One failure of logic is that you explicitly stated in your post that you already expected people to not follow this principle:
“I’m anticipating this post to be a straight shot to meta-irony: I have confidently made a non-normative claim, so expect a couple of negative post votes, absent of material feedback.”
You cannot then claim that using it was reasonable.
Furthermore: regardless of whether the comment was in fact a response to your request in line with your requested guidelines, your first action in this discussion was to publically punish the one person who you believed to be following the guidelines you requested. You clearly have not examined the incentives you are creating here.
Edit: For the meta-meta-irony, I will also state that I downvoted and disagree-voted your comment replying to thenoviceoof for these reasons.
I don’t think that really tracks — both can easily be true:
I can receive a couple of negative post votes, absent of material feedback
I can receive a comment, see that the post vote tally is at “-3”, and be presented with a Bayesian to evaluate: P (Comment was provided in accordance with my explicit request | I made an explicit request and post vote tally is at “-3″)
The public punishment (which, again, I explicitly moderate in the same comment — the selective reading of those arguing against me is quite astonishing) was intentional to try to illustrate how I would see my recommendation playing out:
Contrarian post is made and receives downvote, with Downvoter comment providing justification.
Contrarian should agree or disagree with the Downvoter, and cast their vote accordingly.
Other readers can cast their votes on the comments of the Contrarian and the Downvoter — here we have rationally operationalised truth-seeking by transparently surfacing a weakness of the post via the Downvoters comment, hearing the Contrarian’s yielding or defence, and being able to rate both sides.
Maybe to better operationalise this I should have clarified “Comment Guideline: If you downvote this post, please also add a Reaction or a 30+ character note prepended with “Downvote note:” on what to improve.” because I’ve left myself open to bad-faith critics to miss the point.
My karma is getting obliterated in faithfully trying to facilitate this discourse — do you think that is interesting?
The post is now at “-9” with 8 votes and it’s still unclear to me why — if nothing other than through misunderstanding the motivation behind my reply to @thenoviceoof and so to take retributive action against the post. I wrote 864 words that:
Suggested an opportunity to improve the LessWrong team’s model of supporting contrarian views with stability
Described mechanisms that suppress contrarian views
Described how a person confidently expressing non-normative ideas is easy to dismiss, despite this being a necessary condition of being a visionary free-thinker.
Nobody has engaged with these three points and the damage is done: my contrarian view is suppressed, my future posts and comments hold less weight, and I’m disillusioned by the capacity of folks to engage with contrarian viewpoints in good faith.
If I say “just kidding, I think the current system is perfect” — now I’m part of your in-group and you can revert your downvote?
I note that in your leading argument, you do not claim that you did do a Bayesian evaluation, nor even that someone could have done one and received a result that aligns with your conclusion. Just that it’s possible that someone could be presented with a probability to evaluate (which could have any result). That seems like a bad-faith evasion to me.
Did you actually evaluate the probability?
I have downvoted your comment but will revert it if you actually did do a Bayesian evaluation and present it in a follow-up comment, and it looks correct, and in line with your statement that “It is reasonable that I would conclude that”. That’s the norm you’re establishing in this comment chain, after all. A norm that, if it’s not obvious already, I think is both harmful to constructive discussion and wasteful of people’s time.
After that, then we can move on to the rest of the discussion with the results in mind.
Thank you for your densely rich reactions to my comment — I see validity in most of them.
Did you actually evaluate the probability?
Yes, evaluating the probability in my mind looked like:
Events:
R: I posted the explicit Request as the first line of the post: “Comment Guideline: If you downvote this post, please also add a Reaction or a 30+ character note”
CFP: A Comment was observed that Fulfilled Part of my requested format — it gave a reactionary note, and was on the order of characters I requested (~100 vs. my request of >=30 )
VN: The Vote tally was Negative on the order of a couple of votes
F: A comment was observed that Fulfilled all of my requested format (i.e CFP + it is a Downvoter providing a note) — this is what I evaluated to justify looking for a “Downvote-worthy” implication from their comment.
D: The commenter downvoted
The conditional probability I intuited:
P ( F=1 | R = 1, CFP = 1, VN = 1) = P ( D=1 | R = 1, CFP = 1, VN = 1) = x
My evaluation: I felt that x would be high, i.e >= 0.75 because of the intersection of R, CFP and VN
I admit that I “jumped the gun” by acting on this Bayesian instead of first asking for clarification, and then responding. I had a few motivations for this:
Receiving a comment satisfying my Comment Guideline (event F above) was necessary for me to be able to operationalise my idea: that is to say that I wanted to be able to agree or disagree and state that because of this I was going to “hold them accountable” and vote on their comment positively or negatively.
I was in the mood to write and express myself, and didn’t want to wait — indeed the first-commenter in question only provided clarification 16 hours later. In the meantime, my karma was obliterated.
Given (1), (2) and my high evaluated probability of F, I figured it was worth it on balance to push forwards.
I also see now that giving attention to spelling out why I was assuming that they were “implying that my post should be downvoted” — with the formalised Bayesian above — could have facilitated better rationalist discourse. My excuse would be that my attention was pulled in a few different directions, and I prioritised simply showcasing the operationalised version of my recommendation/ feature request.
On your disagreement to “Nobody has engaged with these points” I think I agree with you and could be more precise with my statement. I think nobody had, at least when I wrote that, engaged directly along the lines of reasoning of one of those points. However, engagement like “You may be interested in a very similar discussion” or “Sometimes, a post or comment seems so far from epistemic virtue as to be not worth spending effort describing all the problems. I mutter “not even wrong”, downvote, and move on.” does engage with those points at a meta-level, in terms of providing constructive feedback for the post.
Okay, so no actual Bayesian calculation, just an intuition.
Your post made the claim that it was substantially likely that most downvotes had no corresponding comment. If we look at this as a set of probabilities over readers R, it seems reasonable to model in terms of P(F_r), P(D_r), P(U_r), P(C_r), and P(CFP_r) for each reader r. C_r is the event of a reader providing a comment at all (whether or not it is a CFP comment).
Your expectation required P(D_r) > P(U_r), since you expected the post to be overall downvoted. This condition also implies that at some point VN holds. You also expected P(F_r | D_r) to be low, say < 0.3. If P(F_r | D_r) were higher, then you could not reasonably expect to see multiple downvotes with no corresponding explanatory comment.
Now let us examine P(CFP_r | C_r). Looking over the site, almost all comments are in some way reactionary to the thing they are commenting on, and all but a tiny minority are >= 30 characters. So P(CFP_r | C_r) > 0.8 is likely in the background and not just under condition F_r. Also looking at other posts, the number of votes seems to be on average about half the number of comments so P(C_r) ~= (1/2) (P(D_r) + P(U_r)).
Having made a specific request (that you did not expect to be followed), did you expect to see fewer comments as a fraction of votes overall, compared with other posts? You didn’t appear to think so, or it should have shown in your reasoning above. Likewise for P(CFP_r | C_r, R).
The condition VN is roughly the case D-U=2 (in this case I think it was exactly D=2, U=0), so your expectation E[C | VN, R] should have been around 1 to 3, and E[CFP | VN, R] about 1 to 2. You should also have expected E[F | VN, R] < 0.6.
So it seems to me to be quite a mistake to conclude P(F_r | CFP_r, R, VN) > 0.75.
That’s even without considering the nature of the comment itself, which made no criticism of your post at all and appeared to be more informative linking it to a previous discussion on the matter than anything else.
Sorry, I think you’re putting far too much weight on something that is not my position.
Your post made the claim that it was substantially likely that most downvotes had no corresponding comment.
My closing line, verbatim:
I’m anticipating this post to be a straight shot to meta-irony: I have confidently made a non-normative claim, so expect a couple of negative post votes, absent of material feedback.
If I thought this was true “it [is] substantially likely that most downvotes had no corresponding comment.”, that would look like me closing with:
I think I will get 10 negative post votes and no material feedback
I’m describing, explicitly in my post, a different phenomenon that contains more nuance: specifically low signal early votes that suppress visibility.
I say:
Sometimes I spend hours putting together a post that I’m proud of, but then receive no feedback besides a couple of downvotes.
and
a down-voting agent can effectively silence my voice just because they disagree with me.
I agree [people don’t owe me their time], but I feel that there is a distinct imbalance where a post can take hours of effort, and be cast aside with a 10-second vibe check and 1 second “downvote click”. I believe that the platform experience for both post authors and readers could be significantly improved by adding a second post-level signal that only takes an additional few seconds — this could be a React like “Difficult to Parse” or a ~30-character tip like “Same ideas posted recently: [link]”.
None of this looks like the claim: “it was substantially likely that most downvotes had no corresponding comment.”
If you want to adjust your calculation, you would need to account for my true position which is that VN (recall: my post having a couple of negative votes) is a prerequisite for a comment that is made to be from a Downvoter.
However, obviously it’s also a small sample size. That means it’s high variability, and we shouldn’t put much weight on it.
To simplify things, even though VN is a prerequisite in my view we can even drop it (due to it holding little weight), so we’re approximately evaluating:
P ( F=1 | R = 1, CFP = 1) = P ( D=1 | R = 1, CFP = 1) = x
Edit:
And implicit in this rebalancing (maybe) — adding this edit to clarify — is that I don’t agree with this:
Now let us examine P(CFP_r | C_r). Looking over the site, almost all comments are in some way reactionary to the thing they are commenting on, and all but a tiny minority are >= 30 characters. So P(CFP_r | C_r) > 0.8 is likely in the background and not just under condition F_r.
I absolutely did not feel that CFP was just background noise. I gave significant weight to the fact that the first line (R) explicitly requested comments of the form of CFP.
One failure of logic is that you explicitly stated in your post that you already expected people to not follow this principle:
“I’m anticipating this post to be a straight shot to meta-irony: I have confidently made a non-normative claim, so expect a couple of negative post votes, absent of material feedback.”
You cannot then claim that using it was reasonable.
Furthermore: regardless of whether the comment was in fact a response to your request in line with your requested guidelines, your first action in this discussion was to publically punish the one person who you believed to be following the guidelines you requested. You clearly have not examined the incentives you are creating here.
Edit: For the meta-meta-irony, I will also state that I downvoted and disagree-voted your comment replying to thenoviceoof for these reasons.
I don’t think that really tracks — both can easily be true:
I can receive a couple of negative post votes, absent of material feedback
I can receive a comment, see that the post vote tally is at “-3”, and be presented with a Bayesian to evaluate: P (Comment was provided in accordance with my explicit request | I made an explicit request and post vote tally is at “-3″)
The public punishment (which, again, I explicitly moderate in the same comment — the selective reading of those arguing against me is quite astonishing) was intentional to try to illustrate how I would see my recommendation playing out:
Contrarian post is made and receives downvote, with Downvoter comment providing justification.
Contrarian should agree or disagree with the Downvoter, and cast their vote accordingly.
Other readers can cast their votes on the comments of the Contrarian and the Downvoter — here we have rationally operationalised truth-seeking by transparently surfacing a weakness of the post via the Downvoters comment, hearing the Contrarian’s yielding or defence, and being able to rate both sides.
Maybe to better operationalise this I should have clarified “Comment Guideline: If you downvote this post, please also add a Reaction or a 30+ character note prepended with “Downvote note:” on what to improve.” because I’ve left myself open to bad-faith critics to miss the point.
My karma is getting obliterated in faithfully trying to facilitate this discourse — do you think that is interesting?
The post is now at “-9” with 8 votes and it’s still unclear to me why — if nothing other than through misunderstanding the motivation behind my reply to @thenoviceoof and so to take retributive action against the post. I wrote 864 words that:
Suggested an opportunity to improve the LessWrong team’s model of supporting contrarian views with stability
Described mechanisms that suppress contrarian views
Described how a person confidently expressing non-normative ideas is easy to dismiss, despite this being a necessary condition of being a visionary free-thinker.
Nobody has engaged with these three points and the damage is done: my contrarian view is suppressed, my future posts and comments hold less weight, and I’m disillusioned by the capacity of folks to engage with contrarian viewpoints in good faith.
If I say “just kidding, I think the current system is perfect” — now I’m part of your in-group and you can revert your downvote?
I note that in your leading argument, you do not claim that you did do a Bayesian evaluation, nor even that someone could have done one and received a result that aligns with your conclusion. Just that it’s possible that someone could be presented with a probability to evaluate (which could have any result). That seems like a bad-faith evasion to me.
Did you actually evaluate the probability?
I have downvoted your comment but will revert it if you actually did do a Bayesian evaluation and present it in a follow-up comment, and it looks correct, and in line with your statement that “It is reasonable that I would conclude that”. That’s the norm you’re establishing in this comment chain, after all. A norm that, if it’s not obvious already, I think is both harmful to constructive discussion and wasteful of people’s time.
After that, then we can move on to the rest of the discussion with the results in mind.
Thank you for your densely rich reactions to my comment — I see validity in most of them.
Yes, evaluating the probability in my mind looked like:
Events:
R: I posted the explicit Request as the first line of the post: “Comment Guideline: If you downvote this post, please also add a Reaction or a 30+ character note”
CFP: A Comment was observed that Fulfilled Part of my requested format — it gave a reactionary note, and was on the order of characters I requested (~100 vs. my request of >=30 )
VN: The Vote tally was Negative on the order of a couple of votes
F: A comment was observed that Fulfilled all of my requested format (i.e CFP + it is a Downvoter providing a note) — this is what I evaluated to justify looking for a “Downvote-worthy” implication from their comment.
D: The commenter downvoted
The conditional probability I intuited:
P ( F=1 | R = 1, CFP = 1, VN = 1) = P ( D=1 | R = 1, CFP = 1, VN = 1) = x
My evaluation: I felt that x would be high, i.e >= 0.75 because of the intersection of R, CFP and VN
I admit that I “jumped the gun” by acting on this Bayesian instead of first asking for clarification, and then responding. I had a few motivations for this:
Receiving a comment satisfying my Comment Guideline (event F above) was necessary for me to be able to operationalise my idea: that is to say that I wanted to be able to agree or disagree and state that because of this I was going to “hold them accountable” and vote on their comment positively or negatively.
I was in the mood to write and express myself, and didn’t want to wait — indeed the first-commenter in question only provided clarification 16 hours later. In the meantime, my karma was obliterated.
Given (1), (2) and my high evaluated probability of F, I figured it was worth it on balance to push forwards.
I also see now that giving attention to spelling out why I was assuming that they were “implying that my post should be downvoted” — with the formalised Bayesian above — could have facilitated better rationalist discourse. My excuse would be that my attention was pulled in a few different directions, and I prioritised simply showcasing the operationalised version of my recommendation/ feature request.
On your disagreement to “Nobody has engaged with these points” I think I agree with you and could be more precise with my statement. I think nobody had, at least when I wrote that, engaged directly along the lines of reasoning of one of those points. However, engagement like “You may be interested in a very similar discussion” or “Sometimes, a post or comment seems so far from epistemic virtue as to be not worth spending effort describing all the problems. I mutter “not even wrong”, downvote, and move on.” does engage with those points at a meta-level, in terms of providing constructive feedback for the post.
Okay, so no actual Bayesian calculation, just an intuition.
Your post made the claim that it was substantially likely that most downvotes had no corresponding comment. If we look at this as a set of probabilities over readers R, it seems reasonable to model in terms of P(F_r), P(D_r), P(U_r), P(C_r), and P(CFP_r) for each reader r. C_r is the event of a reader providing a comment at all (whether or not it is a CFP comment).
Your expectation required P(D_r) > P(U_r), since you expected the post to be overall downvoted. This condition also implies that at some point VN holds. You also expected P(F_r | D_r) to be low, say < 0.3. If P(F_r | D_r) were higher, then you could not reasonably expect to see multiple downvotes with no corresponding explanatory comment.
Now let us examine P(CFP_r | C_r). Looking over the site, almost all comments are in some way reactionary to the thing they are commenting on, and all but a tiny minority are >= 30 characters. So P(CFP_r | C_r) > 0.8 is likely in the background and not just under condition F_r. Also looking at other posts, the number of votes seems to be on average about half the number of comments so P(C_r) ~= (1/2) (P(D_r) + P(U_r)).
Having made a specific request (that you did not expect to be followed), did you expect to see fewer comments as a fraction of votes overall, compared with other posts? You didn’t appear to think so, or it should have shown in your reasoning above. Likewise for P(CFP_r | C_r, R).
The condition VN is roughly the case D-U=2 (in this case I think it was exactly D=2, U=0), so your expectation E[C | VN, R] should have been around 1 to 3, and E[CFP | VN, R] about 1 to 2. You should also have expected E[F | VN, R] < 0.6.
So it seems to me to be quite a mistake to conclude P(F_r | CFP_r, R, VN) > 0.75.
That’s even without considering the nature of the comment itself, which made no criticism of your post at all and appeared to be more informative linking it to a previous discussion on the matter than anything else.
Sorry, I think you’re putting far too much weight on something that is not my position.
My closing line, verbatim:
If I thought this was true “it [is] substantially likely that most downvotes had no corresponding comment.”, that would look like me closing with:
I’m describing, explicitly in my post, a different phenomenon that contains more nuance: specifically low signal early votes that suppress visibility.
I say:
and
and in my discussion with @Drake Morrison :
None of this looks like the claim: “it was substantially likely that most downvotes had no corresponding comment.”
If you want to adjust your calculation, you would need to account for my true position which is that VN (recall: my post having a couple of negative votes) is a prerequisite for a comment that is made to be from a Downvoter.
However, obviously it’s also a small sample size. That means it’s high variability, and we shouldn’t put much weight on it.
To simplify things, even though VN is a prerequisite in my view we can even drop it (due to it holding little weight), so we’re approximately evaluating:
P ( F=1 | R = 1, CFP = 1) = P ( D=1 | R = 1, CFP = 1) = x
Edit:
And implicit in this rebalancing (maybe) — adding this edit to clarify — is that I don’t agree with this:
I absolutely did not feel that CFP was just background noise. I gave significant weight to the fact that the first line (R) explicitly requested comments of the form of CFP.