There’s a whole bunch of information out there—literally more than any one person could/cares to know—and we simply don’t have the time (or often the background) to fully understand certain fields and more importantly to evaluate which claims are true and which aren’t.
In other words, reality is objective and claims should be evaluated based on their evidence, not the person who proposes them.
It would seem to me that these claims aren’t consistent. I agree with the first claim, not with the second. It’s true that experts’ claims are objectively and directly verifiable, but lots of the time checking that direct evidence is not an optimal use of our time. Instead we’re better off deferring to experts (which we actually also do, as you say, on a massive scale).
that bit about the 50% chance of being true is ridiculous even if you don’t have any knowledge going into it—you would simply say “I don’t know if these claims are true”
Well according to the betting interpretation of degrees of belief, this just means that you would, if rational, be willing to accept bets that are based on the claim in question having a 50 % chance of being true (but not bets based on the claim that it has, say, a 51 % chance of being true). But sure, sometimes it can seem a bit contrived to assign a definite probability to claims you know little about.
I guess the main thing I am trying to say that directly ties into your post is that we shouldn’t really care how someone formed their beliefs when evaluating the veracity of a claim; when we should care is:
I don’t agree with that. We use others’ statements as a source of evidence on a massive scale (i.e. we defer to them. Indeed, experiments show that we do this automatically. But if these statements express beliefs that were produced by unreliable processes—e.g. bias—then that’s clearly not a good strategy. Hence we should care very much of whether someone is biased when evaluating the veracity of many claims, for that reason.
Also, as I said, if we find out that someone is biased, then we have little reason to use that person as a source of knowledge.
What I want to stress is the need for cognitive economy. We don’t have time to check the direct evidence for different claims lots of the time (as you yourself admit above) and therefore have to use assessments of others’ reliability. Knowledge about bias is a vital (but not the only) ingredient in our assessments of reliability, and are hence extremely useful.
I’m making a separate reply for the betting thing, only to try to keep the two conversations clean/simple.
Let’s muddle through it:
If I have a box containing an unknown (to you) number of gumballs and I claim that there are an odd number of gumballs, you would actually be quite reasonable in assigning a 50% chance to my claim being true.
If I claim that the gumballs in the box are blue, would you say there is a 50% chance of my claim being true?
What if I claimed that I ate pizza last night?
You might have a certain level of confidence in my accuracy and my reliability as a person to not lie to you; and, if someone was taking bets, you would probably bet on how likely I am to tell the truth, rather than assuming there was a 50% chance that I ate pizza last night.
If you you then notice that my friend, who was with me last night, claims that I in fact ate pasta, then you have to weigh their reliability against mine, and more importantly now you have to start looking for reasons that we came to different conclusions about the same dinner. And finally, you have to weigh the effort it takes to vet our claims against how much you really care what I ate last night.
So, assuming you are rational, would you bet 50⁄50 that I ate pizza? Or would you just say “I don’t know” and refuse to bet in the first place?
This is a bit of a side-track. For the Bayesian interpretation of probability, it’s important to be able to assign a prior probability to any event (since otherwise you can’t calculate the posterior probability, given some piece of evidence that makes the event more or less probable). They do this using, e.g. the much contested principle of indifference. Some people object to this, and argue along your lines that it’s just silly to ascribe probabilities to events we know nothing about. Indeed, the frequentists define an event’s probability as the limit of its relative frequency in a large number of trials. Hence, to them, we can’t ascribe a probability to a one-off event at all.
Hence there is a huge discussion on this already and I don’t think that it’s meaningful for us to address it here. Anyway, you do have a point that one should be a bit cautious ascribing definite probabilities to events we know very little about. An alternative can be to say that the probability is somewhere in the interval from x to y, where x and y are some real numbers betwen 0 and 1.
I agree that it is largely off-topic and don’t feel like discussing it further here—I would like to point out that the principle of indifference specifies that your list of possibilities must be mutually exclusive and exhaustive. In practice, when dealing with multifaceted things such as claims about the effects of changing the minimum wage, an exhaustive list of possible outcomes would result in an assignment of an arbitrarily small probability according to the principle of indifference. The end effect is that it’s a meaningless assignment and you may as well ignore it.
There’s a whole bunch of information out there—literally more than any one person could/cares to know—and we simply don’t have the time (or often the background) to fully understand certain fields and more importantly to evaluate which claims are true and which aren’t.
In other words, reality is objective and claims should be evaluated based on their evidence, not the person who proposes them.
It would seem to me that these claims aren’t consistent. I agree with the first claim, not with the second. It’s true that experts’ claims are objectively and directly verifiable, but lots of the time checking that direct evidence is not an optimal use of our time. Instead we’re better off deferring to experts (which we actually also do, as you say, on a massive scale).
I think we are in agreement but my second statement didn’t have the caveats it should have; I doubt you would disagree with the first half, that reality is objective. You disagreed with the second half, that claims should be evaluated based on evidence—not because it’s a false statement, but rather that, in practice, we cannot reasonably be expected to do this for every claim we encounter. I agree. The unstated caveat is that we should trust the experts until there is a reason to think that their claims are poorly founded, i.e. they have demonstrated bias in their work or there is a lack of consensus among experts in a similar field.
I guess the main thing I am trying to say that directly ties into your post is that we shouldn’t really care how someone formed their beliefs when evaluating the veracity of a claim; when we should care is:
I don’t agree with that. We use others’ statements as a source of evidence on a massive scale (i.e. we defer to them. Indeed, experiments show that we do this automatically. But if these statements express beliefs that were produced by unreliable processes—e.g. bias—then that’s clearly not a good strategy. Hence we should care very much of whether someone is biased when evaluating the veracity of many claims, for that reason.
Hold on now, you did read my bullets right?
When we should care is:
When we suspect that a bias may have lead to a false reporting of real information (in which case we would want independent, unbiased research/reporting)
Notice that I actually did say suspicion of bias is an exception to the “not caring” statement. In other words, unless we have a reason to suspect a bias, (and/or the second bullet) then we probably won’t care. There can be other ways of bad conclusions being drawn; the reason I mention bias is because it is systematic. If we see a trend of a particular person systematically coming to poor conclusions, whatever their reason, then our confidence in their input would fall. On the other hand, experts are human and can make mistakes as well—we should not dismiss someone for being wrong once but for being systematically wrong and unwilling to fix the problem. If we really care about high confidence in something, for instance in the cases where the truth of the claim is important to a lot of people and we want to avoid being mislead if there are a few biased opinions, we seek the consensus.
Now, can we get the consensus all of the time? Unfortunately not. Not even most of the time. So what’s our next line of defense? Well, one of them is journalistic integrity; frankly I don’t even want to go there, but if done properly there are people whose job it is to sort through these very things—but really let’s not go there for now. The last line of defense is yourself and the actual work of checking on things yourself.
If a claim is important enough for you to really care whether or not it’s accurate, then you have to be willing to do a little bit of digging yourself. Now I realize that the entire point of this post was to avoid just that thing and to have computers do it automagically; but really, if it is important enough for you to check on it yourself, rather than just trusting your regular sources of information, then would you be willing not to check just because a program said that this guy was unbiased?
That might be a bit of an unfair characterization of what you’re discussing, but there is a distinction to be made between using online behavior to measure/understand the general population’s belief structure and to check for bias in expert opinions.
I think the idea of understanding the population’s belief structures would still be extremely useful in it’s own right though, per my second bullet in the exceptions to the “don’t care” statement—particularly if someone wants to change a lot of people’s minds about something. If you have a campaign (be it political or social), then understanding how people have structured their beliefs would give you a road map for how best to go about changing them in the way you want. To some extent, this is how it’s already been done historically, but it was not done via raw data analysis.
I feel that this discussion is getting a bit too multifarious, which no doubt has to do with the very abstract nature of my post. I’m not very happy with it. I should probably have started with more comprehensive and clear examples than an abstract and general discussion like this. Anyway, I do intend to give more examples of reverse-engineering-of-belief-structures-examples in the future. Hopefully that’ll make it clearer what I’m trying to do. Here’s one example of reverse engineering-reasoning I’ve already given.
I agree that lots of the time we should “do a bit of digging ourselves”; i.e. look at the direct evidence for P rather than on whether those telling us P or not-P are reliable or not. But I also claim that in many cases deference is extremely cost-efficient and useful. You seem to agree with this—good.
...but there is a distinction to be made between using online behavior to measure/understand the general population’s belief structure and to check for bias in expert opinions.
Sure. But reverse engineering reasoning can also be used to infer expert bias (as shown in this post).
To some extent, this is how it’s already been done historically, but it was not done via raw data analysis.
Yes. People already perform this kind of reverse engineering reasoning, as I said (cf my reference to Marx). What I want to do is to do it more systematically and efficiently.
It would seem to me that these claims aren’t consistent. I agree with the first claim, not with the second. It’s true that experts’ claims are objectively and directly verifiable, but lots of the time checking that direct evidence is not an optimal use of our time. Instead we’re better off deferring to experts (which we actually also do, as you say, on a massive scale).
I wrote a very long post on a related theme—“genetic arguments”—some time ago, by the way.
Well according to the betting interpretation of degrees of belief, this just means that you would, if rational, be willing to accept bets that are based on the claim in question having a 50 % chance of being true (but not bets based on the claim that it has, say, a 51 % chance of being true). But sure, sometimes it can seem a bit contrived to assign a definite probability to claims you know little about.
I don’t agree with that. We use others’ statements as a source of evidence on a massive scale (i.e. we defer to them. Indeed, experiments show that we do this automatically. But if these statements express beliefs that were produced by unreliable processes—e.g. bias—then that’s clearly not a good strategy. Hence we should care very much of whether someone is biased when evaluating the veracity of many claims, for that reason.
Also, as I said, if we find out that someone is biased, then we have little reason to use that person as a source of knowledge.
What I want to stress is the need for cognitive economy. We don’t have time to check the direct evidence for different claims lots of the time (as you yourself admit above) and therefore have to use assessments of others’ reliability. Knowledge about bias is a vital (but not the only) ingredient in our assessments of reliability, and are hence extremely useful.
I’m making a separate reply for the betting thing, only to try to keep the two conversations clean/simple.
Let’s muddle through it: If I have a box containing an unknown (to you) number of gumballs and I claim that there are an odd number of gumballs, you would actually be quite reasonable in assigning a 50% chance to my claim being true.
If I claim that the gumballs in the box are blue, would you say there is a 50% chance of my claim being true?
What if I claimed that I ate pizza last night?
You might have a certain level of confidence in my accuracy and my reliability as a person to not lie to you; and, if someone was taking bets, you would probably bet on how likely I am to tell the truth, rather than assuming there was a 50% chance that I ate pizza last night.
If you you then notice that my friend, who was with me last night, claims that I in fact ate pasta, then you have to weigh their reliability against mine, and more importantly now you have to start looking for reasons that we came to different conclusions about the same dinner. And finally, you have to weigh the effort it takes to vet our claims against how much you really care what I ate last night.
So, assuming you are rational, would you bet 50⁄50 that I ate pizza? Or would you just say “I don’t know” and refuse to bet in the first place?
This is a bit of a side-track. For the Bayesian interpretation of probability, it’s important to be able to assign a prior probability to any event (since otherwise you can’t calculate the posterior probability, given some piece of evidence that makes the event more or less probable). They do this using, e.g. the much contested principle of indifference. Some people object to this, and argue along your lines that it’s just silly to ascribe probabilities to events we know nothing about. Indeed, the frequentists define an event’s probability as the limit of its relative frequency in a large number of trials. Hence, to them, we can’t ascribe a probability to a one-off event at all.
Hence there is a huge discussion on this already and I don’t think that it’s meaningful for us to address it here. Anyway, you do have a point that one should be a bit cautious ascribing definite probabilities to events we know very little about. An alternative can be to say that the probability is somewhere in the interval from x to y, where x and y are some real numbers betwen 0 and 1.
I agree that it is largely off-topic and don’t feel like discussing it further here—I would like to point out that the principle of indifference specifies that your list of possibilities must be mutually exclusive and exhaustive. In practice, when dealing with multifaceted things such as claims about the effects of changing the minimum wage, an exhaustive list of possible outcomes would result in an assignment of an arbitrarily small probability according to the principle of indifference. The end effect is that it’s a meaningless assignment and you may as well ignore it.
I think we are in agreement but my second statement didn’t have the caveats it should have; I doubt you would disagree with the first half, that reality is objective. You disagreed with the second half, that claims should be evaluated based on evidence—not because it’s a false statement, but rather that, in practice, we cannot reasonably be expected to do this for every claim we encounter. I agree. The unstated caveat is that we should trust the experts until there is a reason to think that their claims are poorly founded, i.e. they have demonstrated bias in their work or there is a lack of consensus among experts in a similar field.
Hold on now, you did read my bullets right? When we should care is:
When we suspect that a bias may have lead to a false reporting of real information (in which case we would want independent, unbiased research/reporting)
Notice that I actually did say suspicion of bias is an exception to the “not caring” statement. In other words, unless we have a reason to suspect a bias, (and/or the second bullet) then we probably won’t care. There can be other ways of bad conclusions being drawn; the reason I mention bias is because it is systematic. If we see a trend of a particular person systematically coming to poor conclusions, whatever their reason, then our confidence in their input would fall. On the other hand, experts are human and can make mistakes as well—we should not dismiss someone for being wrong once but for being systematically wrong and unwilling to fix the problem. If we really care about high confidence in something, for instance in the cases where the truth of the claim is important to a lot of people and we want to avoid being mislead if there are a few biased opinions, we seek the consensus.
Now, can we get the consensus all of the time? Unfortunately not. Not even most of the time. So what’s our next line of defense? Well, one of them is journalistic integrity; frankly I don’t even want to go there, but if done properly there are people whose job it is to sort through these very things—but really let’s not go there for now. The last line of defense is yourself and the actual work of checking on things yourself.
If a claim is important enough for you to really care whether or not it’s accurate, then you have to be willing to do a little bit of digging yourself. Now I realize that the entire point of this post was to avoid just that thing and to have computers do it automagically; but really, if it is important enough for you to check on it yourself, rather than just trusting your regular sources of information, then would you be willing not to check just because a program said that this guy was unbiased?
That might be a bit of an unfair characterization of what you’re discussing, but there is a distinction to be made between using online behavior to measure/understand the general population’s belief structure and to check for bias in expert opinions.
I think the idea of understanding the population’s belief structures would still be extremely useful in it’s own right though, per my second bullet in the exceptions to the “don’t care” statement—particularly if someone wants to change a lot of people’s minds about something. If you have a campaign (be it political or social), then understanding how people have structured their beliefs would give you a road map for how best to go about changing them in the way you want. To some extent, this is how it’s already been done historically, but it was not done via raw data analysis.
I feel that this discussion is getting a bit too multifarious, which no doubt has to do with the very abstract nature of my post. I’m not very happy with it. I should probably have started with more comprehensive and clear examples than an abstract and general discussion like this. Anyway, I do intend to give more examples of reverse-engineering-of-belief-structures-examples in the future. Hopefully that’ll make it clearer what I’m trying to do. Here’s one example of reverse engineering-reasoning I’ve already given.
I agree that lots of the time we should “do a bit of digging ourselves”; i.e. look at the direct evidence for P rather than on whether those telling us P or not-P are reliable or not. But I also claim that in many cases deference is extremely cost-efficient and useful. You seem to agree with this—good.
Sure. But reverse engineering reasoning can also be used to infer expert bias (as shown in this post).
Yes. People already perform this kind of reverse engineering reasoning, as I said (cf my reference to Marx). What I want to do is to do it more systematically and efficiently.