I actually meant it as a compliment, thanks for pointing out that it can be received differently. I liked this “quick take” and believe it would have been a high-quality post.
I was not aware that my comment would reduce the number of quick takes and posts, but I feel deleting my comment now just because of the downvotes would also be weird. So, if anyone reads this and felt discouraged by the above, I hope you rather post your things somewhere rather than not at all.
Who was behind the backstop funds in the previous years? Do you know why they would not offer the backstop funds this year?
Meta: this should not have been a quick take, but a post (references, structure, tldr, epistemic status, …)
But absolutely, and yet a big part of EAs seem to be pro-Altman!
What makes you think a big part of EAs are pro-Altman? My impressions is that this is not true, and I cannot come up with any concrete example.
why would you like there to be less people “motivated in part by an impartial care for others, [are] thinking very carefully about how they can best help others [...]”?
edit: please ignore, just saw that titotal asked the same question 10 minutes earlier.
this sounds confused: people who do interpretability already know the model weights (more like they try to interpret the model weights). They are not learning the model weights from some outer behaviour of the system or anything like that.
It is inevitable that the EA brand will now be associated with irrational feminized college students rather than interesting quantitative thinkers willing to bite socially undesirable bullets. In many ways this association will be correct in practice, because we are outnumbered.
I dislike this part, especially the phrase “irrational feminized college students”.
My rational behavior is to allocate all points to either A or B, to maximize the chance that one of them ends up in the top three. If I split my points between two candidates, I face the risk that neither makes it in the top three.
I think you are misunderstanding the mechanics of the elimination here. If you allocate nonzero points to both charities, then after one of A and B will be eliminated, all points will be reallocated to the remaining charity. So, to maximize the chance that one of them ends up in the top three, it doesn’t matter much weather to put 50 points on A and B each, or 99 points on A and 1 point on B (and actually, putting all points on one of A and B will do worse than these).
I didn’t read the article, only skimmed it, but couldn’t find much that supports the headline claim of $1.5 trillion. I tried searching for “$”, but all other dollar amounts are much smaller and don’t add up to $1.5 trillion, so it is unclear what this money will be conretely spent on. Also I couldn’t find a time frame over which this money would be spent.
Maybe you can add a summary of the parts of the article that deal with the headline amount, or say which things make you believe that this much money will likely/maybe be spent on nuclear missiles in the near future?
I’m looking for informed opinions only, preferably with links to respected essays. If you’re Scott Alexander or someone similar, fine, otherwise keep your personal opinions to yourself.
Please link to any on-point references and again, no uninformed “this is what I think” rants
I dislike this attitude when asking these kind of questions, especially when the question itself is a bit ranty (and in my opinion not that well thought-out, with seemingly little research behind it). Its conceivable that some readers can spot flaws in your arguments, but requiring that they are Scott-Alexander-famous or that they find a “respectable” source that supports them discourages them from answering.
I assume then the plan would be that those ballots just won’t continue to matter in the election?
No, it sounds like they will count as equally weighted votes. From the article: “If all of a voter’s points were assigned to candidates which are now eliminated, we’ll pretend that the voter spread their points out equally across the remaining candidates.” I think both would be ok choices
I also assume that the way the renormalization step works is that if everyone gets 10pts and someone voted A − 6pts, B − 2pts, C − 2pts and then A is eliminated, their ballot is then changed to “B − 5pts, C − 5pts”? And if B is also eliminated, their ballot becomes “C − 10pts”?
Basically, yes, according to my understanding. Except that there is no fixed point limit, but every voters points get normalized so that voters have equal weights.
Also do you allow for undervoting (e.g., a ballot that doesn’t allocate points to three options but is just “A − 10pts” full stop)?
I would guess yes, based on the description.
As someone who participated in the discussion I am satisfied with the outcome. In particular, I like that peoples votes still count with equal weight if their favourite charity gets eliminated.
A suggestion to supplement the voting system: The proposed voting system is compatible with ranked-choice voting. Maybe you can offer voters the option to give their vote in the form of ranking, without putting in lots of points. When calculating point totals, this would give all points of the voter to the highest-ranked charity on this voters ballot that has not been eliminated yet (within the current system, this effect can be approximated by giving 90% of points to my first choice, 9% to second, 0.9% to third, and so on).
Why? This could be a simpler way to vote for some people, and its fully compatible with the existing system. Also, the share of each voter corresponds to a few dollars, and at those scales, some EAs might think that all their share should go to the charity they consider to be the best, and not split up among multiple charities (this would correspond to giving all your points to that charity, but under the current system you should also spare some points in case your favourite charitiy gets eliminated).
Also, what is the point limit per voter? Maybe you dont need a limit at all, since you normalize the points anyways to give every voter equal weight...
This maximizes the sum of voter utilities, if you define voter utility to be the amount of money that goes to the charities a voter approves
This definition of “voter utility” feels very different to how EAs think about charities: the definition would imply that you are indifferent between all charities that you approve of. A better definition of “voter utility” would take into account the relative worth of the charities (eg a voter might think that charitiy A is 3x better than charity B, which is 5x better than charity C).
I just read most of the article. It was not that satisfying in this context. Most of it is arguments that we should work together (which I dont disagree with). And I imagine it will be quite hard to convince most AI xrisk people of “whether AI is closer to a stupid ‘stochastic parrot’ or on the ‘verge-of-superintelligence’ doesn’t really matter; ”. If we were to adopt Gideon’s desired framing, it looks like we would need to make sacrifices in epistemics. Related:
The relevant question isn’t “are the important harms to be prioritised the existential harms or the non-existential ones?”, “will AI be agents or not?’, nor ’will AI be stochastic parrots or superintelligence?” Rather, the relevant question is whether we think that power-accumulation and concentration in and through AI systems, at different scales of capability, is extremely risky.
Some of Gideon’s suggestion such as protest or compute governance are already being pursued. Not sure if that counts as bridge building though, because these might be good ideas anyways.
This would be very similar to first-past-the-post (third-past-the-post in this case), and has many of the same drawbacks as first-past-the-post, such as lots of strategic voting. Giving a voice to people who’s favorite charities are not wildly popular seems preferable (as would be the case with ranked-choice voting). The fact that you have 100 votes instead of 1 vote doesnt make much of a difference here (imagine a country where everyone has 99 clones, election systems would mostly still have the same advantages and disadvantages).
You haven’t actually addressed the main question of the previous comment: What would this bridge building look like? Your council example does not match the current reality very well.
It feels like you also sidestep other stuff in the comment, and it is unclear what your position is. Should we spend EA money (or other resources) on “short-term harms”? If yes, is the main reason because funding the marginal AI ethics research is better than the marginal bed-net and the marginal AI xrisk research? Or would the main reason for spending money on “short-term harms” that we buy sympathy with the group of people concerned about “short-term harms”, so we can later pass regulations together with them to reduce both “short-term harm” and AI x-risk.
some related discussion: https://www.lesswrong.com/posts/TYTEJxzeK3jBMq2TZ/your-posts-should-be-on-arxiv and the comments on that post
I think whether a post would be a good fit for an academic journal depends a lot on the concrete article, and might be not worthwhile for some. Maybe you can encourage authors of specific posts directly and point them towards a fitting academic journal?
some thoughts on different mechanisms:
Quadratic voting:
I think this could be fun. An advantage here is that voters have to think about the relative value of different charities, rather than just deciding which are better or worse. This could also be an important aspect when we want people to discuss how they plan to vote/how others should vote. If you want to be explicit about this, you could also consider designing the user interface so that users enter these relative differences of charities directly (e.g. “I vote charity A to be 3 times as good as charity B” rather than “I assign 90 vote credits to charity A and 10 vote credits to charity B”). Note however, that due to the top-3 cutoff, putting in the true relative differences between charities might not be the optimal policy.
A technical remark: If you want only to do payouts for the top three candidates, instead of just relying on the final vote, I think it would be better to rescale the voting credits of each voter after kicking out the charity with the least votes and then repeating the process until there are only 3 charities left. This would reduce tactical voting and would respect voters more who pick unusual charities as their top choices. This process has some similarities with ranked-choice voting. Additionally, users should have the ability to enter large relative differences (or very tiny votes like 1 in a billion), so their votes are still meaningful even after many eliminations.
Approval voting:
I think voting either “approve” or “disapprove” does not match how EAs think about charities. I generally approve a lot of charities within EA space, but would not vote “approve” for these charities.
I worry that a lot of tactical voting can take place here, especially if people can see the current votes or the pre-votes. For example, a person who both approves of the 3rd-placed charity and the 4th-placed charity (by overall popularity), might want to switch their vote to “disapprove” for the (according to them) worse charity. For example, voters are incentivized to give different votes to the 3rd-placed and 4th-placed charity, because there the difference will have the biggest impact on money paid out. Or a person who disapproves of all the top charities might switch a vote from “disapprove” to “approve” so that their vote matters at all.
Ranked-choice voting:
I am assuming here that the elimination process in ranked-choice stops once you reach the top 3 and that votes are then distributed proportionally. I think this would be a good implementation choice (mostly because proportional voting itself would be a decent choice by itself, so doing it for the top 3 seems reasonable). Ranking charities could be more satisfying for voters than having to figure out where to draw the line between “approve” and “disapprove”, or putting in lots of numeric values.
Generally, ranked-choice voting seems like an ok choice.
how well will these allocate funds?:
I am quite unsure here, and finding a best charity based on expressed preferences of lots of people with lots of opinions will be difficult in any case. My best guess here is that ranked-choice voting > quadratic voting > approval voting. A disadvantage of quadratic voting here is that it can happen that some fraction of the money will be paid out to sub-optimal charities (even if everyone agrees that charity C is worse than A and B, then it will likely still be rational for voters to assign non-zero weight to charity C, corresponding to non-zero payout).
understandability:
I think approval voting is easier to understand than ranked-choice voting, which is easier to understand than quadratic voting. This is both for the user interface and for understanding the whole system. Also, the mental effort for making a voting decision is less under ranked-choice and approval voting. I think the precise effects of the voters choices will be difficult to estimate in any system, so keeping
general remarks:
Different voting mechanisms can be useful for different purposes, and paying 3 charities different amounts of money is a different use case than selecting a single president, so not all considerations and analyses of different voting mechanisms will carry over to our particular case. The top-3 rule will incentivize tactical voting in all these systems (whereas in a purely proportional system there would be no tactical voting). Maybe this number should be increased a bit (especially if we use quadratic voting). If there are lots of charities to choose from, it will be quite an effort to evaluate all these charities. Potentially, you could give each voter a small number of charities to compare with each other, and then aggregate the result somehow (although that would be complicated and would change the character of the election). Or there can be two phases of voting, where the first phase narrows it down to 3-5 charities and then the second phase determines the proportions.
My personal preferences:
Obviously, we should have a meta-vote to select the three top voting methods among user-suggested voting methods and then hold three elections with the respective voting methods, each determining how a fraction of the fund (proportional to the vote that the voting method received in the meta-vote) gets distributed. And as for the voting method for this meta-vote, we should use… ok, this meta-voting stuff was not meant entirely seriously.
In my current personal judgement, I prefer quadratic voting over ranked-choice and ranked-choice over approval voting. I might be biased here towards more complex systems. I think an important factor is also that I might like more data about my preferences as a voter: With quadratic voting, I can express my relative preferences between charities quantitatively. With ranked-choice voting, I can rank charities, but cannot say by how much I prefer one charity over another. With approval voting, I can put charities in only two categories.
edit: I should have read the post more carefully
Do you intend to have one final winner or would it be ok to pay out the fund to various charities in different proportions (maybe with a minimum payout to avoid logistical hassle)? In the latter case, a consideration could also be proportional voting. But it is not clear how approval voting and ranked choice would work exactly in those cases.
Also, am I understanding correctly that donating more to that fund does not get you additional votes?
Not OP, but I would guess OP would recommend to pay only proportional to your estimated CO2 footprint. If you rarely travel, it will be much less than $35 (iirc veganism has a much smaller impact on CO2 than intercontinental flights).