The more realistic version of this argument uses AI in place of digital people, which is certainly a possibility. But it's still... put gently... not very grounded.

I don't know what to do with this information. Have never seen a palatable fertility intervention!

Strong +1 on the first point. EA folks have done good work in these areas but it's swamped by the good work done by outsiders that I never see referenced.

Revealed preference is a good way to get a handle on what people value, but its normative foundation is strongest when the tradeoff is internal to people. Eg when we value lives vs income, we would want to use people's revealed preference for how they trade those off because those people are the most affected by our decisions and we want to incorporate their preferences. That normative foundation doesn't really apply to animal welfare where the trade-offs are between people and animals. You may as well use animals revealed preferences for saving humans (ie not at all) and conclude that humans have no worth; it would be nonsensical.

Fair points, I agree that taxation has a lower bar. The bimodal point was illustrative, you could take some other individual characteristics as proxies for the extent of internalities (e.g. education) and weight people by that when estimating.

The dismissal of consumer surplus-based ways to value alcohol consumption is puzzling. The main justification is that "it seems likely to us that consumers are behaving irrationally", but this is an overly broad statement. What fraction of alcohol consumption is irrational? If you believe that 30% of consumers are consuming irrationally high amounts, you could easily exclude the 30% heaviest drinkers and estimate consumer surplus for the remainder. In general, you can choose a population where you believe people are consuming more out of enjoyment than addiction.

This would require a bit of primary investigation, but you could use Nielsen scanner data with alcohol prices and consumption to a) drop the people who drink the most, b) estimate consumer surplus on the remainder. I'm pretty sure Nielsen has similar data in some LMICs if you want a more representative population. My prior is that you would arrive at substantially more than a 10% downward adjustment.

I would also like to hear this, as a well wisher :)

One concern that asymmetrically affects this discussion is measurement error. Democracy or institutional quality or any such impact is measured with a lot of error, so estimates of the impact of X on democracy/institutional quality are going to be super noisy and it's going to be harder to reject the null of no impact - even if there is a real negative impact. Is that something that's received a lot of attention?

In general, it's harder to put stock in "studies have not found evidence of X" than "studies have found evidence of X" since everything is measured with error, so rejecting nulls is harder than failing to reject them.