the luxury of not being "subjective" observers. Bayesianism happens to be entirely cool with that.
I have no problem with starting somewhere, but I don't claim our theories can ever be anything more than a guess, since, as you seem to have agreed, they are ultimately baseless due to infinite regress. In the context of this discussion on justification and the basis of science, I'm ok with Bayesianism that only claims to be decision theory, a formalized account of how we try to temper our guesses by reason and experience with no justification or basis ever being provided, which is also the Popperian view of the epistemic status of science. Bayesianism would then be a methdology to help in our conjectural decisionmaking, but would never elevate our theories beyond the status of a guess, in the sense of them having some sort of justification or basis. Do we disagee here?
Ok, so if I'm understanding you right, you do care about the truth of your beliefs, not just about your confidence in them. So what's the logical relationship between your calculation of confidence in a theory and the truth of that theory? What is the epistemic benefit of confidence calculation, as opposed to a Popperian conjecture? It seems to me that if you are mistaken about the truth of the belief in question (as you would be with regard to the sun rising tomorrow if you went to, say, Iceland in winter), your high calculated confidence does nothing to mitigate your mistake. You are equally wrong as a Popperian who would just say he guessed wrong, despite your high confidence. And if the belief in question is true, it's just as true for the Popperian who only claims it to be a guess, regardless of confidence calculation. So what is the epistemic benefit of the confidence calculation?
To clarify a bit more, I see two questions we are discussing. First, whether Popper's falsificationist "logic of science" is a better description/methodology of science than Bayesianism. We can set that aside for now, as it is not the focus of the topic. The second question that's relevant to the topic at hand is whether you think Bayesianism can provide some sort of justification or rational basis for claims about the truth of our beliefs that elevates them to something more than a guess. We certainly seem to agree that we can temper our guesses using logic and reason and experience, but in the Popperian view all of this is still guesswork, and never elevates the epistemic status of a theory beyond that of a guess. So tell me if and where we disagree on this :)
We don't. I only wish to emphasize that since we have no reason to think that nothing else can, this is not grounds to knock Bayesianism.
Bayesian updates are, as I understand it, the optimal way to update on new evidence in the light of your existing priors, and with sufficient evidence, two Bayesian who start out with very different credences can converge to the same place. The Aumann Agreement theorem proves that two perfect Bayesians who are able to full share all their information and priors must converge to unanimity, don't ask me how practical that happens to be.
Logical truths are just propositions we have very high confidence in, downstream of axioms or valid arguments in which we also vest high confidence. Is it possible, as in, is there a non-zero chance, that you or I could be hallucinating and imagining obvious truth where there is none? Well, the base rate of insanity or hallucination is greater than 1 in 8 billion. In practise, we can ignore this, and people who wish to function will summarily do so.
As I've demonstrated, the binary notion of truth you hold so dear is simply unattainable, so you should take what you can get in that regard.
Popperian falsifiability is simply dysfunctional. Taken at face value, recall those experiments that suggested neutrinos move faster than light? That is evidence that neutrinos move faster than light.
A serious Popperian would immediately give up on the idea that nothing can exceed the speed of light in a vacuum. A sensible Bayesian, which humans (and thus physicists) are naturally inclined to be, would note that this evidence is severely outweighed by all the other observations we have, and while adjusting very slightly in favor of it being possible to exceed the speed of light with a massive object, still sensibly choose to devote the majority of their energy to finding flaws in the experiment.
Which did in fact turn out to exist.
Bayesianism applied rigorously is simply far more robust than Popperian notions can be, and to the extent the latter was deemed to be workable, it was only by (pragmatically) ignoring minor bits of evidence against one's hypothesis.
Looks like we're on the same page on the overall epistemic status of scientific theories, namely that they are not justified by the evidence and always remain conjectural. That's not a knock against Bayesianism, I agree!
The optimal way in order to do what? What would you say is the aim of science?
For me, it's the commonsense notion of truth as correspondence to reality. Of course, we cannot know or be justified in believing that our theories are true, but they can still be true guesses if they correspond to reality, and it is this reality that we are interested in.
You say that this binary notion of truth is unattainable, so what do you replace it with? Probability calculations? What do those achieve? What is their relationship to reality? There are many variants of Bayesiansim, and they are often very fuzzy about this point, so I'm trying to pinpoint your position.
This is not quite right. For a Popperian, accepting the results of a test is a conjecture just like anything else. We are searching for errors in our guesses by testing them against reality - if we suspect a test is wrong, we are very welcome to criticize the test, devise another test, etc. It is only if we accept a negative test result that have to consider the theory being tested to be false, by plain old deductive logic since it is contradicted by the test result. But a serious Popperian is quite capable of being suspicious of an experiment, and looking for flaws in it.
This might sound tautological, but it makes sense (to me). When I express high certainty in a hypothesis, that is tantamount, and fully interchangeable, with me expressing that in expectation, I do not expect to see evidence that contradicts the hypothesis.
When someone uses "true" in the usual sense, they are expressing that they have very minimal expectation of being contraindicated by future observation (they might even think this is precisely zero, which is not true, except by fiat or axiom, but I'm not going to quibble). When truth is used in a logical or mathematical sense, it is assumed that each valid intermediate step starting from axioms loses so little probability mass that you can have almost the same level of certainty in the compound hypothesis. The loss is unavoidable, but once again, in practise we can ignore it.
You use truth in the sense that something is consistently and robustly reproducible in the observed environment or universe (or more abstract spaces). Sure. I have no beef with that.
Things can be true and truer. "Men are usually taller than women" can be true even if I pull out the Amazonian chick I just matched with on Bumble to humble you (well, she's only 6', she'd outstrip me in slippers let alone heels). But the more reliably that men are taller than women, the truer this gets and this is reflected in the probability of any given comparison between a random man and woman supporting it.
"Men are taller than women", when stated baldly and without regard to how real humans speak, can be taken to mean that there exists no woman taller than a man. In other words, the probability of finding a woman taller than a man is 0% (or 0+epsilon% if you want more rigor).
Probability and truth can be converted (I'm not sure if they're perfectly interconvertible, but when you say that 3>2 is true, there is a tiny chance you are actually wrong and crazy or a gamma ray or electron tunneling has fucked up even the ECC memory k your computer, and this can never be eliminated, only reduced to tolerable levels such that with sufficient engineering you can keep on making such true statements till the Heat Death, which is good enough for me).
What is clear to me that noise and error are unavoidable, and hence perfect credence out of bounds for us poor computationally bounded entities. But we're used to fighting back.
I am confused. What is the criteria that a Popperian uses to determine what error even is, without smuggling in a bit of Bayesianism?
A Bayesian would say that even an (in hindsight) erroneous result is still evidence. But with our previous priors, we would (hopefully) decide that the majority of the probability mass is in the result being in error than the hypothesis.
The only criterion for something to count as evidence for a hypothesis is if it is more likely to be seen if the hypothesis is true, and vice versa.
In other words, how are you becoming suspicious without Bayesian reasoning (however subconscious it is)?
I like this discussion, it feels like we're really doing our best to understand each other's position better, as it should be.
Let's start with the easy bit.
That's right! Bayesianism is, on one level, an attempt to formalize exactly this kind of thinking. I'd call it "critical discussion" or "conjecture" or just "reasoning", you can call it Bayesian reasoning, I'm totally fine with that. I think this process is messier than Bayesianism claims, but I also think that plenty of Popperian concepts can do with some formalizing, like "degree of corroboration", "state of the critical discussion", "severity of testing in light of our background knowledge", etc. So on this level I'd say Bayesianism is a worthwhile pursuit of formalizing this kind of thinking, and we can set aside the question of how well it does this for now. I say "on this level" because there are many variants of Bayesianism, and some claim Bayesianism does much more, such as Bayesianism providing a (partial) justification and rational basis for scientific theories instead of induction, which we have agreed it cannot do, i.e. it cannot elevate our theories beyond the status of a guess.
It is with regards to truth and the relationship of Bayesian calculations to external reality that I suspect we disagree, although I'm not quite sure about this. There's definitely still some stuff to clear up here.
No, this is definitely not the way in which I use truth. Like I said, I use truth in the sense of correspondence to reality. Our subjective expectations, reproducibility, and observations are totally irrelevant to truth. The only thing that matters with regard to the truth of a claim is whether our claim corresponds to reality, i.e. whether things really are the way we claim they are. So the claim that moving faster than light is not possible is true only if this corresponds to reality, i.e. if it is really, objectively, impossible to move faster than light in this universe. We can do all kinds of experiments on this, but the only thing that matters with regard to the truth of this statement is whether it is actually possible to move faster than light in reality or not. Of course, we cannot know what is true, but that does not stop our claim from being true if does in fact correspond to reality, despite it being just a guess. The experiments are attempts to eliminate our errors in our guesses about reality. And while we can never be sure (or be partially justified) in thinking what we believe is true, it can still be true, objectively, if we have guessed right, if it really is impossible to move faster than light. So we cannot be justified (fully or partially) in thinking what we believe is true, but our theories can still be true, objectively, if they correspond to reality whether we know it or not.
In my view, this is what science is interested in: how the universe really works regardless of what we think and the evidence we have. And the best we can do in figuring this out is to guess, and temper our guesses by critical reasoning and empirical testing (which is also guesswork) in order to eliminate errors in our guesses.
Bayesianism often gives up on this goal, and confines itself to only our expectations and related calculations. It thus gives up on saying anything about objective reality, and confines itself to talking about our subjective states of expectation and calculations of probabilty regarding those expectations. This, for me, is an unacceptable retreat from figuring out objective reality as the aim of science. I'm interested in how the world really is, I'm not interested in our subjective calculations of probabilty (other than if they can perhaps help in the former).
Would this describe your position though? That we cannot rationally say anything about the external, objective reality, but can calculate our subjective expectations/probabilites using Bayesian reasoning? I'm not quite sure this is your position, although there have been hints that it is and it is common among Bayesians, which is why I was asking questions about the aim of science and the relationship between Bayesian calculations and reality/truth.
And I suspect this is another point of disagreement - I don't think this is right. Having certain truth would be great - if we could be certain about something being true, we could infallibly classify it as true. Awesome! However, we have agreed that this is not possible. But say we can calculate something is probable using Bayesianism. This tells us exactly nothing about whether it is actually true or not, about whether reality really works that way. Certain truth implies truth, yes, but probable truth does not imply truth. Like I said, if you are mistaken about the truth of the belief in question, i.e. if reality does not work that way, your belief being probable does nothing to mitigate your mistake. You still believe something false. If moving faster than light is possible in reality, you are wrong in believing it is not, no matter how probable your belief may be. You are equally wrong as a Popperian who would just say he guessed wrong, despite your probability calculation. And if the belief in question is true, if reality is such that moving faster than light is not possible, this belief just as true for the Popperian who only claims it to be a guess and does not reference probabilty, regardless of whether your beleif is probable or not according to Bayesianism. I see absolutely no logical connection, i.e. possibilty of conversion, between probability and truth. Reality does not care about our probability calculations.
To sum up, I have no issues with Bayesianism as a theory of decision-making. I see science as guesswork tempered by reason and experience, aiming at objectively true theories about reality - although we can never know or be (partially or fully) justified in believing them to be true: we have to guess, and eliminate our errors by experiments, which are also guesswork. I think we may disagee with regard to truth and the aim of science, but I'm not totally clear on your position here. I also think we may disagree on the connection between probability and truth/external reality.
I certainly agree that Bayesianism can't solve the problem of regress or induction. But since nothing can, we have made absolutely no progress in solving it (despite many very intelligent people trying), my strong suspicion is that it's an impossible problem to solve, and so we might as well just make do.
Can you call everything a guess? Yes.
Yet guesses are certainly not made equal. We can be much more certain of some guesses than others, and while we can never get to perfect certainty in anything, we can often get close enough for government work, or even to establish extremely powerful laws of physics.
When a guess becomes so strong that we would be utterly flabbergasted if we suddenly saw evidence it was false, then we can dispense with the formalities and call it a fact, or true.
My stance is that this isn't a form of unwarranted cowardice or retreat, but a pragmatic acknowledgement that we are subjective observers forced to grapple with evidence acquired through sensoria. There is no scope for an actual bird's eye view, we must trudge through the territory and map it as we go.
Our probability theories breakdown utterly, or become garbled, when confronted with things like the possibility of the Simulation Hypothesis being true, Boltzmann brains and so on. I can't rule out with any degree of real confidence that this isn't a simulation, and the laws of physics as we best understand them imply Boltzmann brains are inevitable, let alone the implications if the Universe (and not just the observable part) is truly infinite, at which point every possible arrangement of matter and energy will be forced to tile over and repeat somewhere.
Or I could be crazy. I could be making mistakes.
We cannot eliminate this, and on top of that, we are computationally bounded entities.
Could this be a simulation? Could I be a Boltzmann brain about to vanish into a fizzle? Am I crazy? Regardless of my credence in any of them, I simply choose to operate as if it's not true, since that is the course of action with the highest payoff if they're not true, and if they are, nothing I do matters (we can't even usefully speculate as to what a potential simulation is for, or if the laws of physics at that level correspond to ours).
Thus, if objective truth exists, we can never know with perfect certainty (as always, without assuming the conclusion in advance), and science can only constrain our expectations and improve our ability to model things at the level of our observations of our environment.
Thus, is science useful? Absolutely. But it can never provide perfect certainty, but we can certainly operate without that, and in practise we do. We can converge ever closer to 100% confidence that our map faithfully represents the territory, and that's good enough for me.
And yet we are not Reality itself, or able to stand above it. Hence we should care about our probability calculations.
The closest we have to perfect certainty is in things like logic and math, but those do not actually exist in a vacuum to be poked and prodded, they're instantiated within our brains and other computers, and thus even what you (and I, with usual caveats that I omit when the discussion does not go this deep) might consider self evident truths, like 2=2 being true, could be the product of error in our cognition. Let's just be glad that we think this is extremely, extremely unlikely, and carry on with our lives nonetheless.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link