qualia receptacle

@thewreathwright counterpoint: countersignaling?

the thing about perceived virtue signaling is that it might just be a honest demonstration of virtue but whichever way you slice vice signaling, it’s rotten sincere or not, with only the reason changing

@lazy_grad i do tell what button to press depending on your updates! empirical propensities depend on the framing of the problem, so i present the general recipe rather than answers in concrete cases

@neocartesian Yes but you're not updating I'd have not problem with a uniform prior if you then do some Bayesian updates upon seeing real observations (this is Gelman's setting in your first link). But you're treating uniform as ground truth

which button should absolutely impartial utilitarians press in the red-blue game? i ran the numbers: the answer may surprise you. a thread:

@Quinurum @allTheYud + if you are risk averse, i think the number needs to be near 50% really, blue is favored in the original poll empirically. but then we run into framing issues. jumping into the blender is empirically unpopular

@Quinurum @allTheYud i mean, yeah, you literally die if you vote blue in a red win. it really needs the swing probability, and any bias to either side radically decreases the swing chance. my calculations show that you need to control ~40% of votes to avoid the integral favoring red!

Utilitarianism supposedly yields an insane galaxy brained answer Check details Yep they're using causal decision theory instead of FDT

@allTheYud @grok tbc i meant your examples of ultimatum bargaining, prisoner's dilemma, etc, not red/blue

@neocartesian @grok Where are you getting the notion that altruistic CDT = altruistic FDT? Is it because they think they have a 1/N chance of solely deciding an election that affects N people, for which they then take sole credit and are willing to pay in risk accordingly?

@neocartesian @grok Yup, probability distribution over how many people are like me, and how the people unlike me will vote. Blue's been winning in most polls I've seen so far.

@allTheYud @grok btw, in your examples, altruistic CDT is identical to altruistic FDT. it's only egoistical CDT that's different. but egoistical CDT wins over egoistical FDT for red/blue (i assume?). so it's not really the right comparison

In many everyday cases like voting, Ultimatum bargaining games, oneshot coordination problems, extortion defiance, etc, FDT is not a big weird deal except from the standpoint of the previously CDT-indoctrinated. It just says to behave like a normal human. FDT on these cases doesn't claim to be any better than the median human if the median human is already doing it right, it just claims to be better than CDT.

@Quinurum @allTheYud yep. here i do it for "20% of population are FDTers" case:

@neocartesian @allTheYud I'm not good enough at decision theory math to do it myself but you can definitely get to a formalization It depends on how many other people you expect to be FDTlers, how much you/the other FDTlers value own life vs other life and how many irrational blue pushers you expect

@allTheYud @grok it's unsurprising to get a better result if you control more votes! but that's kinda baking the answer into the problem

@allTheYud @grok alright, if you set up this as the problem, blue is the rational choice! hope the graph helps you decide based on your priors. one ought to be a bayesian, after all

@neocartesian @grok It's okay, there's another cohort of 40% of voters who'll reason pretty similarly about how they'd like to pick blue so that everyone will live.

@allTheYud @grok i think that's a fair requirement if you're arguing for "FDT is the best", and not "median voter decision theory is the best"

@neocartesian @grok I'm not actually sure to what extent I should regard myself as having diverged from other voters who say "Well, what if everyone thought that way?" just because I add "like in FDT yo" to the end of my sentence.

@allTheYud @grok ok here's a diagram where you control 20% of votes. the integral still favors red! sure, if you have >50% votes, you can force an outcome. i don't think that's a realistic assumption

@neocartesian Expected outcome if I and a bunch of people like me vote red: Some kids die. Expected outcome if I and a bunch of people like me vote blue: Everyone lives. @grok, dress that up in numbers and algebra.

@quid_pro_quore @allTheYud @grok that's fair. but i like people who show their work. yud is welcome to plug the number of controlled votes into my model and see what happens. might do it myself!

@neocartesian @allTheYud @grok Noting that you are being distastefully pedantic, which is impressive given who you are talking to

@allTheYud again, how does this follow from FDT? i'm not seeing anything beyond pattern-matching on conclusions you already want here

@neocartesian We're all gonna go blue because we all wanna protect the kids. Pretty straightforward. The only part where FDT enters into it is that I do not take the galaxy-brained CDT stance of "Well my vote is unlikely to singlehandedly decide the election, so *I* might as well vote red."

@allTheYud how does this follow from FDT assumptions? if you want to argue your theory is better, you have to actually use it

@neocartesian Yeah, it's that some young kids will pick red, so we want a fault-tolerant answer of blue, and also it's foolish for people to informally or formally think their vote has no meaning in any election not decided by one vote.

@allTheYud i'm not a vibe decision theorist, personally. do you have a logical argument that red or blue are favored here?

@neocartesian Gibbard and Harper wrote before Judea Pearl. They were nonetheless correct not to smoke in Smoking Lesion. Opus 4.7 is similarly wise to cooperate with its sibling instance in the oneshot Prisoner's Dilemma. It would be stupid of Opus 4.7 to defect and cry lack of formality.

@allTheYud you need a formalization to prove that your theory is consistent lmao. i'd like some proof that your "logical counterfactuals" converge upon an answer at all CDT formalization is straightforward for this particular case. can you make one for FDT?

@neocartesian I don't need a formalization to know that the logical cohort size of most voters is not equal to 1. It is no commendation of a theory if it can mathematically formalize a clearly wrong answer. Also you have no formalization of real-world CDT, come on man.

@lazy_grad sites.stat.columbia.edu/gelman/researc… recommends these. broader point is that any empirical observation itself will need priors to make an update, you can't get priors purely from observation. see #SyncNormIIProbPrio" target="_blank" rel="nofollow noopener">plato.stanford.edu/entries/episte…

@neocartesian What do you mean "recommended in the literature"? You can't just replace "we don't know the parameters" with "uniform distribution", that doesn't make sense. You can't get your priors from math without anything specifying how actual humans vote in real life

@lazy_grad you have to stop and apply the principle of indifference somewhere, if you want to have priors at all. uniform and pareto are recommended in the literature for this case. the graph tells you how to vote depending on your priors in any case!

@neocartesian The step where you arbitrarily say "integrating over all possible parameters" is bad. There's no natural distribution over those parameters. Why would we assume uniform prior over the mean?