Nice! I'll be sure to keep that debate in my prayers! You're definitely much smarter than he is, but his rhetorical skill is always a challenge to deal with (I have no idea why "claims aren't evidence" is such a compelling epistemological principle to so many of Dillahunty's fans, but it is, lol).
Yup. Interestingly, there’s another reason some people think we shouldn’t introduce probabilities in the court room. Suppose someone dies from a bus accident and we know it’s because the brakes on the bus were faulty and negligently not checked on consistently by the company. We dont know which bus in the city is the one that killed the person, but we know there are only 2 bus companies in the city: company A and company B. Suppose 52% of busses are owned by company A. If this is all we know, then it follows a jury should find company A liable and have them pay millions of dollars. Even though there is no “particularized” evidence against company A. Some say there is something unjust about using probabilities this way in a court of justice. If they’re right, all the more reason to separate norms of courts and law from our conversations.
Whether that is or isn’t is separate from whether or not Company A meets the legal “preponderance of the evidence” standard. The argument is supposed to show that that standard can’t be defined simply as “50+% probability”.
Separately, though, I should remark that your approach would seem to recommend that in cases where assign a 95% chance to S killing someone, and 5% chance to S’ doing it, we should split the sentence 95-5.
Good article. I think I feel similarly about the idea that claims have to be "falsifiable"—I don't believe you have an invisible, immaterial dragon in your garage because my priors on invisible, immaterial dragons' existence are very low, not because you haven't done an experiment to prove its existence! If you find some way to convince me that the dragon is actually very simple and intuitive, I don't get to just plug my ears and yell, "Unfalsifiable! Unfalsifiable!"
Yeap. Burden of Proof is almost always a way to express one’s priors (without admitting to them) as well. It’s a tool to make one’s priors seem superior to the priors of another. It hides one’s priors behind an epistemological fog in order to claim supremacy. After all, to claim an asserted point shoulders the burden of proof is itself an implicit prior!
Another day, another Bentham post that I disagree with! Let me try to answer this one.
First, I think there is a lot more to our epistemic stances than a numeric probability/credence, or even a range of them, so by the time we've gone bayesian, a lot of information has already been lost. The bayesian formalization also has the practical problem that we're psychologically very bad at reasoning with numbers beyond a few orders of magnitude; there's not much left of Bayes if we can't psychologically tell the difference between holding a prior of 10^-11 or 10^-22. I've previously said that some of your arguments for theism verge into this territory.
But even ignoring all of that: our systems of beliefs are large interconnected tapestries of views, often held with fragmentary evidence, and faith that someone else has figured it out in more detail, and so on. So when you put two people with opposing beliefs in front of each other, and give them a single piece of evidence from the world to update on, not much happens. Each will make a slight tweak to their system, and they will remain in unexplained opposition.
This is unsatisfactory, so people came up with the idea of debate. You use the emotional energy of finding the other's beliefs incredibly weird and totally unacceptable, to get the two people to play a structured game where they rub their priors and their metapriors against each other, gasping and grunting^W^W^W asserting tentative beliefs, challenging them, going into details and rabbit holes as needed to really challenge their systems.
A debate is a repeated interaction, and out of repeated interactions you get emergent structures. In the case of debate, you quickly find the need to have rules for what it takes to consider something proved - that's what the idea of the burden of proof comes from.
That's also why there are continuing disagreements about how the burden of proof works. The rules of debate are not logically proven themselves. If debate is the game, then figuring out the most productive rules of debate is the meta-game.
So for example, a simple rule can be that the person who makes an assertion has the burden of proof. But then you run into the problem that you can't prove a negative. "There are no flying pigs" -> "You haven't checked every pig". The rule is not 'wrong' in itself, it's just not very productive.
So you can then try a different rule, where existential assertions have the burden of proof, but not universal ones. "There are flying pigs" -> "Show me one". "There are no flying pigs" -> "I accept because I can't show you one".
Either way, you still need a working rule for some kind of burden of proof, otherwise we don't have an agreed protocol for what happens after one says "there are no flying pigs", and each can just sit there thinking the other is weirdly wrong.
I guess this is the steel version of the exchange, though:
MATERIALIST: In this debate, the burden of proof is on you, in the sense that, because you're the one defending Theism, it's your responsibility to actually present the evidence. Everyone else has already reached a well-calibrated Bayesian probability on the subject [lol], so don't expect them to be as motivated as you are to do the philosophical work of defending Theism. Maybe you see something they don't—it's YOUR burden to show it!
THEIST: But you don't get to just assume Atheism in the meantime!
MATERIALIST: Correct. I just assume Materialism or Physicalism.
THEIST: That's just as bad!
MATERIALIST: Well, then I guess I would say I have a pragmatic attachment to Materialism because of a whole bunch of convergent knowledge from many different scientific fields, which has offered a picture of the world that is described very well by the assumption of Materialism/Physicalism; not to mention the predictive power and the miracle of technology. So I'm pragmatically assuming the theory that makes the fewest assumptions.
THEIST: Pff, go try to reconcile mind with matter and tell me Physicalism makes the fewest assumptions!
MATERIALIST: Fine, but like I said, I believe in it because it's useful to and the whole wheel of science keeps spinning when you believe in it. Either way, it's still your responsibility to make the case for Theism because you've chosen to defend it in this debate.
I agree that how burden of proof is typically used is dumb, but that’s also a bit of a strawman. It seems to me that reasonable arguments about burden of proof are often really just arguments over what our priors should be, using different language
In law, we distinguish between the burden of production and the burden of persuasion. The burden of production determines who has the obligation to present facts on a particular matter. The burden of persuasion determines how strong the evidence must be before the fact is established.
The burden of production is the flip side of a presumption. All people are presumed innocent, which means that in the absence of any evidence, a court will provisionally assume they are innocent. This is not an epistemological claim. It is more of a pragmatic claim in accordance with our notions of justice. In civil cases, the plaintiff has the burden of production on each element of his claim, and the defendant has the burden of production on any affirmative defenses, This largely correlates with pragmatic issues and who is more likely to have the evidence. But not completely, (Which is why we have discovery.)
The burden of persuasion determines how compelling the evidence must be to establish a claim. In civil cases, it is preponderance of the evidence, which simply means more likely than not. Some claims require "clear and convincing" evidence. The definition is other vague adjectives, but it generally means a lot more than 50%. For example, in California, a civil plaintiff can get punitive damages for some claims if the plaintiff acted with "malice, fraud. or oppression." Those have to be shown by clean and convincing evidence. And in criminal cases, each element of the crime must be proven beyond a reasonable doubt. Again, the precise definition consists of other adjectives, but it basically means the jury or judge must be really really certain.
The does not fit precisely into a baysian framework, but it is close. But law is pragmatic, and there are issues of justice at work as well. For example, balancing betweel Type I and Type II errors. It is better to convict n guilty person than one innocent one. (BTW, my friend Sasha Volokh has a humorous article where he analyzed the different values of n over time an in different states, complete with a tongue-in-cheek regression analysis with a microscopic r^2 value, some practical advice for criminals, and a discussion of Bentham as an "n-skeptic":
These sorts of rules are designed for pragmatic reasons, not a search for ultimate truth or reframing legal issues in a baysian framework. But the distinction between the burdens of production and persuasion is often helpful in non-legal contexts.
I think you've got confused in semantics. The concept of burden of proof is basically layman's way to talk about bayesian reasoning.
Let's take the example with a commited crime. Suppose no evidence is provided but someone priviledges the hypothesis that you in particular have commited the crime.
We can talk about it in terms of "burden of proof". Say that it lies on the accuser. And until they presented some evidence in favor of you commiting the crime we are not treating it as priviledged hypothesis.
or
We can talk about in terms of bayesian updates, that to move from equiprobable prior we need to receive some evidence that makes you commiting the crime likelier than an average person, and until this evidence is received we can't lawfully update like this.
Both ways are talking about the same thing.
Likewise, when DNA evidence is presented then the burden of proof is moved to someone who claims that we still not supposed to update in favor of you being the main suspect. Maybe the DNA test was counterfeight? Maybe there is a very good reason for your DNA to in the crime scene, unrelated to you being a murderer? These possibilities are themselves not enough for us not to update (or to update back), only when we receive an actual counterevidene that one of them is in fact the case (or much more likely the case, then we initially though), we stop priviledging the hypothesis that you are the murderer, or even decrease our confidence in it. And so on and so forth.
Essentially, the notion of burden of proof is a blunt tool which is able to talk about things being more or less likely than average but without specifying how much, while being fluent bayesian gives you more preceise way to do the same thing, using real numbers from [0,1] for credences. There is no conflict here, and if you find that your probability theoretic reasoning contradict the common sense notion of burden of proof, you might have made some mistake along the way.
If you make a claim but don’t give any reasoning, then you’re not giving me any information I can use to update my beliefs.
(Well, other than the information that you believe in that claim – which in some contexts might be enough to greatly increase my credence in the claim, but in the context of your typical internet debate, not so much.)
If I were a mathematically rational Bayesian agent, I would have no need for arguments; I would have already evaluated every possible argument for every possible claim, and all that would remain is concrete evidence. But as an actual person with finite computational ability, I have to obtain an argument before I can evaluate it. Either you have to tell it to me, or I have to speculate and try to come up with it on my own, and the latter path tends to be quite unproductive.
In that sense there is indeed an epistemic problem. If a statement provides no mechanism for anyone to acquire knowledge, then it’s epistemically useless.
> Me: It’s not my job to educate you. Do the work. Don’t force self-indication assumption understanders to do the emotional labor of teaching you, which is often traumatic and triggering for POSIAA (persons of self-indication assumption understanding).
Ok, but if SIA's strongest soldier said this in an argument, surely one's Bayesian credence in SIA would fall significantly. I feel like there's a justified expectation that someone making a claim can provide evidence, and if they don't, that increases the probability that they're full of hot air.
I think I largely disagree or at least think you may be strawmanning the case for burden of proof.
The way that this is important is a few things:
Bayesianism is pretty hard to do in practice and simplicity over worlds as a prior is an especially difficult part of this process, so we should have heuristics (especially for lay-people and most non specialists) that are more accommodating.
Burden of proof (most times in practice) is basically saying that if you posit another thing without good evidence that should be something much less probable than something that does have good evidence.
On the other hand, I do think this gets into trouble with things that have high probability on priors, but at least pragmatically, I think this is just worth losing and have normal people use burden of proof (to some degree because people have some notion of that exception as well).
You could say that just everyone should be Bayesian (and I’d agree), but this is just not practically plausible.
The way that the question of burden of proof should be framed, in my opinion, is if you could snap your fingers and make everyone not believe in burden of proof would you (all else equal)? I think the obvious answer here is to not do it as burden of proof is a good way of approximating bayes than not having it on the margin.
Citing the law of non-contradiction is a bit of a cop out. Saying you’re unconvinced that X is true is different from confidently asserting that X isn’t true. Whether you’re asserting X is true or false, it falls on you to make that argument rather than me to argue why X is uncertain.
Bro failed to meet the burden of proof in this post. A lot of claims about debate norms, but no hard peer-reviewed scientific evidence. So cringe
Matt Dillahunty would be very annoyed if he read this post. Which, of course, is always a good sign when it comes to questions of epistemology.
Speaking of Matt, wasn’t there going to be a debate between them or did I just make that up?
Yep still happening - plan is doing in on the 12th
Awesome! Are you going to be debating God's existence with him, or a different topic?
God’s existence.
Nice! I'll be sure to keep that debate in my prayers! You're definitely much smarter than he is, but his rhetorical skill is always a challenge to deal with (I have no idea why "claims aren't evidence" is such a compelling epistemological principle to so many of Dillahunty's fans, but it is, lol).
Yup. Interestingly, there’s another reason some people think we shouldn’t introduce probabilities in the court room. Suppose someone dies from a bus accident and we know it’s because the brakes on the bus were faulty and negligently not checked on consistently by the company. We dont know which bus in the city is the one that killed the person, but we know there are only 2 bus companies in the city: company A and company B. Suppose 52% of busses are owned by company A. If this is all we know, then it follows a jury should find company A liable and have them pay millions of dollars. Even though there is no “particularized” evidence against company A. Some say there is something unjust about using probabilities this way in a court of justice. If they’re right, all the more reason to separate norms of courts and law from our conversations.
In your hypothetical bus example, isn't the better solution to fine company A 52% of the total, and company B 48%?
Whether that is or isn’t is separate from whether or not Company A meets the legal “preponderance of the evidence” standard. The argument is supposed to show that that standard can’t be defined simply as “50+% probability”.
Separately, though, I should remark that your approach would seem to recommend that in cases where assign a 95% chance to S killing someone, and 5% chance to S’ doing it, we should split the sentence 95-5.
Good article. I think I feel similarly about the idea that claims have to be "falsifiable"—I don't believe you have an invisible, immaterial dragon in your garage because my priors on invisible, immaterial dragons' existence are very low, not because you haven't done an experiment to prove its existence! If you find some way to convince me that the dragon is actually very simple and intuitive, I don't get to just plug my ears and yell, "Unfalsifiable! Unfalsifiable!"
Yeap. Burden of Proof is almost always a way to express one’s priors (without admitting to them) as well. It’s a tool to make one’s priors seem superior to the priors of another. It hides one’s priors behind an epistemological fog in order to claim supremacy. After all, to claim an asserted point shoulders the burden of proof is itself an implicit prior!
Another day, another Bentham post that I disagree with! Let me try to answer this one.
First, I think there is a lot more to our epistemic stances than a numeric probability/credence, or even a range of them, so by the time we've gone bayesian, a lot of information has already been lost. The bayesian formalization also has the practical problem that we're psychologically very bad at reasoning with numbers beyond a few orders of magnitude; there's not much left of Bayes if we can't psychologically tell the difference between holding a prior of 10^-11 or 10^-22. I've previously said that some of your arguments for theism verge into this territory.
But even ignoring all of that: our systems of beliefs are large interconnected tapestries of views, often held with fragmentary evidence, and faith that someone else has figured it out in more detail, and so on. So when you put two people with opposing beliefs in front of each other, and give them a single piece of evidence from the world to update on, not much happens. Each will make a slight tweak to their system, and they will remain in unexplained opposition.
This is unsatisfactory, so people came up with the idea of debate. You use the emotional energy of finding the other's beliefs incredibly weird and totally unacceptable, to get the two people to play a structured game where they rub their priors and their metapriors against each other, gasping and grunting^W^W^W asserting tentative beliefs, challenging them, going into details and rabbit holes as needed to really challenge their systems.
A debate is a repeated interaction, and out of repeated interactions you get emergent structures. In the case of debate, you quickly find the need to have rules for what it takes to consider something proved - that's what the idea of the burden of proof comes from.
That's also why there are continuing disagreements about how the burden of proof works. The rules of debate are not logically proven themselves. If debate is the game, then figuring out the most productive rules of debate is the meta-game.
So for example, a simple rule can be that the person who makes an assertion has the burden of proof. But then you run into the problem that you can't prove a negative. "There are no flying pigs" -> "You haven't checked every pig". The rule is not 'wrong' in itself, it's just not very productive.
So you can then try a different rule, where existential assertions have the burden of proof, but not universal ones. "There are flying pigs" -> "Show me one". "There are no flying pigs" -> "I accept because I can't show you one".
Either way, you still need a working rule for some kind of burden of proof, otherwise we don't have an agreed protocol for what happens after one says "there are no flying pigs", and each can just sit there thinking the other is weirdly wrong.
Good post
I never thought I'd see the day when you'd like one of my blog posts.
W take. Debates about the burden of proof drive me up the wall.
I guess this is the steel version of the exchange, though:
MATERIALIST: In this debate, the burden of proof is on you, in the sense that, because you're the one defending Theism, it's your responsibility to actually present the evidence. Everyone else has already reached a well-calibrated Bayesian probability on the subject [lol], so don't expect them to be as motivated as you are to do the philosophical work of defending Theism. Maybe you see something they don't—it's YOUR burden to show it!
THEIST: But you don't get to just assume Atheism in the meantime!
MATERIALIST: Correct. I just assume Materialism or Physicalism.
THEIST: That's just as bad!
MATERIALIST: Well, then I guess I would say I have a pragmatic attachment to Materialism because of a whole bunch of convergent knowledge from many different scientific fields, which has offered a picture of the world that is described very well by the assumption of Materialism/Physicalism; not to mention the predictive power and the miracle of technology. So I'm pragmatically assuming the theory that makes the fewest assumptions.
THEIST: Pff, go try to reconcile mind with matter and tell me Physicalism makes the fewest assumptions!
MATERIALIST: Fine, but like I said, I believe in it because it's useful to and the whole wheel of science keeps spinning when you believe in it. Either way, it's still your responsibility to make the case for Theism because you've chosen to defend it in this debate.
I agree that how burden of proof is typically used is dumb, but that’s also a bit of a strawman. It seems to me that reasonable arguments about burden of proof are often really just arguments over what our priors should be, using different language
In law, we distinguish between the burden of production and the burden of persuasion. The burden of production determines who has the obligation to present facts on a particular matter. The burden of persuasion determines how strong the evidence must be before the fact is established.
The burden of production is the flip side of a presumption. All people are presumed innocent, which means that in the absence of any evidence, a court will provisionally assume they are innocent. This is not an epistemological claim. It is more of a pragmatic claim in accordance with our notions of justice. In civil cases, the plaintiff has the burden of production on each element of his claim, and the defendant has the burden of production on any affirmative defenses, This largely correlates with pragmatic issues and who is more likely to have the evidence. But not completely, (Which is why we have discovery.)
The burden of persuasion determines how compelling the evidence must be to establish a claim. In civil cases, it is preponderance of the evidence, which simply means more likely than not. Some claims require "clear and convincing" evidence. The definition is other vague adjectives, but it generally means a lot more than 50%. For example, in California, a civil plaintiff can get punitive damages for some claims if the plaintiff acted with "malice, fraud. or oppression." Those have to be shown by clean and convincing evidence. And in criminal cases, each element of the crime must be proven beyond a reasonable doubt. Again, the precise definition consists of other adjectives, but it basically means the jury or judge must be really really certain.
The does not fit precisely into a baysian framework, but it is close. But law is pragmatic, and there are issues of justice at work as well. For example, balancing betweel Type I and Type II errors. It is better to convict n guilty person than one innocent one. (BTW, my friend Sasha Volokh has a humorous article where he analyzed the different values of n over time an in different states, complete with a tongue-in-cheek regression analysis with a microscopic r^2 value, some practical advice for criminals, and a discussion of Bentham as an "n-skeptic":
https://scholarship.law.upenn.edu/penn_law_review/vol146/iss1/4/ )
Also, sometimes the burden of production shifts.
https://en.wikipedia.org/wiki/Res_ipsa_loquitur
These sorts of rules are designed for pragmatic reasons, not a search for ultimate truth or reframing legal issues in a baysian framework. But the distinction between the burdens of production and persuasion is often helpful in non-legal contexts.
I think you've got confused in semantics. The concept of burden of proof is basically layman's way to talk about bayesian reasoning.
Let's take the example with a commited crime. Suppose no evidence is provided but someone priviledges the hypothesis that you in particular have commited the crime.
We can talk about it in terms of "burden of proof". Say that it lies on the accuser. And until they presented some evidence in favor of you commiting the crime we are not treating it as priviledged hypothesis.
or
We can talk about in terms of bayesian updates, that to move from equiprobable prior we need to receive some evidence that makes you commiting the crime likelier than an average person, and until this evidence is received we can't lawfully update like this.
Both ways are talking about the same thing.
Likewise, when DNA evidence is presented then the burden of proof is moved to someone who claims that we still not supposed to update in favor of you being the main suspect. Maybe the DNA test was counterfeight? Maybe there is a very good reason for your DNA to in the crime scene, unrelated to you being a murderer? These possibilities are themselves not enough for us not to update (or to update back), only when we receive an actual counterevidene that one of them is in fact the case (or much more likely the case, then we initially though), we stop priviledging the hypothesis that you are the murderer, or even decrease our confidence in it. And so on and so forth.
Essentially, the notion of burden of proof is a blunt tool which is able to talk about things being more or less likely than average but without specifying how much, while being fluent bayesian gives you more preceise way to do the same thing, using real numbers from [0,1] for credences. There is no conflict here, and if you find that your probability theoretic reasoning contradict the common sense notion of burden of proof, you might have made some mistake along the way.
If you make a claim but don’t give any reasoning, then you’re not giving me any information I can use to update my beliefs.
(Well, other than the information that you believe in that claim – which in some contexts might be enough to greatly increase my credence in the claim, but in the context of your typical internet debate, not so much.)
If I were a mathematically rational Bayesian agent, I would have no need for arguments; I would have already evaluated every possible argument for every possible claim, and all that would remain is concrete evidence. But as an actual person with finite computational ability, I have to obtain an argument before I can evaluate it. Either you have to tell it to me, or I have to speculate and try to come up with it on my own, and the latter path tends to be quite unproductive.
In that sense there is indeed an epistemic problem. If a statement provides no mechanism for anyone to acquire knowledge, then it’s epistemically useless.
> Me: It’s not my job to educate you. Do the work. Don’t force self-indication assumption understanders to do the emotional labor of teaching you, which is often traumatic and triggering for POSIAA (persons of self-indication assumption understanding).
Ok, but if SIA's strongest soldier said this in an argument, surely one's Bayesian credence in SIA would fall significantly. I feel like there's a justified expectation that someone making a claim can provide evidence, and if they don't, that increases the probability that they're full of hot air.
I think I largely disagree or at least think you may be strawmanning the case for burden of proof.
The way that this is important is a few things:
Bayesianism is pretty hard to do in practice and simplicity over worlds as a prior is an especially difficult part of this process, so we should have heuristics (especially for lay-people and most non specialists) that are more accommodating.
Burden of proof (most times in practice) is basically saying that if you posit another thing without good evidence that should be something much less probable than something that does have good evidence.
On the other hand, I do think this gets into trouble with things that have high probability on priors, but at least pragmatically, I think this is just worth losing and have normal people use burden of proof (to some degree because people have some notion of that exception as well).
You could say that just everyone should be Bayesian (and I’d agree), but this is just not practically plausible.
The way that the question of burden of proof should be framed, in my opinion, is if you could snap your fingers and make everyone not believe in burden of proof would you (all else equal)? I think the obvious answer here is to not do it as burden of proof is a good way of approximating bayes than not having it on the margin.
Citing the law of non-contradiction is a bit of a cop out. Saying you’re unconvinced that X is true is different from confidently asserting that X isn’t true. Whether you’re asserting X is true or false, it falls on you to make that argument rather than me to argue why X is uncertain.