I have two small corrections:

• I'm no longer confident moderate levels of drinking are healthy.
• It appears good progress has been made towards computing the freezing point of water from first principles.

On drinking

I've gone from high confidence one drink a day is mortality decreasing (argued here) to believing more likely than not this is not true. The primary cause of this update was looking into gene wide associational studies, specifically this one. Thanks to Hand Banana for bringing this to my attention. I'm not updating all the way to confidence that low amounts of drinking are bad or high confidence that it's not good because I'm a little suspicious about a study that controls for so many things and I don't have the time to check myself. Sadly this study that could have moved me to high confidence in either direction was cancelled because we can't have nice things.

I have a lot of thoughts that don't neatly connect together on the subject and certainly don't add up to a thesis so I'm just going to list them out:

• As a man if one of the first things someone says about why something is unhealthy is breast cancer, I have to wonder if it's actually neutral to good for me.
• People will absolutely not make the best case for their position and you can't expect them too. People will just say correlation is not causation and be done with it [1]. And if they're asked to give reasons why something is bad for you they'll lead with breast cancer even if their audience is 90% men. If Hand Banana hadn't made an offhand remark about GWAS I'd still be confident in my probably wrong position. [2]
• It's funny that people are very quick to decry any potential benefit of alcohol as associational but immediately accept negative effects like breast cancer as causal. I'd push on this point more but I also believe all the negative effects are causal.
• People often point to wealth as the hidden latent variable but I wonder if it's really mostly race or genetic. I was suspicious about the GWAS study at first because it seemed implausible that controlling for genes would invert the effect but thinking more about it different subpopulations have different histories of drinking. One could imagine a subpopulation with a long history of drinking would produce individuals with a greater affinity for drinking who are also more resilient to alcohol correlated diseases. But even they are better off not drinking.
• The associational studies show all the gain in mortality is due to the reduction in heart disease. This seems like a problem if your theory of the association is it's about rich people drinking more because wealthier people have better cancer outcomes too but moderate drinkers do not.

In conclusion I wish experiments were possible. We shouldn't have to speculate.

On First Principles Freezing Point

In The Water can only get so Cold I said scientists cannot compute the freezing point of water from first principles. But as of 2016 they sort of can. Thanks to Gilad for asking if I was sure this was true. Here's the paper and here's a popular explainer. It's not exactly the strong form where you only take Schrödinger equation and the atomic number of oxygen and hydrogen and calculate. It involves two different approximations. But it's still very impressive. And it undermines my strange philosophical point since maybe with the right approximations we could model the tipping points for all those other catastophized phenomena. Then again maybe not because in the case of water we know the freezing point in advance so at the end of our computation we'll know if or approximations were bad. Who would trust this paper if the freezing point and density of water weren't already known?

[1] Gwern is right and I should have an even higher prior that correlations aren't meaningful and are not "wagging their finger suggestively" or whatever. Usually I do but this time the causation seemed particularly plausible. It's funny how correlation+convincing explanation feels like good evidence but actually it's two bad pieces of evidence that don't add up to a good argument.

[2] I'm reminded of this Nate Silver Tweet. I think it's closer to the opposite though. People are slightly better at having true beliefs than they are at making good arguments. If someone has spent a lot of time justifying and arguing their belief that's probably because on a deeper level they suspect it's not true. Of course you shouldn't update much in either direction on a bad argument. It's mostly noise.