My main answer is “you get more signal out of the system if people assume votes have a ‘correct’ amount.”
Say we _don’t_ have strong upvote / small upvotes. There’s just a single upvote.
You see two comments: one pretty good, one great.
Say there are 100 clones of yourself using LW.
Would you rather both comments get 100 karma? Or the better comment to end up with more?
The only options here are
all 100 clones upvote both equally
all 100 clones only upvote the ‘great’ comment
all 100 clones run some sort of percentage-chance-to-upvote on the ‘pretty-good’ comment
all 100 clones upvote the pretty-good-comment if it has less than 35 karma.
The 3rd and 4th option seem roughly equivalent to me.
Variable-size-votes do enable the possibility of a finer-grained result with everyone voting purely based on their individual assessments, but if everyone voted for a comment exactly based on how good they thought it was, and you _add_ those votes up, then the karma ends up being more a function of “how many people saw the comment” than “how good it is.”
(Wei_Dai’s comments on ‘cost of voting’ and ‘preventing rich-getting-richer’ also play into it)
If you want an idealized system wherein the expectation is “everyone votes their true appraisal of a comment’s value”, then the way I’d implement it would be:
1. Comment karma is hidden until you vote on it (so you don’t anchor as much)
2. The karma displayed isn’t the total, but something more like an average (or a function that’s closer to an average than a total).
(I think this might actually be a good system, although it’s a pretty big change)
I also think that this is likely to be a pretty good system. (I can the outlines of some downsides, but would have to think further before I could describe them satisfactorily; in any case, it does not seem to me, at first glance, that any downsides of this system would exceed those of the current system. But this is not yet a strongly held opinion.)
Are these the only users, or is it me and my 100 clones and also some other users? I’m going to assume the latter, because in the former case the entire karma system is just pointless; correct me if my assumption is mistaken.
In that case, I strongly prefer the 3rd option. (The 2nd is an acceptable fallback position if the 3rd is unavailable.) The 3rd and 4th options do not seem at all equivalent to me; it seems strange to suggest otherwise.
My assumption here is that the outcome of options #3 and #4 are fairly similar. If Great Comment is 3x as good as Good comment, then option #3 looks something like “33% chance of upvoting” and option #4 is “upvote if it’s less than about 33 karma. Possibly downvote if it’s got so much karma that it’s getting sorted higher than the Great Comment.”
How similar the outcomes are depends on some of the precise numbers, but I think the order-of-magnitude* is about the same.
*where by order-of-magnitude I mean “base 4 order of magnitude”, which is the comparison that usually seems most relevant to me.
Option #4 is not “thread-safe”. It therefore can, and very likely will, cause chaotic behavior with unpredictable and potentially quite undesirable attractors.
Edit: Note that we may already observe this taking place.
Doublechecking what you mean by “thread-safe” (multiple people looking at the same thing at the same time making decisions before they see what other people do?)
Cool. That doesn’t seem like that big an issue to me, because the system has built-in error correction – people come back later, see that it’s slid past whichever direction they thought it should go, and can change their vote. It more robustly converges on Good Comment getting a proportionally correct-ish score (whereas the “roll a die in your head, upvote 33% of the time” will some non-trivial portion of the time result in things getting way-the-hell upvoted (or not) that should have been).
I should note: I don’t all think this is necessarily the best approach, just, it’s an approach that seems “reasonable” enough that describing it as ‘defeating the point of the voting system’ doesn’t seem accurate.
The core of the problem remains: it requires users to know what other users are doing (as well as how many other users there are, and how many other users are paying attention to a comment, and other such things). The cognitive overhead is tremendously higher. The potential for error is (thus) also much higher.
This is shocking to me, and I would absolutely love to hear more about this perspective.
My main answer is “you get more signal out of the system if people assume votes have a ‘correct’ amount.”
Say we _don’t_ have strong upvote / small upvotes. There’s just a single upvote.
You see two comments: one pretty good, one great.
Say there are 100 clones of yourself using LW.
Would you rather both comments get 100 karma? Or the better comment to end up with more?
The only options here are
all 100 clones upvote both equally
all 100 clones only upvote the ‘great’ comment
all 100 clones run some sort of percentage-chance-to-upvote on the ‘pretty-good’ comment
all 100 clones upvote the pretty-good-comment if it has less than 35 karma.
The 3rd and 4th option seem roughly equivalent to me.
Variable-size-votes do enable the possibility of a finer-grained result with everyone voting purely based on their individual assessments, but if everyone voted for a comment exactly based on how good they thought it was, and you _add_ those votes up, then the karma ends up being more a function of “how many people saw the comment” than “how good it is.”
(Wei_Dai’s comments on ‘cost of voting’ and ‘preventing rich-getting-richer’ also play into it)
If you want an idealized system wherein the expectation is “everyone votes their true appraisal of a comment’s value”, then the way I’d implement it would be:
1. Comment karma is hidden until you vote on it (so you don’t anchor as much)
2. The karma displayed isn’t the total, but something more like an average (or a function that’s closer to an average than a total).
(I think this might actually be a good system, although it’s a pretty big change)
I also think that this is likely to be a pretty good system. (I can the outlines of some downsides, but would have to think further before I could describe them satisfactorily; in any case, it does not seem to me, at first glance, that any downsides of this system would exceed those of the current system. But this is not yet a strongly held opinion.)
Are these the only users, or is it me and my 100 clones and also some other users? I’m going to assume the latter, because in the former case the entire karma system is just pointless; correct me if my assumption is mistaken.
In that case, I strongly prefer the 3rd option. (The 2nd is an acceptable fallback position if the 3rd is unavailable.) The 3rd and 4th options do not seem at all equivalent to me; it seems strange to suggest otherwise.
My assumption here is that the outcome of options #3 and #4 are fairly similar. If Great Comment is 3x as good as Good comment, then option #3 looks something like “33% chance of upvoting” and option #4 is “upvote if it’s less than about 33 karma. Possibly downvote if it’s got so much karma that it’s getting sorted higher than the Great Comment.”
How similar the outcomes are depends on some of the precise numbers, but I think the order-of-magnitude* is about the same.
*where by order-of-magnitude I mean “base 4 order of magnitude”, which is the comparison that usually seems most relevant to me.
Option #4 is not “thread-safe”. It therefore can, and very likely will, cause chaotic behavior with unpredictable and potentially quite undesirable attractors.
Edit: Note that we may already observe this taking place.
Doublechecking what you mean by “thread-safe” (multiple people looking at the same thing at the same time making decisions before they see what other people do?)
Right.
Cool. That doesn’t seem like that big an issue to me, because the system has built-in error correction – people come back later, see that it’s slid past whichever direction they thought it should go, and can change their vote. It more robustly converges on Good Comment getting a proportionally correct-ish score (whereas the “roll a die in your head, upvote 33% of the time” will some non-trivial portion of the time result in things getting way-the-hell upvoted (or not) that should have been).
I should note: I don’t all think this is necessarily the best approach, just, it’s an approach that seems “reasonable” enough that describing it as ‘defeating the point of the voting system’ doesn’t seem accurate.
The core of the problem remains: it requires users to know what other users are doing (as well as how many other users there are, and how many other users are paying attention to a comment, and other such things). The cognitive overhead is tremendously higher. The potential for error is (thus) also much higher.