The good delusion: has effective altruism broken bad?

I'm not sure that you can square the "EA is precisely donating to best reviewed charities" and "tens of millions to AI safety is money well spent."

A huge problem is that people are developing this longtermism prediction model based on nothing more than sci-fi books and hunches. You need only look at how seriously places like SSC take Roko's Basilisk to see that they aren't modeling the future of AI in a reasonable way.

It is like donating to an organization that will prevent Vulcan, God of Volcanoes from destroying civilization.

Yes, government is bad at optimizing for moral good. Which is why billionaires should give their wealth to their own workers. If Tim Cook paid the people building iPhones more it would do far more good than him donating the money that comes from increased margins to some charity. The workers could decide what they need for themselves!

Instead, the idea seems to be that people are not properly qualified to identify their own needs and such decision need to be made by funneling profits to a CEO who then has superior knowledge of how to help the world and donates that money as he sees fit.

> There are a variety of ways in which such a person could be less moral than the pitifully ineffective non-billionaires.

I think we might be talking past each other a bit.

I'm saying, given no other information, someone who has given away billions, effectively, as in saved lives, is "morally ahead" of someone who hasn't similarly done so.

Of course they could've done other things that make them morally worse. Of course the way they donated money could be net-negative, if they caused negative things in the world.

But in the absence of other information, if I told you someone had donated a billion dollars to, I don't know, give food to hungry people, do you really think that is a morally neutral act? Like, if someone is a firefighter that has saved hundreds of lives by their heroism, do you think they are literally morally the same as other people who haven't? If that's not what you're talking about, I'm not sure what you are talking about.

Btw, I don't think morality is just an issue of summing up all your good versus bad deeds. I'm not exactly sure how to even define "ahead morally". A firefighter who has saved lives but also abused their kids - how do you even deal with a case like that? I think people are complex, they can do both good and bad, and the good that firefighter did is good, while the bad is bad, period.

I think one thing you might be objecting to is the idea that you can say something like "person X did 100 good deeds, and 99 bad deeds, therefore they are net good" or even "it's worth giving money to a cause that causes 99 bad deeds and 100 good deeds cause it's net positive." And I agree, that's a valid criticism and a wrong way to think about things.

The solution is to tax the shit out of rich people and hand it over to the rest of society. I mean the maths has been done: whatever the rich give out in charity is always less than what they would pay in taxes. The whole effective altruism thing is an ego trip experience. Wow I get my name on a hospital wing!

"If you define EA as "figuring out how to deploy a given set of resources so as to do the most good," your conclusions are going to depend on your value system: how do you define and measure good?"

I think more subtly it also depends on your calculation methodology. Rationalists talk a lot about the value of numbers for their Bayesian analyses and their "shut up and calculate" motto, but to my mind they overestimate the accuracy they have in their calculations by a lot. Of course I want to go with the 99% likely-to-help choice over the 25%-likely-to-help choice, but along with the fact my value system is indeed much different than theirs (in some important ways, infinitely so), I think all their Bayesian calculations need error bars in them. I think some rationalists would claim their probabilities have that error built into them, and I disagree. I think 75% likely is very different from 75% +/- 20% [1] when you start feeding that number through a chain of other calculations. A chain of calculations that nominally results in a 99.99% probability is a lot less impressive if a more honest calculation that propagates error bars correctly results in "somewhere between 2% and 99.9999%". It becomes more clear that the 99.99% is a very suspect number, even if it is nominally the middle of the distribution, especially given the number of times out in the real world the "highly improbable" ends up happening when it turns out that the assessment of the outcome being "highly improbable" was wrong. The probability distributions in the real world are highly pathological much more often than a naive analysis would suggest.

When analyzing the virtues of stealing billions to donate it to an effective charity, you better not just think that the value of the charity is higher than the cost of the loss of the money, you better be really sure your calculation is correct. I think the rationalists could argue amoungst themselves endlessly on the first point, but I do not believe they have any way whatsoever to establish their second point to my satisfaction, which renders the entire conversation moot, when there is no way they could ever convince me that their calculation is correct regardless of what it is. Even when my value system does in fact line up with theirs, and there is non-trivial overlap, I disagree with their calculation methodologies quite substantially as I think they consistently act with much more confidence then is justified.

Against that, I'm comfortable with the societally-evolved (evolution is important too, right, rationalists?) standard that in addition to the first-order effects of theft, one much consider the second-order effects of the dangers of normalizing outright theft because of the fact that someone else thinks they could use the resources better. Even if that is locally true, if everyone globally acts like that, global utility will plummet because everybody always thinks everybody else could spend their money better and the resulting chaos will not be helpful. (The Kantian imperative is completely inadequate as a theoretical foundation of philosophy but I do find it a useful practical tool, as long as one is aware to not place too much confidence in an extended analysis using it... the theme of this post comes up yet again.) The world is not improved by turning everyone into a central authority over all resource allocation in the world.

[1]: I know real error bars in probability space would never look like this, but mathematical accuracy would distract from my point here.