accumulated 30 points of karma from what seems to me to be… unimpressive as presented?
I upvoted on the value of the comment as additional source data (IIRC when the comment had much lower karma). This value shouldn’t be diminished by questionable interpretation/attitude bundled with it, since the interpretation can be discarded, but the data can’t be magicked up.
This is a general consideration that applies to communications that provoke a much stronger urge to mute them, for example those that defend detestable positions. If such communications bring you new relevant data, even data that doesn’t significantly change your understanding of the situation, they are still precious, the effects of processing them and not ignoring them sum up over all such instances. (I think the comment to this post most rich in relevant data is prevlev-anon’s, which I strong-upvoted.)
This makes sense to me in my first pass of thinking about it, and I agree.
There’s something subtle and extremely hard to pull off (perhaps impossible) in: “in the wishing world, what do we think a shared voting policy should be, such that the aggregate of everyone voting consistently according to that policy leaves all comments in approximately the same order that a single extremely perceptive and high-quality reasoner would rank them?”
As opposed to comments just trending toward infinities.
This works out for the earlier top level comments (that see similar voter turnout), the absolute numbers just scale with popularity of the post. If something is not in its place in your ideal ranking, it’s possible to use the vote to move it that way. Vote weights do a little bit to try improving the quality (or value lock-in) of the ranking.
One issue with the system is the zero equilibrium on controversial things, with the last voters randomly winning irrespective of the actual distribution of opinion. It’s unclear how to get something more informative for such situations, but this should be kept in mind as a use case for any reform.
I’m trying to apply the ITT to your position, and I’m pretty sure I’m failing (and for the avoidance of doubt I believe that you are generally very well informed, capable and are here engaging in good faith, so I anticipate that the failing is mine, not yours). I hope that you can help me better understand your position:
My background assumptions (not stated or endorsed by you):
Conditional on a contribution (a post, a comment) being all of (a) subject to a reasonably clear interpretation (for the reader alone, if that is the only value the reader is optimising for, or otherwise for some (weighted?) significant portion of the reader community), (b) with content that is relevant and important to a question that the reader considers important (most usually the question under discussion), and (c) that is substantially true, and it is evident that it is true from the content as it is presented (for the reader alone, or the reader community), then…
My agreement with the value that I think you’re chasing:
… I agree that there is at least an important value at stake here, and the reader upvoting a contribution that meets those conditions may serve that important value.
Further elaboration of my background assumptions:
If (a) (clear interpretation) is missing, then the reader won’t know there’s value there to reward, or must (should?) at least balance the harms that I think are clear from the reader or others misinterpreting the data offered.
If (b) (content is relevant) is missing, then… perhaps you like rewarding random facts? I didn’t eat breakfast this morning. This is clear and true, but I really don’t expect to be rewarded for sharing it.
If (c) (evident truth) is missing, then either (not evident) you don’t know whether to reward the contribution or not, or (not true) surely the value is negative?
My statement of my confusion:
Now, you didn’t state these three conditions, so you obviously get to reject my claim of their importance… yet I’ve pretty roundly convinced myself that they’re important, and that (absent some very clever but probably nit-picky edge case, which I’ve been around Lesswrong long enough to know is quite likely to show up) you’re likely to agree (other readers should note just how wildly I’m inferring here, and if Vladimir_Nesov doesn’t respond please don’t assume that they actually implied any of this). You also report that you upvoted orthonormal’s comment (I infer orthonormal’s comment instead of RyanCarey’s, because you quoted “30 points of karma”, which didn’t apply to RyanCarey’s comment). So I’m trying to work out what interpretation you took from orthonormal’s comment (and the clearest interpretation I managed to find is the one I detailed in my earlier comment: that orthonormal based their opinion overwhelmingly on their first impression and didn’t update on subsequent data), whether you think the comment shared relevant data (did you think orthonormal’s first impression was valuable data pertaining to whether Leverage and Geoff were bad? did you think the data relevant to some other valuable thing you were tracking, that might not be what other readers would take from the comment?), and whether you think that orthonormal’s data was self-evidently true (do you have other reason to believe that orthonormal’s first impressions are spectacular? did you see some other flaw in the reasoning I my earlier comment?)
So, I’m confused. What were you rewarding with your upvote? Were you rewarding (orthonormal’s) behaviour, that you expect will be useful to you but misleading for others, or rewarding behaviour that you expect would be useful on balance to your comment’s readers (if so, what and how)? If my model is just so wildly wrong that none of these questions make sense to answer, can you help me understand where I fell over?
(To the inevitable commenter who would, absent this addition, jump in and tell me that I clearly don’t know what an ITT is: I know that what I have written here is not what it looks like to try to pass an ITT — I did try, internally, to see whether I could convince myself that I could pass Vladimir_Nesov’s ITT, and it was clear to me that I could not. This is me identifying where I failed — highlighting my confusion — not trying to show you what I did.)
There is an important class of claims detailed enough to either be largely accurate or intentional lies, their distortion can’t be achieved with mere lack of understanding or motivated cognition. These can be found even in very strange places, and still be informative when taken out of context.
The claim I see here is that orthonormal used a test for dicey character with reasonable precision. The described collateral damage of just one positive reading signals that it doesn’t trigger all the time, and there was at least one solid true positive. The wording also vaguely suggests that there aren’t too many other positive readings, in which case the precision is even higher than the collateral damage signals.
Since base rate is lower than the implied precision, a positive reading works as evidence. For the opposite claim, that someone has an OK character, evidence of this form can’t have similar strength, since the base rate is already high and there is no room for precision to get significantly higher.
It’s still not strong evidence, and directly it’s only about character in the sense of low-level intuitive and emotional inclinations. This is in turn only weak evidence of actual behavior, since people often live their lives “out of character”, it’s the deliberative reasoning that matters for who someone actually is as a person. Internal urges are only a risk factor and a psychological inconvenience for someone who disagrees with their own urges and can’t or won’t retrain them, it’s not an important defining characteristic and not relevant in most contexts. This must even be purposefully disregarded in some contexts to prevent discrimination.
Edit: I managed to fumble terminology in the original version of this comment and said “specificity” instead of “precision” or “positive predictive value”, which is what I actually meant. It’s true that specificity of the test is also not low (much higher even), and for basically the same reasons, but high specificity doesn’t make a positive reading positive evidence.
I upvoted on the value of the comment as additional source data (IIRC when the comment had much lower karma). This value shouldn’t be diminished by questionable interpretation/attitude bundled with it, since the interpretation can be discarded, but the data can’t be magicked up.
This is a general consideration that applies to communications that provoke a much stronger urge to mute them, for example those that defend detestable positions. If such communications bring you new relevant data, even data that doesn’t significantly change your understanding of the situation, they are still precious, the effects of processing them and not ignoring them sum up over all such instances. (I think the comment to this post most rich in relevant data is prevlev-anon’s, which I strong-upvoted.)
This makes sense to me in my first pass of thinking about it, and I agree.
There’s something subtle and extremely hard to pull off (perhaps impossible) in: “in the wishing world, what do we think a shared voting policy should be, such that the aggregate of everyone voting consistently according to that policy leaves all comments in approximately the same order that a single extremely perceptive and high-quality reasoner would rank them?”
As opposed to comments just trending toward infinities.
This works out for the earlier top level comments (that see similar voter turnout), the absolute numbers just scale with popularity of the post. If something is not in its place in your ideal ranking, it’s possible to use the vote to move it that way. Vote weights do a little bit to try improving the quality (or value lock-in) of the ranking.
One issue with the system is the zero equilibrium on controversial things, with the last voters randomly winning irrespective of the actual distribution of opinion. It’s unclear how to get something more informative for such situations, but this should be kept in mind as a use case for any reform.
I’m trying to apply the ITT to your position, and I’m pretty sure I’m failing (and for the avoidance of doubt I believe that you are generally very well informed, capable and are here engaging in good faith, so I anticipate that the failing is mine, not yours). I hope that you can help me better understand your position:
My background assumptions (not stated or endorsed by you):
Conditional on a contribution (a post, a comment) being all of (a) subject to a reasonably clear interpretation (for the reader alone, if that is the only value the reader is optimising for, or otherwise for some (weighted?) significant portion of the reader community), (b) with content that is relevant and important to a question that the reader considers important (most usually the question under discussion), and (c) that is substantially true, and it is evident that it is true from the content as it is presented (for the reader alone, or the reader community), then…
My agreement with the value that I think you’re chasing:
… I agree that there is at least an important value at stake here, and the reader upvoting a contribution that meets those conditions may serve that important value.
Further elaboration of my background assumptions:
If (a) (clear interpretation) is missing, then the reader won’t know there’s value there to reward, or must (should?) at least balance the harms that I think are clear from the reader or others misinterpreting the data offered.
If (b) (content is relevant) is missing, then… perhaps you like rewarding random facts? I didn’t eat breakfast this morning. This is clear and true, but I really don’t expect to be rewarded for sharing it.
If (c) (evident truth) is missing, then either (not evident) you don’t know whether to reward the contribution or not, or (not true) surely the value is negative?
My statement of my confusion:
Now, you didn’t state these three conditions, so you obviously get to reject my claim of their importance… yet I’ve pretty roundly convinced myself that they’re important, and that (absent some very clever but probably nit-picky edge case, which I’ve been around Lesswrong long enough to know is quite likely to show up) you’re likely to agree (other readers should note just how wildly I’m inferring here, and if Vladimir_Nesov doesn’t respond please don’t assume that they actually implied any of this). You also report that you upvoted orthonormal’s comment (I infer orthonormal’s comment instead of RyanCarey’s, because you quoted “30 points of karma”, which didn’t apply to RyanCarey’s comment). So I’m trying to work out what interpretation you took from orthonormal’s comment (and the clearest interpretation I managed to find is the one I detailed in my earlier comment: that orthonormal based their opinion overwhelmingly on their first impression and didn’t update on subsequent data), whether you think the comment shared relevant data (did you think orthonormal’s first impression was valuable data pertaining to whether Leverage and Geoff were bad? did you think the data relevant to some other valuable thing you were tracking, that might not be what other readers would take from the comment?), and whether you think that orthonormal’s data was self-evidently true (do you have other reason to believe that orthonormal’s first impressions are spectacular? did you see some other flaw in the reasoning I my earlier comment?)
So, I’m confused. What were you rewarding with your upvote? Were you rewarding (orthonormal’s) behaviour, that you expect will be useful to you but misleading for others, or rewarding behaviour that you expect would be useful on balance to your comment’s readers (if so, what and how)? If my model is just so wildly wrong that none of these questions make sense to answer, can you help me understand where I fell over?
(To the inevitable commenter who would, absent this addition, jump in and tell me that I clearly don’t know what an ITT is: I know that what I have written here is not what it looks like to try to pass an ITT — I did try, internally, to see whether I could convince myself that I could pass Vladimir_Nesov’s ITT, and it was clear to me that I could not. This is me identifying where I failed — highlighting my confusion — not trying to show you what I did.)
Edit 6hrs after posting: formatting only (I keep expecting Github Flavoured Markdown, instead of vanilla Markdown).
There is an important class of claims detailed enough to either be largely accurate or intentional lies, their distortion can’t be achieved with mere lack of understanding or motivated cognition. These can be found even in very strange places, and still be informative when taken out of context.
The claim I see here is that orthonormal used a test for dicey character with reasonable precision. The described collateral damage of just one positive reading signals that it doesn’t trigger all the time, and there was at least one solid true positive. The wording also vaguely suggests that there aren’t too many other positive readings, in which case the precision is even higher than the collateral damage signals.
Since base rate is lower than the implied precision, a positive reading works as evidence. For the opposite claim, that someone has an OK character, evidence of this form can’t have similar strength, since the base rate is already high and there is no room for precision to get significantly higher.
It’s still not strong evidence, and directly it’s only about character in the sense of low-level intuitive and emotional inclinations. This is in turn only weak evidence of actual behavior, since people often live their lives “out of character”, it’s the deliberative reasoning that matters for who someone actually is as a person. Internal urges are only a risk factor and a psychological inconvenience for someone who disagrees with their own urges and can’t or won’t retrain them, it’s not an important defining characteristic and not relevant in most contexts. This must even be purposefully disregarded in some contexts to prevent discrimination.
Edit: I managed to fumble terminology in the original version of this comment and said “specificity” instead of “precision” or “positive predictive value”, which is what I actually meant. It’s true that specificity of the test is also not low (much higher even), and for basically the same reasons, but high specificity doesn’t make a positive reading positive evidence.