David Balan and I exchanged a few posts here over the last few weeks on paternalism, and we had a one hour debate a week ago (audio here). David and I are similarly expert by conventional measures, so this is more disagreeing with an equal than
Nick, it isn't clear to me how wide a range of "western" paternalism David embraces. Germans allow legal prostitution, for example. Is Germany "western", and are laws against prostitution part of the "western" set? I suspect I do place more weight on dead people than he does, and/or have different opinions about what they would believe about us.
Robin, do you think that Balan can claim to agree with the vast bulk of contemporary Western opinion (and extending quite far beyond that since many other societies might see the paternalism we have now in our society as doing on balance more good than bad - I'm not sure whether most contemporary Muslims, Chinese, Africans, etc. would think that mandatory schooling, seat belts, not selling alcohol to minors, and the rest of it are a net bad for Americans). Does your disagreement with him hinge on your (a) placing a much greater weight than he does on the opinions of people who are dead, and (b) assuming that those dead people would believe that present paternalism is on balance bad?
That reminds me of my justification for not being religious: the majority of people in the world are not Christian, the majority of people in the world are not Muslim, the majority of people in the world are not Hindu, the majority of people in the world are not Buddhist, etc...
So I can't pick any religion without being in a minority! I'm not sure the conclusion really follows though. Something I'm still working on.
Nick, the waste-cutting analogy clearly failed to be clear; I'll just drop it.
Each society's paternalism policies are usually supported by a majority there. And if societies tended to support the paternalism in other socities, then the majorities of all the societies would together be a very strong majority, which one should think twice or twenty times before opposing.
But, as I explained to Hal, the majority in our society also opposes most of the paternalisms in other societies now and in the past. If other societies also disagree with each other, then it is not clear there is a majority you can agree with. My guess, thought it is only a guess, is that the minorities in each society that oppose paternalism agree with each other more across societies; they'd like all the minorities everywhere to be left alone.
Robin, (a) many particular paternalistic policies enjoy majority support, yet you seem to oppose these policies; (b) even if no particular paternalistic policy enjoyed majority support, we can still look at the fact that most people believe it's better to accept the actual system of paternalistic policies, and the way it may be tweaked over time, than to eliminate it wholesale; whereas you seem to hold the opposite view.
I'm not here directly questioning those views, but I'm questioning your apparent claim in your post, where you seem to be justifying your disagreement with Balan by suggesting that you are agreeing with the majoriy while Balan's views are those of a minority.
I'm not sure how you mean to apply the waste-cutting example to the present case. In the analogy, is waste-cutting=paternalism? So most people are in favor of waste-cutting and in favor of paternalism, even though for most people there are also some specific instances of waste-cutting and some specific instances of paternalism that they oppose, although different instances for different people. If so, the analogy would be that Balan argues for paternalism and for waste-cutting, while you are arguing against paternalism and against waste-cutting? But are you really againt cutting waste in government? And would you really think you were in the majority in holding that view?
Btw, if we're going to make a disagreement case study of this debate between you and Balan, it would be useful also to hear Balan's side of the story.
Nick, I meant that while most people may support some paternalism, they don't necessarily support the same paternalism policies. A similar example would be that most people support cutting "waste" in government spending, as long as it isn't the "waste" they personally benefit from.
Hal, to use a simple average of opinion as the best estimate, you'd have to assume similar error rates. This is subject worth more thought.
Robin says "the fact that the vast majority support some paternalism seems like the fact that the vast majority would support an absolute dictator, if only they could pick the individual to be the dictator."
I don't think these alleged facts are alike. Almost all of the citizens support paternalism; almost all oppose dictatorship.
The vast majority would support having themselves thrown off a tall building, if only they could magically sprout wings at the crucial moment and fly off into the sunset. This does not imply that the vast majority support having themselves thrown off tall buildings.
Robin, that is a good point about mutually inconsistent paternalism. But suppose we accept the general principle that more minds are wiser than fewer. Then it still might be reasonable to support local paternalism even though each locality has different practices. As Eliezer describes, this should reduce mean squared error, right?
The ideal situation would then be a global worldwide democracy. No surprise there, I'm sure most people in our culture would agree with this goal, and many in other cultures as well. It is safe to assume that such a democracy would practice paternalism, since it is widely supported among the public. Would you support paternalism under such a government?
Hal, I think you're following your ideas off a cliff. The average person tries to do better-than-average, so if you only try to be as good as the average person, you'll do worse than average because you aren't striving as hard as the average person.
Hal, a problem is that these societies disagree with each other about their paternalisms. If each society believed that each other society was right in their paternalisms, then you could agree with them all at once. But when they each claim that other societies had it wrong, you can't agree with them all at once.
Okay, then I think I would have to support those kinds of paternalistic policies as well. Given the uncertainty that exists on the best way to live, and the pervasiveness of individual bias, if the majority of society strongly believes in certain practices then that would be good enough grounds in most cases to accept those beliefs as our best guide to what is correct. If everyone believes in following Sharia, Islamic law, then I should be very hesitant to put my skepticism above everyone else's opinion.
Again, this is based on the principle that in general, the majority is more likely to be right than the minority, and the individual. I am still mulling over Eliezer's comments to try to understand how mean squared error fits into this picture.
Matt C, your claim is of course true in many case; the question is which cases.
Richard, I am not a history expert, and so can't confirm your claim.
Nick, the fact that the vast majority support some paternalism seems like the fact that the vast majority would support an absolute dictator, if only they could pick the individual to be the dictator.
Hal, most paternalism in most societies has been supported by a substantial majority of that society. Consider rules against gays or divorce, against alternative religions, or rules in Islamic societies today. Consider racial slavery in the U.S., justified by saying blacks couldn't take care of themselves. Even rules limiting women were supported by enough women to get majority support.
Hal, to the extent that people have different biases randomly distributed around a mean, I would expect larger group sizes - and they could be quite small group sizes, on the order of 10,000 say - to reduce in variance to that mean. The mean can still be biased, that is, far from the correct answer. For example, people's beliefs about the Emperor of China's height are Gaussian-distributed around 220 centimeters with a standard deviation of 10 centimeters, but the correct answer is 200 centimeters. Now, for technical reasons, the expectation of squared error does decrease as you poll more and more Chinese, but the expectation doesn't go to zero. For example, a guess of 210 has a squared error of 100, while a guess of 230 has a squared error of 900, so if you have a 50/50 chance of making only one of these guesses - polling a single Chinese - your average squared error will be 900 + 100 / 2 = 500. However, if you poll both of them and average their guesses together to 220, the squared error will be 400, which is lower.
On the other hand, if you were working with expected error instead of expected squared error, no amount of averaging would make a difference. The standard formulation of "bias-variance decomposition" assumes squared error, and it lets you neatly partition the portion of squared error that comes from variance - individual differences - and bias - error that comes from the individual average still being wrong.
So, if you're squaring the errors that people make, then you're guaranteed to reduce the average squared error by polling more people. But if you're not squaring the errors, it doesn't make a difference.
For different biases to cancel out in the intuitive sense, not just the technical sense of squared error and bias-variance decomposition, the different biases would have to be on different sides of the correct answer.
Eliezer, to the extent that different people have different biases, wouldn't you expect them to cancel out as we move to larger group sizes? What do you think of the general principle that larger groups would on average be less biased than smaller ones?
Nick, it isn't clear to me how wide a range of "western" paternalism David embraces. Germans allow legal prostitution, for example. Is Germany "western", and are laws against prostitution part of the "western" set? I suspect I do place more weight on dead people than he does, and/or have different opinions about what they would believe about us.
Robin, do you think that Balan can claim to agree with the vast bulk of contemporary Western opinion (and extending quite far beyond that since many other societies might see the paternalism we have now in our society as doing on balance more good than bad - I'm not sure whether most contemporary Muslims, Chinese, Africans, etc. would think that mandatory schooling, seat belts, not selling alcohol to minors, and the rest of it are a net bad for Americans). Does your disagreement with him hinge on your (a) placing a much greater weight than he does on the opinions of people who are dead, and (b) assuming that those dead people would believe that present paternalism is on balance bad?
That reminds me of my justification for not being religious: the majority of people in the world are not Christian, the majority of people in the world are not Muslim, the majority of people in the world are not Hindu, the majority of people in the world are not Buddhist, etc...
So I can't pick any religion without being in a minority! I'm not sure the conclusion really follows though. Something I'm still working on.
Nick, the waste-cutting analogy clearly failed to be clear; I'll just drop it.
Each society's paternalism policies are usually supported by a majority there. And if societies tended to support the paternalism in other socities, then the majorities of all the societies would together be a very strong majority, which one should think twice or twenty times before opposing.
But, as I explained to Hal, the majority in our society also opposes most of the paternalisms in other societies now and in the past. If other societies also disagree with each other, then it is not clear there is a majority you can agree with. My guess, thought it is only a guess, is that the minorities in each society that oppose paternalism agree with each other more across societies; they'd like all the minorities everywhere to be left alone.
Btw, if we're going to make a disagreement case study of this debate between you and Balan, it would be useful also to hear Balan's side of the story.
Right here, Nick.
Robin, (a) many particular paternalistic policies enjoy majority support, yet you seem to oppose these policies; (b) even if no particular paternalistic policy enjoyed majority support, we can still look at the fact that most people believe it's better to accept the actual system of paternalistic policies, and the way it may be tweaked over time, than to eliminate it wholesale; whereas you seem to hold the opposite view.
I'm not here directly questioning those views, but I'm questioning your apparent claim in your post, where you seem to be justifying your disagreement with Balan by suggesting that you are agreeing with the majoriy while Balan's views are those of a minority.
I'm not sure how you mean to apply the waste-cutting example to the present case. In the analogy, is waste-cutting=paternalism? So most people are in favor of waste-cutting and in favor of paternalism, even though for most people there are also some specific instances of waste-cutting and some specific instances of paternalism that they oppose, although different instances for different people. If so, the analogy would be that Balan argues for paternalism and for waste-cutting, while you are arguing against paternalism and against waste-cutting? But are you really againt cutting waste in government? And would you really think you were in the majority in holding that view?
Btw, if we're going to make a disagreement case study of this debate between you and Balan, it would be useful also to hear Balan's side of the story.
Nick, I meant that while most people may support some paternalism, they don't necessarily support the same paternalism policies. A similar example would be that most people support cutting "waste" in government spending, as long as it isn't the "waste" they personally benefit from.
Hal, to use a simple average of opinion as the best estimate, you'd have to assume similar error rates. This is subject worth more thought.
Hal, not if the temptation to regulate is inherently biased in its execution - then you can reduce total error by reducing regulation.
(Statistical arguments shouldn't really ought to be invoked here. Why use squared error instead of error? It's the wrong grounds for the argument.)
Robin says "the fact that the vast majority support some paternalism seems like the fact that the vast majority would support an absolute dictator, if only they could pick the individual to be the dictator."
I don't think these alleged facts are alike. Almost all of the citizens support paternalism; almost all oppose dictatorship.
The vast majority would support having themselves thrown off a tall building, if only they could magically sprout wings at the crucial moment and fly off into the sunset. This does not imply that the vast majority support having themselves thrown off tall buildings.
Robin, that is a good point about mutually inconsistent paternalism. But suppose we accept the general principle that more minds are wiser than fewer. Then it still might be reasonable to support local paternalism even though each locality has different practices. As Eliezer describes, this should reduce mean squared error, right?
The ideal situation would then be a global worldwide democracy. No surprise there, I'm sure most people in our culture would agree with this goal, and many in other cultures as well. It is safe to assume that such a democracy would practice paternalism, since it is widely supported among the public. Would you support paternalism under such a government?
Hal, I think you're following your ideas off a cliff. The average person tries to do better-than-average, so if you only try to be as good as the average person, you'll do worse than average because you aren't striving as hard as the average person.
Hal, a problem is that these societies disagree with each other about their paternalisms. If each society believed that each other society was right in their paternalisms, then you could agree with them all at once. But when they each claim that other societies had it wrong, you can't agree with them all at once.
Okay, then I think I would have to support those kinds of paternalistic policies as well. Given the uncertainty that exists on the best way to live, and the pervasiveness of individual bias, if the majority of society strongly believes in certain practices then that would be good enough grounds in most cases to accept those beliefs as our best guide to what is correct. If everyone believes in following Sharia, Islamic law, then I should be very hesitant to put my skepticism above everyone else's opinion.
Again, this is based on the principle that in general, the majority is more likely to be right than the minority, and the individual. I am still mulling over Eliezer's comments to try to understand how mean squared error fits into this picture.
Matt C, your claim is of course true in many case; the question is which cases.
Richard, I am not a history expert, and so can't confirm your claim.
Nick, the fact that the vast majority support some paternalism seems like the fact that the vast majority would support an absolute dictator, if only they could pick the individual to be the dictator.
Hal, most paternalism in most societies has been supported by a substantial majority of that society. Consider rules against gays or divorce, against alternative religions, or rules in Islamic societies today. Consider racial slavery in the U.S., justified by saying blacks couldn't take care of themselves. Even rules limiting women were supported by enough women to get majority support.
Hal, to the extent that people have different biases randomly distributed around a mean, I would expect larger group sizes - and they could be quite small group sizes, on the order of 10,000 say - to reduce in variance to that mean. The mean can still be biased, that is, far from the correct answer. For example, people's beliefs about the Emperor of China's height are Gaussian-distributed around 220 centimeters with a standard deviation of 10 centimeters, but the correct answer is 200 centimeters. Now, for technical reasons, the expectation of squared error does decrease as you poll more and more Chinese, but the expectation doesn't go to zero. For example, a guess of 210 has a squared error of 100, while a guess of 230 has a squared error of 900, so if you have a 50/50 chance of making only one of these guesses - polling a single Chinese - your average squared error will be 900 + 100 / 2 = 500. However, if you poll both of them and average their guesses together to 220, the squared error will be 400, which is lower.
On the other hand, if you were working with expected error instead of expected squared error, no amount of averaging would make a difference. The standard formulation of "bias-variance decomposition" assumes squared error, and it lets you neatly partition the portion of squared error that comes from variance - individual differences - and bias - error that comes from the individual average still being wrong.
So, if you're squaring the errors that people make, then you're guaranteed to reduce the average squared error by polling more people. But if you're not squaring the errors, it doesn't make a difference.
For different biases to cancel out in the intuitive sense, not just the technical sense of squared error and bias-variance decomposition, the different biases would have to be on different sides of the correct answer.
Eliezer, to the extent that different people have different biases, wouldn't you expect them to cancel out as we move to larger group sizes? What do you think of the general principle that larger groups would on average be less biased than smaller ones?