Here's an attempted reconstruction of Mills' argument. I'm not endorsing this argument (although there are parts of it with which I sympathize), but I think it is a lot better than the case for Mills as you present it in your post:
If a friend asked me whether she should vote in the upcoming Presidential election, I would advise her not to. It would be an inconvenience, and the chance of her vote making a difference to the outcome in my state is minuscule. From a consequentialist point of view, there is a good argument that it would be (mildly) unethical for her to vote, given the non-negligible cost and the negligible benefit. So if I were her personal ethical adviser, I would advise her not to vote. This analysis applies not just to my friend, but to most people in my state. So I might conclude that I would encourage significant good if I launched a large-scale state-wide media blitz discouraging voter turn-out. But this would be a bad idea! What is sound ethical advice directed at an individual is irresponsible when directed at the aggregate.
80k strongly encourages professional philanthropism over political activism, based on an individualist analysis. Any individual's chance of making a difference as an activist is small, much smaller than his chance of making a difference as a professional philanthropist. Directed at individuals, this might be sound ethical advice. But the message has pernicious consequences when directed at the aggregate, as 80k intends.
It is possible for political activism to move society towards a fundamental systemic change that would massively reduce global injustice and suffering. However, this requires a cadre of dedicated activists. Replaceability does not hold of political activism; if one morally serious and engaged activist is lured away from activism, it depletes the cadre. Now any single activist leaving (or not joining) the cadre will not significantly affect the chances of revolution succeeding. But if there is a message in the zeitgeist that discourages political participation, instead encouraging potential revolutionaries to participate in the capitalist system, this can significantly impact the chance of revolutionary success. So 80k's message is dangerous If enough motivated and passionate young people are convinced by their argument.
It's sort of like an n-person prisoner's dilemma, where each individual's (ethically) dominant strategy is to defect (conform with the capitalist system and be a philanthropist), but the Nash equilibrium is not the Pareto optimum. This kind of analysis is not uncommon in the Marxist literature. Analytic Marxists (like Jon Elster) interpret class consciousness as a stage of development at which individuals regard their strategy in a game as representative of the strategy of everyone in their socio-economic class. This changes the game so that certain strategies which would otherwise be individually attractive but which lead to unfortunate consequences if adopted in the aggregate are rendered individually unattractive. [It's been a while since I've read this stuff, so I may be misremembering, but this is what I recall.]
...If a friend asked me whether she should vote in the upcoming Presidential election, I would advise her not to. It would be an inconvenience, and the chance of her vote making a difference to the outcome in my state is minuscule. From a consequentialist point of view, there is a good argument that it would be (mildly) unethical for her to vote, given the non-negligible cost and the negligible benefit. So if I were her personal ethical adviser, I would advise her not to vote. This analysis applies not just to my friend, but to most people in my state. So I
The current issue of the Oxford Left Review has a debate between socialist Pete Mills and two 80,000 hours people, Ben Todd and Sebastian Farquhar: The Ethical Careers Debate, p4-9. I'm interested in it because I want to understand why people object to the ideas of 80,000 hours. A paraphrasing:
As a socialist, Mills really doesn't like the argument that the best way to help the world's poor is probably to work in heavily capitalist industries. He seems to be avoiding engaging with Todd and Farquhar's arguments, especially replaceability. He also really doesn't like looking at things in terms of numbers, I think because numbers suggest certainty. When I calculate that in 50 years of giving away $40K a year you save 1000 lives at $2K each, that's not saying the number is exactly 1000. It's saying 1000 is my best guess, and unless I can come up with a better guess it's the estimate I should use when choosing between this career path and other ones. He also doesn't seem to understand prediction and probability: "every revolution is impossible, until it is inevitable" may be how it feels for those living under an oppressive regime but it's not our best probability estimate. [1]
In a previous discussion a friend also was mislead calculations. When I said "one can avert infant deaths for about $500 each" their response was "What do they do with the 500 dollars? That doesn't seem to make sense. Do they give the infant a $500 anti-death pill? How do you know it really takes a constant stream of $500 for each infant?". Have other people run into this? Bad calculations also tend to be distributed widely, with people saying things like "one pint of blood can save up to three lives" when the expected marginal lives saved is actually tiny. Maybe we should focus less on estimates of effectiveness in smart-giving advocacy? Is there a way to show the huge difference in effect between the best charities and most charities without using these?
Maybe I should have way more of these discussions, enough that I can collect statistics on what arguments and examples work and which don't.
(I also posted this on my blog)
[1] Which is not to say you can't have big jumps in probability estimates. I could put the chance of revolution at 5% somewhere based on historical data but then hear some new information about how one has just started and sounds really promising which bumps my estimate up to 70%. But expected value calculations for jobs can work with numbers like these, it's just "impossible" and "inevitable" that break estimates.